Compare commits

..
2 Commits
Author SHA1 Message Date
henrygd 5f74438bd7 update 2025-12-21 16:44:20 -05:00
henrygd ea354ec030 initial nvml collector 2025-12-21 14:33:04 -05:00
433 changed files with 12101 additions and 81112 deletions
+2 -2
View File
@@ -1,6 +1,6 @@
# Node.js dependencies # Node.js dependencies
node_modules/ node_modules
**/node_modules/ internalsite/node_modules
# Go build artifacts and binaries # Go build artifacts and binaries
build build
-2
View File
@@ -1,2 +0,0 @@
# Everything needs to be reviewed by Hank
* @henrygd
-19
View File
@@ -1,19 +0,0 @@
body:
- type: dropdown
id: component
attributes:
label: Component
description: Which part of Beszel is this about?
options:
- Hub
- Agent
- Hub & Agent
default: 0
validations:
required: true
- type: textarea
attributes:
label: Description
description: Please describe in detail what you want to share.
validations:
required: true
+14 -54
View File
@@ -1,54 +1,19 @@
body: body:
- type: checkboxes - type: markdown
id: terms
attributes: attributes:
label: Welcome! value: |
description: | ### Before opening a discussion:
Thank you for reaching out to the Beszel community for support! To help us assist you better, please make sure to review the following points before submitting your request:
Please note: - Check the [common issues guide](https://beszel.dev/guide/common-issues).
- For translation-related issues or requests, please use the [Crowdin project](https://crowdin.com/project/beszel). - Search existing [issues](https://github.com/henrygd/beszel/issues) and [discussions](https://github.com/henrygd/beszel/discussions) (including closed).
**- Please do not submit support reqeusts that are specific to ZFS. We plan to add integration with ZFS utilities in the near future.**
options:
- label: I have read the [Documentation](https://beszel.dev/guide/getting-started)
required: true
- label: I have checked the [Common Issues Guide](https://beszel.dev/guide/common-issues) and my problem was not mentioned there.
required: true
- label: I have searched open and closed issues and discussions and my problem was not mentioned before.
required: true
- label: I have verified I am using the latest version available. You can check the latest release [here](https://github.com/henrygd/beszel/releases).
required: true
- type: dropdown
id: component
attributes:
label: Component
description: Which part of Beszel is this about?
options:
- Hub
- Agent
- Hub & Agent
default: 0
validations:
required: true
- type: textarea - type: textarea
id: description id: description
attributes: attributes:
label: Problem Description label: Description
description: | description: A clear and concise description of the issue or question. If applicable, add screenshots to help explain your problem.
How to write a good bug report?
- Respect the issue template as much as possible.
- The title should be short and descriptive.
- Explain the conditions which led you to report this issue: the context.
- The context should lead to something, a problem that youre facing.
- Remain clear and concise.
- Format your messages to help the reader focus on what matters and understand the structure of your message, use [Markdown syntax](https://help.github.com/articles/github-flavored-markdown)
validations: validations:
required: true required: true
- type: input - type: input
id: system id: system
attributes: attributes:
@@ -56,15 +21,13 @@ body:
placeholder: linux/amd64 (agent), freebsd/arm64 (hub) placeholder: linux/amd64 (agent), freebsd/arm64 (hub)
validations: validations:
required: true required: true
- type: input
# - type: input id: version
# id: version attributes:
# attributes: label: Beszel version
# label: Beszel version placeholder: 0.9.1
# placeholder: 0.9.1 validations:
# validations: required: true
# required: true
- type: dropdown - type: dropdown
id: install-method id: install-method
attributes: attributes:
@@ -78,21 +41,18 @@ body:
- Other (please describe above) - Other (please describe above)
validations: validations:
required: true required: true
- type: textarea - type: textarea
id: config id: config
attributes: attributes:
label: Configuration label: Configuration
description: Please provide any relevant service configuration description: Please provide any relevant service configuration
render: yaml render: yaml
- type: textarea - type: textarea
id: hub-logs id: hub-logs
attributes: attributes:
label: Hub Logs label: Hub Logs
description: Check the logs page in PocketBase (`/_/#/logs`) for relevant errors (copy JSON). description: Check the logs page in PocketBase (`/_/#/logs`) for relevant errors (copy JSON).
render: json render: json
- type: textarea - type: textarea
id: agent-logs id: agent-logs
attributes: attributes:
+52 -51
View File
@@ -1,30 +1,8 @@
name: 🐛 Bug report name: 🐛 Bug report
description: Use this template to report a bug or issue. description: Report a new bug or issue.
title: '[Bug]: ' title: '[Bug]: '
labels: ['bug'] labels: ['bug', "needs confirmation"]
body: body:
- type: checkboxes
attributes:
label: Welcome!
description: |
The issue tracker is for reporting bugs and feature requests only. For end-user related support questions, please use the **[GitHub Discussions](https://github.com/henrygd/beszel/discussions/new?category=support)** instead
Please note:
- For translation-related issues or requests, please use the [Crowdin project](https://crowdin.com/project/beszel).
- To request a change or feature, use the [feature request form](https://github.com/henrygd/beszel/issues/new?template=feature_request.yml).
- Any issues that can be resolved by consulting the documentation or by reviewing existing open or closed issues will be closed.
**- Please do not submit bugs that are specific to ZFS. We plan to add integration with ZFS utilities in the near future.**
options:
- label: I have read the [Documentation](https://beszel.dev/guide/getting-started)
required: true
- label: I have checked the [Common Issues Guide](https://beszel.dev/guide/common-issues) and my problem was not mentioned there.
required: true
- label: I have searched open and closed issues and my problem was not mentioned before.
required: true
- label: I have verified I am using the latest version available. You can check the latest release [here](https://github.com/henrygd/beszel/releases).
required: true
- type: dropdown - type: dropdown
id: component id: component
attributes: attributes:
@@ -34,53 +12,81 @@ body:
- Hub - Hub
- Agent - Agent
- Hub & Agent - Hub & Agent
default: 0
validations: validations:
required: true required: true
- type: markdown
attributes:
value: |
### Thanks for taking the time to fill out this bug report!
- For more general support, please [start a support thread](https://github.com/henrygd/beszel/discussions/new?category=support).
- To request a change or feature, use the [feature request form](https://github.com/henrygd/beszel/issues/new?template=feature_request.yml).
- Please do not submit bugs that are specific to ZFS. We plan to add integration with ZFS utilities in the near future.
### Before submitting a bug report:
- Check the [common issues guide](https://beszel.dev/guide/common-issues).
- Search existing [issues](https://github.com/henrygd/beszel/issues) and [discussions](https://github.com/henrygd/beszel/discussions) (including closed).
- type: textarea - type: textarea
id: description id: description
attributes: attributes:
label: Problem Description label: Description
description: | description: Explain the issue you experienced clearly and concisely.
How to write a good bug report? placeholder: I went to the coffee pot and it was empty.
- Respect the issue template as much as possible.
- The title should be short and descriptive.
- Explain the conditions which led you to report this issue: the context.
- The context should lead to something, a problem that youre facing.
- Remain clear and concise.
- Format your messages to help the reader focus on what matters and understand the structure of your message, use [Markdown syntax](https://help.github.com/articles/github-flavored-markdown)
validations: validations:
required: true required: true
- type: textarea - type: textarea
id: expected-behavior id: expected-behavior
attributes: attributes:
label: Expected Behavior label: Expected Behavior
description: | description: In a perfect world, what should have happened?
In a perfect world, what should have happened?
**Important:** Be specific. Vague descriptions like "it should work" are not helpful.
placeholder: When I got to the coffee pot, it should have been full. placeholder: When I got to the coffee pot, it should have been full.
validations: validations:
required: true required: true
- type: textarea - type: textarea
id: steps-to-reproduce id: steps-to-reproduce
attributes: attributes:
label: Steps to Reproduce label: Steps to Reproduce
description: | description: Describe how to reproduce the issue in repeatable steps.
Provide detailed, numbered steps that someone else can follow to reproduce the issue.
**Important:** Vague descriptions like "it doesn't work" or "it's broken" will result in the issue being closed.
Include specific actions, URLs, button clicks, and any relevant data or configuration.
placeholder: | placeholder: |
1. Go to the coffee pot. 1. Go to the coffee pot.
2. Make more coffee. 2. Make more coffee.
3. Pour it into a cup. 3. Pour it into a cup.
4. Observe that the cup is empty instead of full.
validations: validations:
required: true required: true
- type: dropdown
id: category
attributes:
label: Category
description: Which category does this relate to most?
options:
- Metrics
- Charts & Visualization
- Settings & Configuration
- Notifications & Alerts
- Authentication
- Installation
- Performance
- UI / UX
- Other
validations:
required: true
- type: dropdown
id: metrics
attributes:
label: Affected Metrics
description: If applicable, which specific metric does this relate to most?
options:
- CPU
- Memory
- Storage
- Network
- Containers
- GPU
- Sensors
- Other
validations:
required: true
- type: input - type: input
id: system id: system
attributes: attributes:
@@ -88,7 +94,6 @@ body:
placeholder: linux/amd64 (agent), freebsd/arm64 (hub) placeholder: linux/amd64 (agent), freebsd/arm64 (hub)
validations: validations:
required: true required: true
- type: input - type: input
id: version id: version
attributes: attributes:
@@ -96,7 +101,6 @@ body:
placeholder: 0.9.1 placeholder: 0.9.1
validations: validations:
required: true required: true
- type: dropdown - type: dropdown
id: install-method id: install-method
attributes: attributes:
@@ -110,21 +114,18 @@ body:
- Other (please describe above) - Other (please describe above)
validations: validations:
required: true required: true
- type: textarea - type: textarea
id: config id: config
attributes: attributes:
label: Configuration label: Configuration
description: Please provide any relevant service configuration description: Please provide any relevant service configuration
render: yaml render: yaml
- type: textarea - type: textarea
id: hub-logs id: hub-logs
attributes: attributes:
label: Hub Logs label: Hub Logs
description: Check the logs page in PocketBase (`/_/#/logs`) for relevant errors (copy JSON). description: Check the logs page in PocketBase (`/_/#/logs`) for relevant errors (copy JSON).
render: json render: json
- type: textarea - type: textarea
id: agent-logs id: agent-logs
attributes: attributes:
-3
View File
@@ -1,8 +1,5 @@
blank_issues_enabled: false blank_issues_enabled: false
contact_links: contact_links:
- name: 🗣️ Translations
url: https://crowdin.com/project/beszel
about: Please report translation issues and request new translations here.
- name: 💬 Support and questions - name: 💬 Support and questions
url: https://github.com/henrygd/beszel/discussions url: https://github.com/henrygd/beszel/discussions
about: Ask and answer questions here. about: Ask and answer questions here.
+51 -32
View File
@@ -1,25 +1,8 @@
name: 🚀 Feature request name: 🚀 Feature request
description: Request a new feature or change. description: Request a new feature or change.
title: "[Feature]: " title: "[Feature]: "
labels: ["enhancement"] labels: ["enhancement", "needs review"]
body: body:
- type: checkboxes
attributes:
label: Welcome!
description: |
The issue tracker is for reporting bugs and feature requests only. For end-user related support questions, please use the **[GitHub Discussions](https://github.com/henrygd/beszel/discussions)** instead
Please note:
- For **Bug reports**, use the [Bug Form](https://github.com/henrygd/beszel/issues/new?template=bug_report.yml).
- Any requests for new translations should be requested within the [crowdin project](https://crowdin.com/project/beszel).
- Create one issue per feature request. This helps us keep track of requests and prioritize them accordingly.
options:
- label: I have searched open and closed feature requests to make sure this or similar feature request does not already exist.
required: true
- label: This is a feature request, not a bug report or support question.
required: true
- type: dropdown - type: dropdown
id: component id: component
attributes: attributes:
@@ -29,29 +12,65 @@ body:
- Hub - Hub
- Agent - Agent
- Hub & Agent - Hub & Agent
default: 0
validations: validations:
required: true required: true
- type: markdown
- type: textarea
id: description
attributes: attributes:
label: Description value: Before submitting, please search existing [issues](https://github.com/henrygd/beszel/issues) and [discussions](https://github.com/henrygd/beszel/discussions) (including closed).
description: | - type: textarea
Describe the solution or feature you'd like. Explain what problem this solves or what value it adds. attributes:
**Important:** Be specific and detailed. Vague requests like "make it better" will be closed. label: Describe the feature you would like to see
placeholder: |
Example:
- What is the feature?
- What problem does it solve?
- How should it work?
validations: validations:
required: true required: true
- type: textarea - type: textarea
id: motivation id: motivation
attributes: attributes:
label: Motivation / Use Case label: Motivation / Use Case
description: Why do you want this feature? What problem does it solve? description: Why do you want this feature? What problem does it solve?
validations:
required: true
- type: textarea
attributes:
label: Describe how you would like to see this feature implemented
validations:
required: true
- type: textarea
id: logs
attributes:
label: Screenshots
description: Please attach any relevant screenshots, such as images from your current solution or similar implementations.
validations:
required: false
- type: dropdown
id: category
attributes:
label: Category
description: Which category does this relate to most?
options:
- Metrics
- Charts & Visualization
- Settings & Configuration
- Notifications & Alerts
- Authentication
- Installation
- Performance
- UI / UX
- Other
validations:
required: true
- type: dropdown
id: metrics
attributes:
label: Affected Metrics
description: If applicable, which specific metric does this relate to most?
options:
- CPU
- Memory
- Storage
- Network
- Containers
- GPU
- Sensors
- Other
validations: validations:
required: true required: true
-12
View File
@@ -1,12 +0,0 @@
version: 2
updates:
- package-ecosystem: gomod
directory: /
schedule:
interval: weekly
- package-ecosystem: github-actions
directory: /
schedule:
interval: weekly
+9 -37
View File
@@ -41,7 +41,7 @@ jobs:
# henrygd/beszel-agent-nvidia # henrygd/beszel-agent-nvidia
- image: henrygd/beszel-agent-nvidia - image: henrygd/beszel-agent-nvidia
dockerfile: ./internal/dockerfile_agent_nvidia dockerfile: ./internal/dockerfile_agent_nvidia
platforms: linux/amd64,linux/arm64 platforms: linux/amd64
registry: docker.io registry: docker.io
username_secret: DOCKERHUB_USERNAME username_secret: DOCKERHUB_USERNAME
password_secret: DOCKERHUB_TOKEN password_secret: DOCKERHUB_TOKEN
@@ -52,19 +52,6 @@ jobs:
type=semver,pattern={{major}} type=semver,pattern={{major}}
type=raw,value={{sha}},enable=${{ github.ref_type != 'tag' }} type=raw,value={{sha}},enable=${{ github.ref_type != 'tag' }}
# henrygd/beszel-agent-nvidia:slim
- image: henrygd/beszel-agent-nvidia
dockerfile: ./internal/dockerfile_agent_nvidia_slim
platforms: linux/amd64,linux/arm64
registry: docker.io
username_secret: DOCKERHUB_USERNAME
password_secret: DOCKERHUB_TOKEN
tags: |
type=raw,value=slim
type=semver,pattern={{version}}-slim
type=semver,pattern={{major}}.{{minor}}-slim
type=semver,pattern={{major}}-slim
# henrygd/beszel-agent-intel # henrygd/beszel-agent-intel
- image: henrygd/beszel-agent-intel - image: henrygd/beszel-agent-intel
dockerfile: ./internal/dockerfile_agent_intel dockerfile: ./internal/dockerfile_agent_intel
@@ -109,7 +96,7 @@ jobs:
# ghcr.io/henrygd/beszel-agent-nvidia # ghcr.io/henrygd/beszel-agent-nvidia
- image: ghcr.io/${{ github.repository }}/beszel-agent-nvidia - image: ghcr.io/${{ github.repository }}/beszel-agent-nvidia
dockerfile: ./internal/dockerfile_agent_nvidia dockerfile: ./internal/dockerfile_agent_nvidia
platforms: linux/amd64,linux/arm64 platforms: linux/amd64
registry: ghcr.io registry: ghcr.io
username: ${{ github.actor }} username: ${{ github.actor }}
password_secret: GITHUB_TOKEN password_secret: GITHUB_TOKEN
@@ -120,19 +107,6 @@ jobs:
type=semver,pattern={{major}} type=semver,pattern={{major}}
type=raw,value={{sha}},enable=${{ github.ref_type != 'tag' }} type=raw,value={{sha}},enable=${{ github.ref_type != 'tag' }}
# ghcr.io/henrygd/beszel-agent-nvidia:slim
- image: ghcr.io/${{ github.repository }}/beszel-agent-nvidia
dockerfile: ./internal/dockerfile_agent_nvidia_slim
platforms: linux/amd64,linux/arm64
registry: ghcr.io
username: ${{ github.actor }}
password_secret: GITHUB_TOKEN
tags: |
type=raw,value=slim
type=semver,pattern={{version}}-slim
type=semver,pattern={{major}}.{{minor}}-slim
type=semver,pattern={{major}}-slim
# ghcr.io/henrygd/beszel-agent-intel # ghcr.io/henrygd/beszel-agent-intel
- image: ghcr.io/${{ github.repository }}/beszel-agent-intel - image: ghcr.io/${{ github.repository }}/beszel-agent-intel
dockerfile: ./internal/dockerfile_agent_intel dockerfile: ./internal/dockerfile_agent_intel
@@ -178,7 +152,7 @@ jobs:
steps: steps:
- name: Checkout - name: Checkout
uses: actions/checkout@v7 uses: actions/checkout@v4
- name: Set up bun - name: Set up bun
uses: oven-sh/setup-bun@v2 uses: oven-sh/setup-bun@v2
@@ -190,14 +164,14 @@ jobs:
run: bun run --cwd ./internal/site build run: bun run --cwd ./internal/site build
- name: Set up QEMU - name: Set up QEMU
uses: docker/setup-qemu-action@v4 uses: docker/setup-qemu-action@v3
- name: Set up Docker Buildx - name: Set up Docker Buildx
uses: docker/setup-buildx-action@v4 uses: docker/setup-buildx-action@v3
- name: Docker metadata - name: Docker metadata
id: metadata id: metadata
uses: docker/metadata-action@v6 uses: docker/metadata-action@v5
with: with:
images: ${{ matrix.image }} images: ${{ matrix.image }}
tags: ${{ matrix.tags }} tags: ${{ matrix.tags }}
@@ -207,7 +181,7 @@ jobs:
env: env:
password_secret_exists: ${{ secrets[matrix.password_secret] != '' && 'true' || 'false' }} password_secret_exists: ${{ secrets[matrix.password_secret] != '' && 'true' || 'false' }}
if: github.event_name != 'pull_request' && env.password_secret_exists == 'true' if: github.event_name != 'pull_request' && env.password_secret_exists == 'true'
uses: docker/login-action@v4 uses: docker/login-action@v3
with: with:
username: ${{ matrix.username || secrets[matrix.username_secret] }} username: ${{ matrix.username || secrets[matrix.username_secret] }}
password: ${{ secrets[matrix.password_secret] }} password: ${{ secrets[matrix.password_secret] }}
@@ -216,13 +190,11 @@ jobs:
# Build and push Docker image with Buildx (don't push on PR) # Build and push Docker image with Buildx (don't push on PR)
# https://github.com/docker/build-push-action # https://github.com/docker/build-push-action
- name: Build and push Docker image - name: Build and push Docker image
uses: docker/build-push-action@v7 uses: docker/build-push-action@v5
with: with:
context: ./ context: ./
file: ${{ matrix.dockerfile }} file: ${{ matrix.dockerfile }}
platforms: ${{ matrix.platforms || 'linux/amd64,linux/arm64,linux/arm/v6,linux/arm/v7' }} platforms: ${{ matrix.platforms || 'linux/amd64,linux/arm64,linux/arm/v7' }}
push: ${{ github.ref_type == 'tag' && secrets[matrix.password_secret] != '' }} push: ${{ github.ref_type == 'tag' && secrets[matrix.password_secret] != '' }}
provenance: mode=max
sbom: true
tags: ${{ steps.metadata.outputs.tags }} tags: ${{ steps.metadata.outputs.tags }}
labels: ${{ steps.metadata.outputs.labels }} labels: ${{ steps.metadata.outputs.labels }}
-109
View File
@@ -1,109 +0,0 @@
name: Helm charts
on:
pull_request:
paths:
- "supplemental/helm/**"
push:
branches:
- main
paths:
- "supplemental/helm/**"
permissions:
contents: read
packages: write
env:
OCI_REGISTRY: ghcr.io/henrygd/beszel-charts
jobs:
changes:
name: Detect changed charts
runs-on: ubuntu-latest
outputs:
charts: ${{ steps.changes.outputs.charts }}
steps:
- name: Checkout repository
uses: actions/checkout@v7
with:
fetch-depth: 0
- name: Detect changed charts
id: changes
env:
BASE_SHA: ${{ github.event_name == 'pull_request' && github.event.pull_request.base.sha || github.event.before }}
run: |
charts=()
for name in beszel-agent beszel-hub; do
path="supplemental/helm/$name"
if ! git diff --quiet "$BASE_SHA" "$GITHUB_SHA" -- "$path"; then
charts+=("$name|$path")
fi
done
printf '%s\n' "${charts[@]}" \
| jq -Rsc 'split("\n") | map(select(length > 0) | split("|") | {name: .[0], path: .[1]})' \
| xargs -0 printf 'charts=%s\n' >> "$GITHUB_OUTPUT"
validate-and-publish:
name: ${{ github.event_name == 'push' && 'Publish' || 'Validate' }} ${{ matrix.chart.name }}
needs: changes
if: needs.changes.outputs.charts != '[]'
runs-on: ubuntu-latest
strategy:
fail-fast: false
matrix:
chart: ${{ fromJSON(needs.changes.outputs.charts) }}
steps:
- name: Checkout repository
uses: actions/checkout@v7
- name: Set up Helm
uses: azure/setup-helm@v5
- name: Lint chart
run: helm lint "${{ matrix.chart.path }}" --set env.KEY=ci-placeholder
- name: Render chart
run: helm template "${{ matrix.chart.name }}" "${{ matrix.chart.path }}" --set env.KEY=ci-placeholder > /dev/null
- name: Package chart
id: package
env:
CHART_NAME: ${{ matrix.chart.name }}
CHART_PATH: ${{ matrix.chart.path }}
run: |
version=$(awk '/^version:/ { print $2 }' "$CHART_PATH/Chart.yaml")
test -n "$version"
mkdir -p .helm-packages
helm package "$CHART_PATH" --destination .helm-packages
package=".helm-packages/${CHART_NAME}-${version}.tgz"
test -f "$package"
echo "version=$version" >> "$GITHUB_OUTPUT"
echo "package=$package" >> "$GITHUB_OUTPUT"
- name: Log in to GHCR
env:
GITHUB_TOKEN: ${{ github.token }}
run: echo "$GITHUB_TOKEN" | helm registry login ghcr.io --username "$GITHUB_ACTOR" --password-stdin
- name: Check chart version is unpublished
env:
CHART_NAME: ${{ matrix.chart.name }}
CHART_VERSION: ${{ steps.package.outputs.version }}
run: |
chart="oci://${OCI_REGISTRY}/${CHART_NAME}"
if helm show chart "$chart" --version "$CHART_VERSION" > /dev/null 2>&1; then
echo "${CHART_NAME} ${CHART_VERSION} is already published. Bump version in Chart.yaml." >&2
exit 1
fi
- name: Publish chart
if: github.event_name == 'push'
run: helm push "${{ steps.package.outputs.package }}" "oci://${OCI_REGISTRY}"
+4 -10
View File
@@ -6,7 +6,6 @@ on:
workflow_dispatch: workflow_dispatch:
permissions: permissions:
actions: write
issues: write issues: write
pull-requests: write pull-requests: write
@@ -15,7 +14,7 @@ jobs:
name: Lock Inactive Issues name: Lock Inactive Issues
runs-on: ubuntu-24.04 runs-on: ubuntu-24.04
steps: steps:
- uses: klaasnicolaas/action-inactivity-lock@v2.0.1 - uses: klaasnicolaas/action-inactivity-lock@v1.1.3
id: lock id: lock
with: with:
days-inactive-issues: 14 days-inactive-issues: 14
@@ -29,7 +28,7 @@ jobs:
runs-on: ubuntu-24.04 runs-on: ubuntu-24.04
steps: steps:
- name: Close Stale Issues - name: Close Stale Issues
uses: actions/stale@v11 uses: actions/stale@v10
with: with:
repo-token: ${{ secrets.GITHUB_TOKEN }} repo-token: ${{ secrets.GITHUB_TOKEN }}
@@ -49,16 +48,11 @@ jobs:
# Action can not skip PRs, set it to 100 years to cover it. # Action can not skip PRs, set it to 100 years to cover it.
days-before-pr-stale: 36524 days-before-pr-stale: 36524
# Max issues to process before early exit. Next run resumes from cache. GH API limit: 5000.
operations-per-run: 1500
# Labels # Labels
stale-issue-label: 'stale' stale-issue-label: 'stale'
remove-stale-when-updated: true remove-stale-when-updated: true
any-of-labels: 'awaiting-requester' only-issue-labels: 'awaiting-requester'
exempt-issue-labels: 'enhancement'
# Exemptions # Exemptions
exempt-assignees: true exempt-assignees: true
exempt-milestones: true
exempt-milestones: true
+82
View File
@@ -0,0 +1,82 @@
name: Label issues from dropdowns
on:
issues:
types: [opened]
jobs:
label_from_dropdown:
runs-on: ubuntu-latest
permissions:
issues: write
steps:
- name: Apply labels based on dropdown choices
uses: actions/github-script@v7
with:
script: |
const issueNumber = context.issue.number;
const owner = context.repo.owner;
const repo = context.repo.repo;
// Get the issue body
const body = context.payload.issue.body;
// Helper to find dropdown value in the body (assuming markdown format)
function extractSectionValue(heading) {
const regex = new RegExp(`### ${heading}\\s+([\\s\\S]*?)(?:\\n###|$)`, 'i');
const match = body.match(regex);
if (match) {
// Get the first non-empty line after the heading
const lines = match[1].split('\n').map(l => l.trim()).filter(Boolean);
return lines[0] || null;
}
return null;
}
// Extract dropdown selections
const category = extractSectionValue('Category');
const metrics = extractSectionValue('Affected Metrics');
const component = extractSectionValue('Component');
// Build labels to add
let labelsToAdd = [];
if (category) labelsToAdd.push(category);
if (metrics) labelsToAdd.push(metrics);
if (component) labelsToAdd.push(component);
// Get existing labels in the repo
const { data: existingLabels } = await github.rest.issues.listLabelsForRepo({
owner,
repo,
per_page: 100
});
const existingLabelNames = existingLabels.map(l => l.name);
// Find labels that need to be created
const labelsToCreate = labelsToAdd.filter(label => !existingLabelNames.includes(label));
// Create missing labels (with a default color)
for (const label of labelsToCreate) {
try {
await github.rest.issues.createLabel({
owner,
repo,
name: label,
color: 'ededed' // light gray, you can pick any hex color
});
} catch (e) {
// Ignore if label already exists (race condition), otherwise rethrow
if (!e || e.status !== 422) throw e;
}
}
// Now apply all labels (they all exist now)
if (labelsToAdd.length > 0) {
await github.rest.issues.addLabels({
owner,
repo,
issue_number: issueNumber,
labels: labelsToAdd
});
}
+5 -5
View File
@@ -13,7 +13,7 @@ jobs:
runs-on: ubuntu-latest runs-on: ubuntu-latest
steps: steps:
- name: Checkout - name: Checkout
uses: actions/checkout@v7 uses: actions/checkout@v4
with: with:
fetch-depth: 0 fetch-depth: 0
@@ -27,12 +27,12 @@ jobs:
run: bun run --cwd ./internal/site build run: bun run --cwd ./internal/site build
- name: Set up Go - name: Set up Go
uses: actions/setup-go@v7 uses: actions/setup-go@v5
with: with:
go-version: stable go-version: "^1.22.1"
- name: Set up .NET - name: Set up .NET
uses: actions/setup-dotnet@v6 uses: actions/setup-dotnet@v4
with: with:
dotnet-version: "9.0.x" dotnet-version: "9.0.x"
@@ -42,7 +42,7 @@ jobs:
shell: bash shell: bash
- name: GoReleaser beszel - name: GoReleaser beszel
uses: goreleaser/goreleaser-action@v7 uses: goreleaser/goreleaser-action@v6
with: with:
workdir: ./ workdir: ./
distribution: goreleaser distribution: goreleaser
-101
View File
@@ -1,101 +0,0 @@
name: Update Helm charts
on:
release:
types:
- published
permissions:
contents: write
pull-requests: write
concurrency:
group: update-helm-charts
cancel-in-progress: false
jobs:
update:
name: Propose chart update
if: ${{ github.repository_owner == 'henrygd' && startsWith(github.event.release.tag_name, 'v') && !github.event.release.prerelease }}
runs-on: ubuntu-latest
env:
BRANCH: automation/update-helm-app-version
RELEASE_TAG: ${{ github.event.release.tag_name }}
AUTOMATION_TOKEN: ${{ secrets.CR_TOKEN || github.token }}
steps:
- name: Checkout main
uses: actions/checkout@v7
with:
ref: main
token: ${{ env.AUTOMATION_TOKEN }}
- name: Update chart versions
id: update
run: |
version="${RELEASE_TAG#v}"
if [[ ! "$version" =~ ^[0-9]+\.[0-9]+\.[0-9]+$ ]]; then
echo "Unsupported software release version: $version" >&2
exit 1
fi
changed=false
for chart in supplemental/helm/beszel-agent supplemental/helm/beszel-hub; do
current_app_version=$(awk -F '"' '/^appVersion:/ { print $2 }' "$chart/Chart.yaml")
if [[ "$current_app_version" == "$version" ]]; then
echo "$chart already uses appVersion $version"
continue
fi
newest_version=$(printf '%s\n' "$current_app_version" "$version" | sort -V | tail -n 1)
if [[ "$newest_version" != "$version" ]]; then
echo "Skipping stale update of $chart from $current_app_version to $version"
continue
fi
chart_version=$(awk '/^version:/ { print $2 }' "$chart/Chart.yaml")
if [[ ! "$chart_version" =~ ^([0-9]+)\.([0-9]+)\.([0-9]+)$ ]]; then
echo "Unsupported chart version in $chart/Chart.yaml: $chart_version" >&2
exit 1
fi
next_chart_version="${BASH_REMATCH[1]}.${BASH_REMATCH[2]}.$((BASH_REMATCH[3] + 1))"
NEW_APP_VERSION="$version" NEW_CHART_VERSION="$next_chart_version" \
perl -pi -e 's/^appVersion:.*$/appVersion: "$ENV{NEW_APP_VERSION}"/; s/^version:.*$/version: $ENV{NEW_CHART_VERSION}/' \
"$chart/Chart.yaml"
OLD_APP_VERSION="$current_app_version" NEW_APP_VERSION="$version" \
perl -pi -e 's/\Q$ENV{OLD_APP_VERSION}\E/$ENV{NEW_APP_VERSION}/g' "$chart/README.md"
echo "$chart: appVersion $current_app_version -> $version, chart $chart_version -> $next_chart_version"
changed=true
done
echo "changed=$changed" >> "$GITHUB_OUTPUT"
- name: Open or update pull request
if: steps.update.outputs.changed == 'true'
env:
GH_TOKEN: ${{ env.AUTOMATION_TOKEN }}
run: |
version="${RELEASE_TAG#v}"
title="chore(helm): update app version to ${version}"
body="Updates the Helm charts for [Beszel ${version}](${GITHUB_SERVER_URL}/${GITHUB_REPOSITORY}/releases/tag/${RELEASE_TAG}) and bumps their chart patch versions. Merging this pull request publishes the updated charts to GHCR."
git config user.name "github-actions[bot]"
git config user.email "41898282+github-actions[bot]@users.noreply.github.com"
git checkout -B "$BRANCH"
git add supplemental/helm/beszel-agent/Chart.yaml \
supplemental/helm/beszel-agent/README.md \
supplemental/helm/beszel-hub/Chart.yaml \
supplemental/helm/beszel-hub/README.md
git commit -m "$title"
git fetch origin "$BRANCH" || true
git push --force-with-lease origin "HEAD:refs/heads/${BRANCH}"
pr_number=$(gh pr list --head "$BRANCH" --base main --state open --json number --jq '.[0].number')
if [[ -n "$pr_number" ]]; then
gh pr edit "$pr_number" --title "$title" --body "$body"
else
gh pr create --base main --head "$BRANCH" --title "$title" --body "$body"
fi
+7 -3
View File
@@ -2,6 +2,10 @@
name: VulnCheck name: VulnCheck
on: on:
pull_request:
branches:
- main
push: push:
branches: branches:
- main - main
@@ -15,11 +19,11 @@ jobs:
runs-on: ubuntu-latest runs-on: ubuntu-latest
steps: steps:
- name: Check out code into the Go module directory - name: Check out code into the Go module directory
uses: actions/checkout@v7 uses: actions/checkout@v4
- name: Set up Go - name: Set up Go
uses: actions/setup-go@v7 uses: actions/setup-go@v5
with: with:
go-version: stable go-version: 1.25.x
# cached: false # cached: false
- name: Get official govulncheck - name: Get official govulncheck
run: go install golang.org/x/vuln/cmd/govulncheck@latest run: go install golang.org/x/vuln/cmd/govulncheck@latest
+1 -3
View File
@@ -3,13 +3,13 @@ pb_data
data data
temp temp
.vscode .vscode
beszel-agent
beszel_data beszel_data
beszel_data* beszel_data*
dist dist
*.exe *.exe
internal/cmd/hub/hub internal/cmd/hub/hub
internal/cmd/agent/agent internal/cmd/agent/agent
agent.test
node_modules node_modules
build build
*timestamp* *timestamp*
@@ -20,5 +20,3 @@ __debug_*
agent/lhm/obj agent/lhm/obj
agent/lhm/bin agent/lhm/bin
dockerfile_agent_dev dockerfile_agent_dev
.cr-release-packages
.tmp
+3 -33
View File
@@ -31,16 +31,12 @@ builds:
goarch: arm64 goarch: arm64
- goos: freebsd - goos: freebsd
goarch: arm goarch: arm
- goos: darwin
goarch: arm
- id: beszel-agent - id: beszel-agent
binary: beszel-agent binary: beszel-agent
main: internal/cmd/agent/agent.go main: internal/cmd/agent/agent.go
env: env:
- CGO_ENABLED=0 - CGO_ENABLED=0
ldflags:
- -s -w -X github.com/henrygd/beszel/internal/ghupdate.buildGOARM={{ .Arm }}
goos: goos:
- linux - linux
- darwin - darwin
@@ -56,10 +52,6 @@ builds:
- mipsle - mipsle
- mips - mips
- ppc64le - ppc64le
goarm:
- "5"
- "6"
- "7"
gomips: gomips:
- hardfloat - hardfloat
- softfloat - softfloat
@@ -79,25 +71,11 @@ builds:
gomips: hardfloat gomips: hardfloat
- goos: windows - goos: windows
goarch: arm goarch: arm
- goos: darwin
goarch: arm
- goos: darwin - goos: darwin
goarch: riscv64 goarch: riscv64
- goos: windows - goos: windows
goarch: riscv64 goarch: riscv64
- id: beszel-agent-linux-amd64-glibc
binary: beszel-agent
main: internal/cmd/agent/agent.go
env:
- CGO_ENABLED=0
flags:
- -tags=glibc
goos:
- linux
goarch:
- amd64
archives: archives:
- id: beszel-agent - id: beszel-agent
formats: [tar.gz] formats: [tar.gz]
@@ -107,20 +85,10 @@ archives:
{{ .Binary }}_ {{ .Binary }}_
{{- .Os }}_ {{- .Os }}_
{{- .Arch }} {{- .Arch }}
{{- if ne .Arm "6" }}{{ with .Arm }}v{{ . }}{{ end }}{{ end }}
format_overrides: format_overrides:
- goos: windows - goos: windows
formats: [zip] formats: [zip]
- id: beszel-agent-linux-amd64-glibc
formats: [tar.gz]
ids:
- beszel-agent-linux-amd64-glibc
name_template: >-
{{ .Binary }}_
{{- .Os }}_
{{- .Arch }}_glibc
- id: beszel - id: beszel
formats: [tar.gz] formats: [tar.gz]
ids: ids:
@@ -169,7 +137,9 @@ nfpms:
- debconf - debconf
scripts: scripts:
templates: ./supplemental/debian/templates templates: ./supplemental/debian/templates
config: ./supplemental/debian/config.sh # Currently broken due to a bug in goreleaser
# https://github.com/goreleaser/goreleaser/issues/5487
#config: ./supplemental/debian/config.sh
scoops: scoops:
- ids: [beszel-agent] - ids: [beszel-agent]
+5 -38
View File
@@ -3,40 +3,6 @@ OS ?= $(shell go env GOOS)
ARCH ?= $(shell go env GOARCH) ARCH ?= $(shell go env GOARCH)
# Skip building the web UI if true # Skip building the web UI if true
SKIP_WEB ?= false SKIP_WEB ?= false
# Controls NVML/glibc agent build tag behavior:
# - auto (default): enable on linux/amd64 glibc hosts
# - true: always enable
# - false: always disable
NVML ?= auto
# Detect glibc host for local linux/amd64 builds.
HOST_GLIBC := $(shell \
if [ "$(OS)" = "linux" ] && [ "$(ARCH)" = "amd64" ]; then \
for p in /lib64/ld-linux-x86-64.so.2 /lib/x86_64-linux-gnu/ld-linux-x86-64.so.2 /lib/ld-linux-x86-64.so.2; do \
[ -e "$$p" ] && { echo true; exit 0; }; \
done; \
if command -v ldd >/dev/null 2>&1; then \
if ldd --version 2>&1 | tr '[:upper:]' '[:lower:]' | awk '/gnu libc|glibc/{found=1} END{exit !found}'; then \
echo true; \
else \
echo false; \
fi; \
else \
echo false; \
fi; \
else \
echo false; \
fi)
# Enable glibc build tag for NVML on supported Linux builds.
AGENT_GO_TAGS :=
ifeq ($(NVML),true)
AGENT_GO_TAGS := -tags glibc
else ifeq ($(NVML),auto)
ifeq ($(HOST_GLIBC),true)
AGENT_GO_TAGS := -tags glibc
endif
endif
# Set executable extension based on target OS # Set executable extension based on target OS
EXE_EXT := $(if $(filter windows,$(OS)),.exe,) EXE_EXT := $(if $(filter windows,$(OS)),.exe,)
@@ -51,8 +17,9 @@ clean:
lint: lint:
golangci-lint run golangci-lint run
test: export GOEXPERIMENT=synctest
test: test:
go test -tags='testing no_ui' ./... go test -tags=testing ./...
tidy: tidy:
go mod tidy go mod tidy
@@ -87,7 +54,7 @@ fetch-smartctl-conditional:
# Update build-agent to include conditional .NET build # Update build-agent to include conditional .NET build
build-agent: tidy build-dotnet-conditional fetch-smartctl-conditional build-agent: tidy build-dotnet-conditional fetch-smartctl-conditional
GOOS=$(OS) GOARCH=$(ARCH) go build $(AGENT_GO_TAGS) -o ./build/beszel-agent_$(OS)_$(ARCH)$(EXE_EXT) -ldflags "-w -s" ./internal/cmd/agent GOOS=$(OS) GOARCH=$(ARCH) go build -o ./build/beszel-agent_$(OS)_$(ARCH)$(EXE_EXT) -ldflags "-w -s" ./internal/cmd/agent
build-hub: tidy $(if $(filter false,$(SKIP_WEB)),build-web-ui) build-hub: tidy $(if $(filter false,$(SKIP_WEB)),build-web-ui)
GOOS=$(OS) GOARCH=$(ARCH) go build -o ./build/beszel_$(OS)_$(ARCH)$(EXE_EXT) -ldflags "-w -s" ./internal/cmd/hub GOOS=$(OS) GOARCH=$(ARCH) go build -o ./build/beszel_$(OS)_$(ARCH)$(EXE_EXT) -ldflags "-w -s" ./internal/cmd/hub
@@ -123,9 +90,9 @@ dev-hub:
dev-agent: dev-agent:
@if command -v entr >/dev/null 2>&1; then \ @if command -v entr >/dev/null 2>&1; then \
find ./internal/cmd/agent/*.go ./agent/*.go | entr -r go run $(AGENT_GO_TAGS) github.com/henrygd/beszel/internal/cmd/agent; \ find ./internal/cmd/agent/*.go ./agent/*.go | entr -r go run github.com/henrygd/beszel/internal/cmd/agent; \
else \ else \
go run $(AGENT_GO_TAGS) github.com/henrygd/beszel/internal/cmd/agent; \ go run github.com/henrygd/beszel/internal/cmd/agent; \
fi fi
build-dotnet: build-dotnet:
+2 -4
View File
@@ -2,8 +2,6 @@
## Reporting a Vulnerability ## Reporting a Vulnerability
**PLEASE ONLY USE SECURITY ADVISORIES FOR REAL HIGH SEVERITY VULNERABILITIES.** If you find a vulnerability in the latest version, please [submit a private advisory](https://github.com/henrygd/beszel/security/advisories/new).
If you find a vulnerability in the latest version, and it is not high severity, open an issue instead of an advisory. If it's low severity (use best judgement) you may open an issue instead of an advisory.
I am overwhelmed with advisories, often erroneous, which are clearly found and written by AI. I don't have the capacity to review all of them.
+58 -53
View File
@@ -5,7 +5,11 @@
package agent package agent
import ( import (
"crypto/sha256"
"encoding/hex"
"log/slog" "log/slog"
"os"
"path/filepath"
"strings" "strings"
"sync" "sync"
"time" "time"
@@ -13,14 +17,12 @@ import (
"github.com/gliderlabs/ssh" "github.com/gliderlabs/ssh"
"github.com/henrygd/beszel" "github.com/henrygd/beszel"
"github.com/henrygd/beszel/agent/deltatracker" "github.com/henrygd/beszel/agent/deltatracker"
"github.com/henrygd/beszel/agent/utils"
"github.com/henrygd/beszel/internal/common" "github.com/henrygd/beszel/internal/common"
"github.com/henrygd/beszel/internal/entities/system" "github.com/henrygd/beszel/internal/entities/system"
"github.com/shirou/gopsutil/v4/host"
gossh "golang.org/x/crypto/ssh" gossh "golang.org/x/crypto/ssh"
) )
const defaultDataCacheTimeMs uint16 = 60_000
type Agent struct { type Agent struct {
sync.Mutex // Used to lock agent while collecting data sync.Mutex // Used to lock agent while collecting data
debug bool // true if LOG_LEVEL is set to debug debug bool // true if LOG_LEVEL is set to debug
@@ -38,7 +40,6 @@ type Agent struct {
sensorConfig *SensorConfig // Sensors config sensorConfig *SensorConfig // Sensors config
systemInfo system.Info // Host system info (dynamic) systemInfo system.Info // Host system info (dynamic)
systemDetails system.Details // Host system details (static, once-per-connection) systemDetails system.Details // Host system details (static, once-per-connection)
detailsDirty bool // Whether system details have changed and need to be resent
gpuManager *GPUManager // Manages GPU data gpuManager *GPUManager // Manages GPU data
cache *systemDataCache // Cache for system stats based on cache time cache *systemDataCache // Cache for system stats based on cache time
connectionManager *ConnectionManager // Channel to signal connection events connectionManager *ConnectionManager // Channel to signal connection events
@@ -48,8 +49,6 @@ type Agent struct {
keys []gossh.PublicKey // SSH public keys keys []gossh.PublicKey // SSH public keys
smartManager *SmartManager // Manages SMART data smartManager *SmartManager // Manages SMART data
systemdManager *systemdManager // Manages systemd services systemdManager *systemdManager // Manages systemd services
monitorManager *MonitorManager // Manages network monitors
storagePoolManager *StoragePoolManager // Manages storage pool and dataset data
} }
// NewAgent creates a new agent with the given data directory for persisting data. // NewAgent creates a new agent with the given data directory for persisting data.
@@ -66,18 +65,18 @@ func NewAgent(dataDir ...string) (agent *Agent, err error) {
agent.netIoStats = make(map[uint16]system.NetIoStats) agent.netIoStats = make(map[uint16]system.NetIoStats)
agent.netInterfaceDeltaTrackers = make(map[uint16]*deltatracker.DeltaTracker[string, uint64]) agent.netInterfaceDeltaTrackers = make(map[uint16]*deltatracker.DeltaTracker[string, uint64])
agent.dataDir, err = GetDataDir(dataDir...) agent.dataDir, err = getDataDir(dataDir...)
if err != nil { if err != nil {
slog.Warn("Data directory not found") slog.Warn("Data directory not found")
} else { } else {
slog.Info("Data directory", "path", agent.dataDir) slog.Info("Data directory", "path", agent.dataDir)
} }
agent.memCalc, _ = utils.GetEnv("MEM_CALC") agent.memCalc, _ = GetEnv("MEM_CALC")
agent.sensorConfig = agent.newSensorConfig() agent.sensorConfig = agent.newSensorConfig()
// Parse disk usage cache duration (e.g., "15m", "1h") to avoid waking sleeping disks // Parse disk usage cache duration (e.g., "15m", "1h") to avoid waking sleeping disks
if diskUsageCache, exists := utils.GetEnv("DISK_USAGE_CACHE"); exists { if diskUsageCache, exists := GetEnv("DISK_USAGE_CACHE"); exists {
if duration, err := time.ParseDuration(diskUsageCache); err == nil { if duration, err := time.ParseDuration(diskUsageCache); err == nil {
agent.diskUsageCacheDuration = duration agent.diskUsageCacheDuration = duration
slog.Info("DISK_USAGE_CACHE", "duration", duration) slog.Info("DISK_USAGE_CACHE", "duration", duration)
@@ -87,7 +86,7 @@ func NewAgent(dataDir ...string) (agent *Agent, err error) {
} }
// Set up slog with a log level determined by the LOG_LEVEL env var // Set up slog with a log level determined by the LOG_LEVEL env var
if logLevelStr, exists := utils.GetEnv("LOG_LEVEL"); exists { if logLevelStr, exists := GetEnv("LOG_LEVEL"); exists {
switch strings.ToLower(logLevelStr) { switch strings.ToLower(logLevelStr) {
case "debug": case "debug":
agent.debug = true agent.debug = true
@@ -102,13 +101,13 @@ func NewAgent(dataDir ...string) (agent *Agent, err error) {
slog.Debug(beszel.Version) slog.Debug(beszel.Version)
// initialize docker manager // initialize docker manager
agent.dockerManager = newDockerManager(agent) agent.dockerManager = newDockerManager()
// initialize system info // initialize system info
agent.refreshSystemDetails() agent.refreshSystemDetails()
// SMART_INTERVAL env var to update smart data at this interval // SMART_INTERVAL env var to update smart data at this interval
if smartIntervalEnv, exists := utils.GetEnv("SMART_INTERVAL"); exists { if smartIntervalEnv, exists := GetEnv("SMART_INTERVAL"); exists {
if duration, err := time.ParseDuration(smartIntervalEnv); err == nil && duration > 0 { if duration, err := time.ParseDuration(smartIntervalEnv); err == nil && duration > 0 {
agent.systemDetails.SmartInterval = duration agent.systemDetails.SmartInterval = duration
slog.Info("SMART_INTERVAL", "duration", duration) slog.Info("SMART_INTERVAL", "duration", duration)
@@ -123,22 +122,6 @@ func NewAgent(dataDir ...string) (agent *Agent, err error) {
// initialize handler registry // initialize handler registry
agent.handlerRegistry = NewHandlerRegistry() agent.handlerRegistry = NewHandlerRegistry()
// initialize monitor manager
agent.monitorManager = newMonitorManager()
agent.storagePoolManager = newStoragePoolManager()
// Retain ZFS_INTERVAL for the shared storage pool detail refresh interval.
if zfsIntervalEnv, exists := utils.GetEnv("ZFS_INTERVAL"); exists {
if duration, err := time.ParseDuration(zfsIntervalEnv); err == nil && duration > 0 {
agent.storagePoolManager.detailInterval = duration
agent.systemDetails.ZfsInterval = duration
slog.Info("ZFS_INTERVAL", "duration", duration)
} else {
slog.Warn("Invalid ZFS_INTERVAL", "err", err)
}
}
// initialize disk info // initialize disk info
agent.initializeDiskInfo() agent.initializeDiskInfo()
@@ -163,12 +146,21 @@ func NewAgent(dataDir ...string) (agent *Agent, err error) {
// if debugging, print stats // if debugging, print stats
if agent.debug { if agent.debug {
slog.Debug("Stats", "data", agent.gatherStats(common.DataRequestOptions{CacheTimeMs: defaultDataCacheTimeMs, IncludeDetails: true})) slog.Debug("Stats", "data", agent.gatherStats(common.DataRequestOptions{CacheTimeMs: 60_000, IncludeDetails: true}))
} }
return agent, nil return agent, nil
} }
// GetEnv retrieves an environment variable with a "BESZEL_AGENT_" prefix, or falls back to the unprefixed key.
func GetEnv(key string) (value string, exists bool) {
if value, exists = os.LookupEnv("BESZEL_AGENT_" + key); exists {
return value, exists
}
// Fallback to the old unprefixed key
return os.LookupEnv(key)
}
func (a *Agent) gatherStats(options common.DataRequestOptions) *system.CombinedData { func (a *Agent) gatherStats(options common.DataRequestOptions) *system.CombinedData {
a.Lock() a.Lock()
defer a.Unlock() defer a.Unlock()
@@ -185,6 +177,11 @@ func (a *Agent) gatherStats(options common.DataRequestOptions) *system.CombinedD
Info: a.systemInfo, Info: a.systemInfo,
} }
// Include static system details only when requested
if options.IncludeDetails {
data.Details = &a.systemDetails
}
// slog.Info("System data", "data", data, "cacheTimeMs", cacheTimeMs) // slog.Info("System data", "data", data, "cacheTimeMs", cacheTimeMs)
if a.dockerManager != nil { if a.dockerManager != nil {
@@ -196,13 +193,8 @@ func (a *Agent) gatherStats(options common.DataRequestOptions) *system.CombinedD
} }
} }
if a.monitorManager != nil {
data.Monitors = a.monitorManager.GetResults(cacheTimeMs)
slog.Debug("Monitors", "data", data.Monitors)
}
// skip updating systemd services if cache time is not the default 60sec interval // skip updating systemd services if cache time is not the default 60sec interval
if a.systemdManager != nil && cacheTimeMs == defaultDataCacheTimeMs { if a.systemdManager != nil && cacheTimeMs == 60_000 {
totalCount := uint16(a.systemdManager.getServiceStatsCount()) totalCount := uint16(a.systemdManager.getServiceStatsCount())
if totalCount > 0 { if totalCount > 0 {
numFailed := a.systemdManager.getFailedServiceCount() numFailed := a.systemdManager.getFailedServiceCount()
@@ -210,25 +202,13 @@ func (a *Agent) gatherStats(options common.DataRequestOptions) *system.CombinedD
} }
if a.systemdManager.hasFreshStats { if a.systemdManager.hasFreshStats {
data.SystemdServices = a.systemdManager.getServiceStats(nil, false) data.SystemdServices = a.systemdManager.getServiceStats(nil, false)
data.SystemdServicesUpdated = true
// Preserve an explicit zero count so the hub can distinguish a fresh
// empty snapshot from a response that omitted systemd data.
if totalCount == 0 {
data.Info.Services = []uint16{0, 0}
}
} }
} }
data.Stats.ExtraFs = make(map[string]*system.FsStats) data.Stats.ExtraFs = make(map[string]*system.FsStats)
data.Info.ExtraFsPct = make(map[string]float64) data.Info.ExtraFsPct = make(map[string]float64)
for name, stats := range a.fsStats { for name, stats := range a.fsStats {
if stats.Root { if !stats.Root && stats.DiskTotal > 0 {
if stats.Name != "" {
data.Info.RootDiskName = stats.Name
}
continue
}
if stats.DiskTotal > 0 {
// Use custom name if available, otherwise use device name // Use custom name if available, otherwise use device name
key := name key := name
if stats.Name != "" { if stats.Name != "" {
@@ -237,7 +217,7 @@ func (a *Agent) gatherStats(options common.DataRequestOptions) *system.CombinedD
data.Stats.ExtraFs[key] = stats data.Stats.ExtraFs[key] = stats
// Add percentages to Info struct for dashboard // Add percentages to Info struct for dashboard
if stats.DiskTotal > 0 { if stats.DiskTotal > 0 {
pct := utils.TwoDecimals((stats.DiskUsed / stats.DiskTotal) * 100) pct := twoDecimals((stats.DiskUsed / stats.DiskTotal) * 100)
data.Info.ExtraFsPct[key] = pct data.Info.ExtraFsPct[key] = pct
} }
} }
@@ -245,16 +225,41 @@ func (a *Agent) gatherStats(options common.DataRequestOptions) *system.CombinedD
slog.Debug("Extra FS", "data", data.Stats.ExtraFs) slog.Debug("Extra FS", "data", data.Stats.ExtraFs)
a.cache.Set(data, cacheTimeMs) a.cache.Set(data, cacheTimeMs)
return data
return a.attachSystemDetails(data, cacheTimeMs, options.IncludeDetails)
} }
// Start initializes and starts the agent with optional WebSocket connection // StartAgent initializes and starts the agent with optional WebSocket connection
func (a *Agent) Start(serverOptions ServerOptions) error { func (a *Agent) Start(serverOptions ServerOptions) error {
a.keys = serverOptions.Keys a.keys = serverOptions.Keys
return a.connectionManager.Start(serverOptions) return a.connectionManager.Start(serverOptions)
} }
func (a *Agent) getFingerprint() string { func (a *Agent) getFingerprint() string {
return GetFingerprint(a.dataDir, a.systemDetails.Hostname, a.systemDetails.CpuModel) // first look for a fingerprint in the data directory
if a.dataDir != "" {
if fp, err := os.ReadFile(filepath.Join(a.dataDir, "fingerprint")); err == nil {
return string(fp)
}
}
// if no fingerprint is found, generate one
fingerprint, err := host.HostID()
// we ignore a commonly known "product_uuid" known not to be unique
if err != nil || fingerprint == "" || fingerprint == "03000200-0400-0500-0006-000700080009" {
fingerprint = a.systemDetails.Hostname + a.systemDetails.CpuModel
}
// hash fingerprint
sum := sha256.Sum256([]byte(fingerprint))
fingerprint = hex.EncodeToString(sum[:24])
// save fingerprint to data directory
if a.dataDir != "" {
err = os.WriteFile(filepath.Join(a.dataDir, "fingerprint"), []byte(fingerprint), 0644)
if err != nil {
slog.Warn("Failed to save fingerprint", "err", err)
}
}
return fingerprint
} }
+1
View File
@@ -1,4 +1,5 @@
//go:build testing //go:build testing
// +build testing
package agent package agent
+1
View File
@@ -1,4 +1,5 @@
//go:build testing //go:build testing
// +build testing
package agent package agent
+67 -56
View File
@@ -1,73 +1,84 @@
// Package battery provides battery information for the host and connected devices. //go:build !freebsd
// Package battery provides functions to check if the system has a battery and to get the battery stats.
package battery package battery
import ( import (
"errors" "errors"
"sort" "log/slog"
"strconv" "math"
"strings"
"github.com/distatus/battery"
) )
const ( var (
stateUnknown uint8 = iota systemHasBattery = false
stateEmpty haveCheckedBattery = false
stateFull
stateCharging
stateDischarging
stateIdle
) )
// Battery is a readable battery reported by the operating system. // HasReadableBattery checks if the system has a battery and returns true if it does.
type Battery struct { func HasReadableBattery() bool {
Name string if haveCheckedBattery {
Percent uint8 return systemHasBattery
State uint8
FullChargeCapacity uint64
HasFullChargeCapacity bool
System bool
}
var errNoBatteries = errors.New("no readable batteries")
// normalizeBatteries supplies stable fallback names and disambiguates duplicates.
func normalizeBatteries(batteries []Battery) []Battery {
nameCounts := make(map[string]int, len(batteries))
for i := range batteries {
// Names come from firmware (e.g. sysfs model_name) and are not guaranteed to
// be valid UTF-8. Invalid bytes are rejected when the hub decodes the CBOR
// payload, which drops every metric for the system, so strip them here.
name := strings.TrimSpace(strings.ToValidUTF8(batteries[i].Name, ""))
if name == "" {
name = "Battery " + strconv.Itoa(i+1)
}
nameCounts[name]++
if nameCounts[name] > 1 {
name += " (" + strconv.Itoa(nameCounts[name]) + ")"
}
batteries[i].Name = name
} }
return batteries haveCheckedBattery = true
batteries, err := battery.GetAll()
for _, bat := range batteries {
if bat != nil && (bat.Full > 0 || bat.Design > 0) {
systemHasBattery = true
break
}
}
if !systemHasBattery {
slog.Debug("No battery found", "err", err)
}
return systemHasBattery
} }
// Primary returns the representative battery. Reported full-charge capacity wins, // GetBatteryStats returns the current battery percent and charge state
// then system-scoped devices, then name for deterministic ties. // percent = (current charge of all batteries) / (sum of designed/full capacity of all batteries)
func Primary(batteries []Battery) (Battery, bool) { func GetBatteryStats() (batteryPercent uint8, batteryState uint8, err error) {
if !HasReadableBattery() {
return batteryPercent, batteryState, errors.ErrUnsupported
}
batteries, err := battery.GetAll()
// we'll handle errors later by skipping batteries with errors, rather
// than skipping everything because of the presence of some errors.
if len(batteries) == 0 { if len(batteries) == 0 {
return Battery{}, false return batteryPercent, batteryState, errors.New("no batteries")
} }
ordered := append([]Battery(nil), batteries...)
sort.SliceStable(ordered, func(i, j int) bool { totalCapacity := float64(0)
a, b := ordered[i], ordered[j] totalCharge := float64(0)
if a.HasFullChargeCapacity != b.HasFullChargeCapacity { errs, partialErrs := err.(battery.Errors)
return a.HasFullChargeCapacity
batteryState = math.MaxUint8
for i, bat := range batteries {
if partialErrs && errs[i] != nil {
// if there were some errors, like missing data, skip it
continue
} }
if a.HasFullChargeCapacity && a.FullChargeCapacity != b.FullChargeCapacity { if bat == nil || bat.Full == 0 {
return a.FullChargeCapacity > b.FullChargeCapacity // skip batteries with no capacity. Charge is unlikely to ever be zero, but
// we can't guarantee that, so don't skip based on charge.
continue
} }
if a.System != b.System { totalCapacity += bat.Full
return a.System totalCharge += bat.Current
if bat.State.Raw >= 0 {
batteryState = uint8(bat.State.Raw)
} }
return a.Name < b.Name }
})
return ordered[0], true if totalCapacity == 0 || batteryState == math.MaxUint8 {
// for macs there's sometimes a ghost battery with 0 capacity
// https://github.com/distatus/battery/issues/34
// Instead of skipping over those batteries, we'll check for total 0 capacity
// and return an error. This also prevents a divide by zero.
return batteryPercent, batteryState, errors.New("no battery capacity")
}
batteryPercent = uint8(totalCharge / totalCapacity * 100)
return batteryPercent, batteryState, nil
} }
-76
View File
@@ -1,76 +0,0 @@
//go:build darwin
package battery
import (
"os/exec"
"howett.net/plist"
)
type macBattery struct {
CurrentCapacity int `plist:"CurrentCapacity"`
MaxCapacity int `plist:"MaxCapacity"`
FullyCharged bool `plist:"FullyCharged"`
IsCharging bool `plist:"IsCharging"`
ExternalConnected bool `plist:"ExternalConnected"`
}
func readMacBatteries() ([]macBattery, error) {
out, err := exec.Command("ioreg", "-n", "AppleSmartBattery", "-r", "-a").Output()
if err != nil {
return nil, err
}
if len(out) == 0 {
return nil, nil
}
var batteries []macBattery
if _, err := plist.Unmarshal(out, &batteries); err != nil {
return nil, err
}
return batteries, nil
}
func HasReadableBattery() bool {
batteries, _ := GetBatteryStats()
return len(batteries) > 0
}
// GetBatteryStats returns every readable battery reported by macOS.
func GetBatteryStats() ([]Battery, error) {
batteries, err := readMacBatteries()
if err != nil {
return nil, err
}
if len(batteries) == 0 {
return nil, errNoBatteries
}
result := make([]Battery, 0, len(batteries))
for _, bat := range batteries {
if bat.MaxCapacity <= 0 {
// skip ghost batteries with 0 capacity
// https://github.com/distatus/battery/issues/34
continue
}
percent := min(max(float64(bat.CurrentCapacity)/float64(bat.MaxCapacity)*100, 0), 100)
state := stateUnknown
switch {
case !bat.ExternalConnected:
state = stateDischarging
case bat.IsCharging:
state = stateCharging
case bat.CurrentCapacity == 0:
state = stateEmpty
case !bat.FullyCharged:
state = stateIdle
default:
state = stateFull
}
result = append(result, Battery{Name: "Primary", Percent: uint8(percent), State: state,
FullChargeCapacity: uint64(bat.MaxCapacity), HasFullChargeCapacity: true, System: true})
}
if len(result) == 0 {
return nil, errNoBatteries
}
return normalizeBatteries(result), nil
}
+13
View File
@@ -0,0 +1,13 @@
//go:build freebsd
package battery
import "errors"
func HasReadableBattery() bool {
return false
}
func GetBatteryStats() (uint8, uint8, error) {
return 0, 0, errors.ErrUnsupported
}
-85
View File
@@ -1,85 +0,0 @@
//go:build linux
package battery
import (
"os"
"path/filepath"
"strconv"
"github.com/henrygd/beszel/agent/utils"
)
var batteryRoot = "/sys/class/power_supply"
// HasReadableBattery reports whether collection currently finds a readable battery.
func HasReadableBattery() bool {
batteries, _ := GetBatteryStats()
return len(batteries) > 0
}
func parseSysfsState(status string) uint8 {
switch status {
case "Empty":
return stateEmpty
case "Full":
return stateFull
case "Charging":
return stateCharging
case "Discharging":
return stateDischarging
case "Not charging":
return stateIdle
default:
return stateUnknown
}
}
// GetBatteryStats re-enumerates power supplies and returns every readable battery.
func GetBatteryStats() ([]Battery, error) {
entries, err := os.ReadDir(batteryRoot)
if err != nil {
return nil, err
}
batteries := make([]Battery, 0, len(entries))
for _, entry := range entries {
path := filepath.Join(batteryRoot, entry.Name())
if utils.ReadStringFile(filepath.Join(path, "type")) != "Battery" {
continue
}
capStr, ok := utils.ReadStringFileOK(filepath.Join(path, "capacity"))
if !ok {
continue
}
cap, parseErr := strconv.Atoi(capStr)
if parseErr != nil {
continue
}
cap = min(max(cap, 0), 100)
name := utils.ReadStringFile(filepath.Join(path, "model_name"))
if name == "" {
name = utils.ReadStringFile(filepath.Join(path, "model"))
}
if name == "" {
name = entry.Name()
}
battery := Battery{
Name: name,
Percent: uint8(cap),
State: parseSysfsState(utils.ReadStringFile(filepath.Join(path, "status"))),
System: utils.ReadStringFile(filepath.Join(path, "scope")) != "Device",
}
for _, fullName := range []string{"charge_full", "energy_full"} {
if parsed, ok := utils.ReadUintFile(filepath.Join(path, fullName)); ok && parsed > 0 {
battery.FullChargeCapacity = parsed
battery.HasFullChargeCapacity = true
break
}
}
batteries = append(batteries, battery)
}
if len(batteries) == 0 {
return nil, errNoBatteries
}
return normalizeBatteries(batteries), nil
}
-109
View File
@@ -1,109 +0,0 @@
//go:build testing && linux
package battery
import (
"os"
"path/filepath"
"testing"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
type fakeBattery struct{ id, name, capacity, status, full, scope string }
func setupFakeSysfs(t *testing.T) (string, func(fakeBattery)) {
t.Helper()
root := t.TempDir()
previousRoot := batteryRoot
batteryRoot = root
t.Cleanup(func() { batteryRoot = previousRoot })
write := func(path, value string) {
t.Helper()
require.NoError(t, os.MkdirAll(filepath.Dir(path), 0o755))
require.NoError(t, os.WriteFile(path, []byte(value), 0o644))
}
add := func(b fakeBattery) {
t.Helper()
dir := filepath.Join(root, b.id)
write(filepath.Join(dir, "type"), "Battery")
if b.capacity != "" {
write(filepath.Join(dir, "capacity"), b.capacity)
}
write(filepath.Join(dir, "status"), b.status)
if b.name != "" {
write(filepath.Join(dir, "model_name"), b.name)
}
if b.full != "" {
write(filepath.Join(dir, "energy_full"), b.full)
}
if b.scope != "" {
write(filepath.Join(dir, "scope"), b.scope)
}
}
return root, add
}
func TestParseSysfsState(t *testing.T) {
assert.Equal(t, stateEmpty, parseSysfsState("Empty"))
assert.Equal(t, stateFull, parseSysfsState("Full"))
assert.Equal(t, stateCharging, parseSysfsState("Charging"))
assert.Equal(t, stateDischarging, parseSysfsState("Discharging"))
assert.Equal(t, stateIdle, parseSysfsState("Not charging"))
assert.Equal(t, stateUnknown, parseSysfsState("SomethingElse"))
}
func TestGetBatteryStatsMultipleNamedAndPrimary(t *testing.T) {
_, add := setupFakeSysfs(t)
add(fakeBattery{id: "BAT0", name: "Primary", capacity: "105", status: "Charging", full: "5000", scope: "System"})
add(fakeBattery{id: "hidpp_battery_0", name: "MX Keys S", capacity: "55", status: "Unknown", full: "900", scope: "Device"})
batteries, err := GetBatteryStats()
require.NoError(t, err)
require.Len(t, batteries, 2)
assert.Equal(t, "Primary", batteries[0].Name)
assert.Equal(t, uint8(100), batteries[0].Percent)
assert.Equal(t, stateUnknown, batteries[1].State)
primary, ok := Primary(batteries)
require.True(t, ok)
assert.Equal(t, "Primary", primary.Name)
}
func TestGetBatteryStatsFallbackDuplicatesAndUnreadable(t *testing.T) {
root, add := setupFakeSysfs(t)
add(fakeBattery{id: "BAT0", name: "Keyboard", capacity: "80", status: "Discharging"})
add(fakeBattery{id: "BAT1", name: "Keyboard", capacity: "-4", status: "SomethingWeird"})
add(fakeBattery{id: "BAT2", capacity: "not-a-number", status: "Charging"})
add(fakeBattery{id: "BAT3", capacity: "42", status: "Full"})
ac := filepath.Join(root, "AC0")
require.NoError(t, os.MkdirAll(ac, 0o755))
require.NoError(t, os.WriteFile(filepath.Join(ac, "type"), []byte("Mains"), 0o644))
batteries, err := GetBatteryStats()
require.NoError(t, err)
require.Len(t, batteries, 3)
assert.Equal(t, "Keyboard", batteries[0].Name)
assert.Equal(t, "Keyboard (2)", batteries[1].Name)
assert.Equal(t, uint8(0), batteries[1].Percent)
assert.Equal(t, "BAT3", batteries[2].Name)
}
func TestGetBatteryStatsHotPlugReenumerates(t *testing.T) {
_, add := setupFakeSysfs(t)
_, err := GetBatteryStats()
assert.Error(t, err)
assert.False(t, HasReadableBattery())
add(fakeBattery{id: "BAT0", capacity: "64", status: "Discharging"})
batteries, err := GetBatteryStats()
require.NoError(t, err)
assert.True(t, HasReadableBattery())
require.Len(t, batteries, 1)
assert.Equal(t, uint8(64), batteries[0].Percent)
}
func TestGetBatteryStatsNoReadableCapacity(t *testing.T) {
_, add := setupFakeSysfs(t)
add(fakeBattery{id: "BAT0", status: "Charging"})
_, err := GetBatteryStats()
assert.Error(t, err)
assert.False(t, HasReadableBattery())
}
-13
View File
@@ -1,13 +0,0 @@
//go:build !darwin && !linux && !windows
package battery
import "errors"
func HasReadableBattery() bool {
return false
}
func GetBatteryStats() ([]Battery, error) {
return nil, errors.ErrUnsupported
}
-48
View File
@@ -1,48 +0,0 @@
package battery
import (
"testing"
"unicode/utf8"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
func TestPrimarySelection(t *testing.T) {
tests := []struct {
name string
bats []Battery
want string
}{
{"largest reported capacity", []Battery{{Name: "Small", FullChargeCapacity: 20, HasFullChargeCapacity: true, System: true}, {Name: "Large", FullChargeCapacity: 80, HasFullChargeCapacity: true}}, "Large"},
{"reported ranks over missing", []Battery{{Name: "Unknown", System: true}, {Name: "Known", FullChargeCapacity: 1, HasFullChargeCapacity: true}}, "Known"},
{"system wins capacity tie", []Battery{{Name: "Peripheral", FullChargeCapacity: 50, HasFullChargeCapacity: true}, {Name: "System", FullChargeCapacity: 50, HasFullChargeCapacity: true, System: true}}, "System"},
{"name resolves final tie", []Battery{{Name: "Zed"}, {Name: "Alpha"}}, "Alpha"},
}
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
got, ok := Primary(tt.bats)
require.True(t, ok)
assert.Equal(t, tt.want, got.Name)
})
}
_, ok := Primary(nil)
assert.False(t, ok)
}
func TestNormalizeBatteriesFallbackNames(t *testing.T) {
bats := normalizeBatteries([]Battery{{}, {}, {Name: "Mouse"}, {Name: "Mouse"}})
assert.Equal(t, []string{"Battery 1", "Battery 2", "Mouse", "Mouse (2)"}, []string{bats[0].Name, bats[1].Name, bats[2].Name, bats[3].Name})
}
func TestNormalizeBatteriesStripsInvalidUTF8(t *testing.T) {
// Firmware occasionally reports names that are not valid UTF-8 (a ThinkPad
// reporting "LNV-5B11K63024@\xd0" in model_name is a real example).
bats := normalizeBatteries([]Battery{{Name: "LNV-5B11K63024@\xd0"}, {Name: "\xff\xfe"}})
assert.Equal(t, "LNV-5B11K63024@", bats[0].Name)
// A name made up entirely of invalid bytes falls back to the generic name.
assert.Equal(t, "Battery 2", bats[1].Name)
for _, b := range bats {
assert.True(t, utf8.ValidString(b.Name))
}
}
-291
View File
@@ -1,291 +0,0 @@
//go:build windows
// Most of the Windows battery code is based on
// distatus/battery by Karol 'Kenji Takahashi' Woźniak
package battery
import (
"errors"
"syscall"
"unsafe"
"golang.org/x/sys/windows"
)
type batteryQueryInformation struct {
BatteryTag uint32
InformationLevel int32
AtRate int32
}
type batteryInformation struct {
Capabilities uint32
Technology uint8
Reserved [3]uint8
Chemistry [4]uint8
DesignedCapacity uint32
FullChargedCapacity uint32
DefaultAlert1 uint32
DefaultAlert2 uint32
CriticalBias uint32
CycleCount uint32
}
type batteryWaitStatus struct {
BatteryTag uint32
Timeout uint32
PowerState uint32
LowCapacity uint32
HighCapacity uint32
}
type batteryStatus struct {
PowerState uint32
Capacity uint32
Voltage uint32
Rate int32
}
type winGUID struct {
Data1 uint32
Data2 uint16
Data3 uint16
Data4 [8]byte
}
type spDeviceInterfaceData struct {
cbSize uint32
InterfaceClassGuid winGUID
Flags uint32
Reserved uint
}
var guidDeviceBattery = winGUID{
0x72631e54,
0x78A4,
0x11d0,
[8]byte{0xbc, 0xf7, 0x00, 0xaa, 0x00, 0xb7, 0xb3, 0x2a},
}
var (
setupapi = &windows.LazyDLL{Name: "setupapi.dll", System: true}
setupDiGetClassDevsW = setupapi.NewProc("SetupDiGetClassDevsW")
setupDiEnumDeviceInterfaces = setupapi.NewProc("SetupDiEnumDeviceInterfaces")
setupDiGetDeviceInterfaceDetailW = setupapi.NewProc("SetupDiGetDeviceInterfaceDetailW")
setupDiDestroyDeviceInfoList = setupapi.NewProc("SetupDiDestroyDeviceInfoList")
)
// winBatteryGet reads one battery by index.
// Returns error == errNotFound when there are no more batteries.
var errNotFound = errors.New("no more batteries")
func setupDiSetup(proc *windows.LazyProc, nargs, a1, a2, a3, a4, a5, a6 uintptr) (uintptr, error) {
_ = nargs
r1, _, errno := syscall.SyscallN(proc.Addr(), a1, a2, a3, a4, a5, a6)
if windows.Handle(r1) == windows.InvalidHandle {
if errno != 0 {
return 0, error(errno)
}
return 0, syscall.EINVAL
}
return r1, nil
}
func setupDiCall(proc *windows.LazyProc, nargs, a1, a2, a3, a4, a5, a6 uintptr) syscall.Errno {
_ = nargs
r1, _, errno := syscall.SyscallN(proc.Addr(), a1, a2, a3, a4, a5, a6)
if r1 == 0 {
if errno != 0 {
return errno
}
return syscall.EINVAL
}
return 0
}
func readWinBatteryState(powerState uint32) uint8 {
switch {
case powerState&0x00000004 != 0:
return stateCharging
case powerState&0x00000008 != 0:
return stateEmpty
case powerState&0x00000002 != 0:
return stateDischarging
case powerState&0x00000001 != 0:
return stateFull
default:
return stateUnknown
}
}
func winBatteryGet(idx int) (Battery, error) {
hdev, err := setupDiSetup(
setupDiGetClassDevsW,
4,
uintptr(unsafe.Pointer(&guidDeviceBattery)),
0, 0,
2|16, // DIGCF_PRESENT|DIGCF_DEVICEINTERFACE
0, 0,
)
if err != nil {
return Battery{}, err
}
defer syscall.SyscallN(setupDiDestroyDeviceInfoList.Addr(), hdev)
var did spDeviceInterfaceData
did.cbSize = uint32(unsafe.Sizeof(did))
errno := setupDiCall(
setupDiEnumDeviceInterfaces,
5,
hdev, 0,
uintptr(unsafe.Pointer(&guidDeviceBattery)),
uintptr(idx),
uintptr(unsafe.Pointer(&did)),
0,
)
if errno == 259 { // ERROR_NO_MORE_ITEMS
return Battery{}, errNotFound
}
if errno != 0 {
return Battery{}, errno
}
var cbRequired uint32
errno = setupDiCall(
setupDiGetDeviceInterfaceDetailW,
6,
hdev,
uintptr(unsafe.Pointer(&did)),
0, 0,
uintptr(unsafe.Pointer(&cbRequired)),
0,
)
if errno != 0 && errno != 122 { // ERROR_INSUFFICIENT_BUFFER
return Battery{}, errno
}
didd := make([]uint16, cbRequired/2)
cbSize := (*uint32)(unsafe.Pointer(&didd[0]))
if unsafe.Sizeof(uint(0)) == 8 {
*cbSize = 8
} else {
*cbSize = 6
}
errno = setupDiCall(
setupDiGetDeviceInterfaceDetailW,
6,
hdev,
uintptr(unsafe.Pointer(&did)),
uintptr(unsafe.Pointer(&didd[0])),
uintptr(cbRequired),
uintptr(unsafe.Pointer(&cbRequired)),
0,
)
if errno != 0 {
return Battery{}, errno
}
devicePath := &didd[2:][0]
handle, err := windows.CreateFile(
devicePath,
windows.GENERIC_READ|windows.GENERIC_WRITE,
windows.FILE_SHARE_READ|windows.FILE_SHARE_WRITE,
nil,
windows.OPEN_EXISTING,
windows.FILE_ATTRIBUTE_NORMAL,
0,
)
if err != nil {
return Battery{}, err
}
defer windows.CloseHandle(handle)
var dwOut uint32
var dwWait uint32
var bqi batteryQueryInformation
err = windows.DeviceIoControl(
handle,
2703424, // IOCTL_BATTERY_QUERY_TAG
(*byte)(unsafe.Pointer(&dwWait)),
uint32(unsafe.Sizeof(dwWait)),
(*byte)(unsafe.Pointer(&bqi.BatteryTag)),
uint32(unsafe.Sizeof(bqi.BatteryTag)),
&dwOut, nil,
)
if err != nil || bqi.BatteryTag == 0 {
return Battery{}, errors.New("battery tag not returned")
}
var bi batteryInformation
if err = windows.DeviceIoControl(
handle,
2703428, // IOCTL_BATTERY_QUERY_INFORMATION
(*byte)(unsafe.Pointer(&bqi)),
uint32(unsafe.Sizeof(bqi)),
(*byte)(unsafe.Pointer(&bi)),
uint32(unsafe.Sizeof(bi)),
&dwOut, nil,
); err != nil {
return Battery{}, err
}
// BatteryDeviceName is optional, so retain the deterministic fallback on error.
name := ""
nameQuery := bqi
nameQuery.InformationLevel = 4 // BatteryDeviceName
nameBuffer := make([]uint16, 128)
if err := windows.DeviceIoControl(
handle, 2703428,
(*byte)(unsafe.Pointer(&nameQuery)), uint32(unsafe.Sizeof(nameQuery)),
(*byte)(unsafe.Pointer(&nameBuffer[0])), uint32(len(nameBuffer)*2),
&dwOut, nil,
); err == nil {
name = windows.UTF16ToString(nameBuffer)
}
bws := batteryWaitStatus{BatteryTag: bqi.BatteryTag}
var bs batteryStatus
if err = windows.DeviceIoControl(
handle,
2703436, // IOCTL_BATTERY_QUERY_STATUS
(*byte)(unsafe.Pointer(&bws)),
uint32(unsafe.Sizeof(bws)),
(*byte)(unsafe.Pointer(&bs)),
uint32(unsafe.Sizeof(bs)),
&dwOut, nil,
); err != nil {
return Battery{}, err
}
if bs.Capacity == 0xffffffff || bi.FullChargedCapacity == 0 || bi.FullChargedCapacity == 0xffffffff {
return Battery{}, errors.New("battery capacity unknown")
}
percent := min(float64(bs.Capacity)/float64(bi.FullChargedCapacity)*100, 100)
return Battery{Name: name, Percent: uint8(percent), State: readWinBatteryState(bs.PowerState),
FullChargeCapacity: uint64(bi.FullChargedCapacity), HasFullChargeCapacity: true, System: true}, nil
}
// HasReadableBattery checks if the system has a battery and returns true if it does.
func HasReadableBattery() bool {
batteries, _ := GetBatteryStats()
return len(batteries) > 0
}
// GetBatteryStats returns every readable battery reported by Windows.
func GetBatteryStats() ([]Battery, error) {
batteries := make([]Battery, 0, 2)
for i := 0; ; i++ {
battery, bErr := winBatteryGet(i)
if errors.Is(bErr, errNotFound) {
break
}
if bErr != nil {
continue
}
batteries = append(batteries, battery)
}
if len(batteries) == 0 {
return nil, errNoBatteries
}
return normalizeBatteries(batteries), nil
}
-26
View File
@@ -1,26 +0,0 @@
// Package btrfs reads btrfs filesystem state from sysfs.
package btrfs
// Filesystem is a mounted btrfs filesystem read from /sys/fs/btrfs/<uuid>.
type Filesystem struct {
UUID string // stable filesystem UUID from sysfs
MountID string // kernel filesystem identity for matching monitored mounts
IODevice string // sole member block-device name, empty for multi-device/unknown pools
Name string // label, else first mountpoint, else UUID
Size uint64 // effective usable capacity, or raw member capacity when Raw
Raw bool // capacity and usage are physical bytes, unsuitable for disk alerts
Alloc uint64 // raw bytes allocated to data, metadata and system chunks
Health string // ONLINE, or DEGRADED when a device is missing
NRead uint64 // cumulative bytes read across member devices
NWrite uint64 // cumulative bytes written across member devices
Devices []Device
}
// Device is one member device (devinfo/<devid>) with its error counters.
type Device struct {
Name string // "devid N"; sysfs does not expose the block device path
State string // ONLINE or MISSING
ReadErrs uint64
WriteErrs uint64
CorruptionErrs uint64
}
-285
View File
@@ -1,285 +0,0 @@
//go:build linux
package btrfs
import (
"errors"
"fmt"
"os"
"path/filepath"
"strconv"
"strings"
"unsafe"
"github.com/henrygd/beszel/agent/utils"
"golang.org/x/sys/unix"
)
var (
sysfsPath = "/sys/fs/btrfs"
mountsPath = "/proc/self/mounts"
mountinfoPath = "/proc/self/mountinfo"
mountUUID = MountID
deviceSize = ioctlDeviceSize
filesystemUsage = statfsUsage
)
// Filesystems returns all mounted btrfs filesystems, or nil when there are none.
func Filesystems() ([]Filesystem, error) {
entries, err := os.ReadDir(sysfsPath)
if errors.Is(err, os.ErrNotExist) {
return nil, nil
}
if err != nil {
return nil, err
}
mounts := mountpointsByDevice()
var filesystems []Filesystem
for _, entry := range entries {
if !entry.IsDir() || entry.Name() == "features" {
continue
}
fs, err := readFilesystem(filepath.Join(sysfsPath, entry.Name()), mounts)
if err != nil {
return nil, fmt.Errorf("btrfs %s: %w", entry.Name(), err)
}
filesystems = append(filesystems, fs)
}
return filesystems, nil
}
func readFilesystem(dir string, mounts map[string]string) (Filesystem, error) {
fs := Filesystem{UUID: filepath.Base(dir), Name: utils.ReadStringFile(filepath.Join(dir, "label")), Health: "UNKNOWN"}
for _, kind := range []string{"data", "metadata", "system"} {
if value, ok := utils.ReadUintFile(filepath.Join(dir, "allocation", kind, "disk_used")); ok {
fs.Alloc += value
}
}
// devices/<name> links to the block device's sysfs directory.
devices, err := os.ReadDir(filepath.Join(dir, "devices"))
if err != nil && !errors.Is(err, os.ErrNotExist) {
return fs, err
}
mountpoint := mounts["uuid:"+fs.UUID]
if fs.Name == "" {
fs.Name = mountpoint
}
var backingSize uint64
for _, dev := range devices {
if mountpoint == "" {
mountpoint = mounts[dev.Name()]
}
if fs.Name == "" {
fs.Name = mountpoint
}
devDir := filepath.Join(dir, "devices", dev.Name())
if size, ok := utils.ReadUintFile(filepath.Join(devDir, "size")); ok {
backingSize += size * 512
}
if stat := strings.Fields(utils.ReadStringFile(filepath.Join(devDir, "stat"))); len(stat) >= 7 {
fs.NRead += parseUint(stat[2]) * 512
fs.NWrite += parseUint(stat[6]) * 512
}
}
devids, err := os.ReadDir(filepath.Join(dir, "devinfo"))
if err != nil && !errors.Is(err, os.ErrNotExist) {
return fs, err
}
capacityAvailable := len(devids) > 0
healthKnown := len(devids) > 0
for _, devid := range devids {
devDir := filepath.Join(dir, "devinfo", devid.Name())
// Replacement targets do not add filesystem capacity.
replaceTarget, _ := utils.ReadUintFile(filepath.Join(devDir, "replace_target"))
if replaceTarget != 1 {
devid, err := strconv.ParseUint(devid.Name(), 10, 64)
if err != nil {
return fs, err
}
size, err := deviceSize(mountpoint, devid)
if err != nil {
capacityAvailable = false
}
fs.Size += size
}
dev := Device{Name: "devid " + devid.Name(), State: "ONLINE"}
missing := utils.ReadStringFile(filepath.Join(devDir, "missing"))
if missing != "0" && missing != "1" {
healthKnown = false
dev.State = "UNKNOWN"
}
if missing == "1" {
dev.State = "MISSING"
fs.Health = "DEGRADED"
}
for line := range strings.Lines(utils.ReadStringFile(filepath.Join(devDir, "error_stats"))) {
if fields := strings.Fields(line); len(fields) == 2 {
switch fields[0] {
case "read_errs":
dev.ReadErrs = parseUint(fields[1])
case "write_errs":
dev.WriteErrs = parseUint(fields[1])
case "corruption_errs":
dev.CorruptionErrs = parseUint(fields[1])
}
}
}
fs.Devices = append(fs.Devices, dev)
}
// Use one capacity source for the whole filesystem: device IDs cannot be
// reliably matched to block-device names in sysfs. A partial ioctl result
// must not be added to the complete backing-device total.
if !capacityAvailable {
fs.Size = backingSize
}
if fs.Health != "DEGRADED" && healthKnown {
fs.Health = "ONLINE"
}
fs.MountID = mountUUID(mountpoint)
if len(devices) == 1 && len(devids) == 1 && fs.Health == "ONLINE" {
fs.IODevice = devices[0].Name()
}
fs.Raw = true
if used, available, err := filesystemUsage(mountpoint); err == nil {
// Effective capacity excludes reserved/unavailable space, so Size-Alloc
// is available to applications and the usage ratio matches df.
fs.Size, fs.Alloc, fs.Raw = used+available, used, false
}
if fs.Name == "" {
fs.Name = filepath.Base(dir)
}
return fs, nil
}
// mountpointsByDevice prefers UUID matches from mountinfo and retains source
// device names as a fallback for environments where FS_INFO is unavailable.
func mountpointsByDevice() map[string]string {
mounts := mountpointsByUUID(utils.ReadStringFile(mountinfoPath), mountUUID)
for line := range strings.Lines(utils.ReadStringFile(mountsPath)) {
fields := strings.Fields(line)
if len(fields) < 3 || fields[2] != "btrfs" {
continue
}
device := fields[0]
if resolved, err := filepath.EvalSymlinks(device); err == nil {
device = resolved
}
if _, seen := mounts[filepath.Base(device)]; !seen {
mounts[filepath.Base(device)] = unescapeMountPath(fields[1])
}
}
return mounts
}
func parseUint(s string) uint64 {
n, _ := strconv.ParseUint(s, 10, 64)
return n
}
// ioctlDeviceSize reads Btrfs's recorded device size, which can be smaller
// than the block device after a filesystem resize. BTRFS_IOC_DEV_INFO is
// _IOWR(0x94, 30, struct btrfs_ioctl_dev_info_args), a 4096-byte ABI structure.
func ioctlDeviceSize(mountpoint string, devid uint64) (uint64, error) {
if mountpoint == "" {
return 0, errors.New("no accessible mountpoint")
}
f, err := os.Open(mountpoint)
if err != nil {
return 0, err
}
defer f.Close()
args := struct {
Devid uint64
UUID [16]byte
BytesUsed uint64
TotalBytes uint64
Reserved [4096 - 40]byte
}{Devid: devid}
_, _, errno := unix.Syscall(unix.SYS_IOCTL, f.Fd(), 0xd000941e, uintptr(unsafe.Pointer(&args)))
if errno != 0 {
return 0, errno
}
return args.TotalBytes, nil
}
// The filesystem magic is unsigned even when Statfs_t.Type is int32.
func isBtrfs(stat *unix.Statfs_t) bool {
return uint32(stat.Type) == unix.BTRFS_SUPER_MAGIC
}
func statfsUsage(path string) (used, available uint64, err error) {
if path == "" {
return 0, 0, errors.New("no accessible mountpoint")
}
var stat unix.Statfs_t
if err = unix.Statfs(path, &stat); err != nil {
return
}
if !isBtrfs(&stat) {
return 0, 0, errors.New("mountpoint is not Btrfs")
}
blockSize := uint64(stat.Bsize)
return (stat.Blocks - min(stat.Blocks, stat.Bfree)) * blockSize, min(stat.Blocks, stat.Bavail) * blockSize, nil
}
// MountID returns the filesystem UUID via BTRFS_IOC_FS_INFO. Unlike statfs
// f_fsid, this identity is shared by all subvolumes and bind mounts.
func MountID(path string) string {
if path == "" {
return ""
}
var stat unix.Statfs_t
if unix.Statfs(path, &stat) != nil || !isBtrfs(&stat) {
return ""
}
f, err := os.Open(path)
if err != nil {
return ""
}
defer f.Close()
args := struct {
MaxID uint64
NumDevices uint64
FSID [16]byte
Reserved [992]byte
}{}
// _IOR(0x94, 31, 1024). Reuse the platform's read-direction bits;
// MIPS/PowerPC use a different encoding than asm-generic.
request := uintptr(unix.FS_IOC_GETFLAGS&0xe0000000) | 0x0400941f
_, _, errno := unix.Syscall(unix.SYS_IOCTL, f.Fd(), request, uintptr(unsafe.Pointer(&args)))
if errno != 0 {
return ""
}
id := args.FSID
return fmt.Sprintf("%x-%x-%x-%x-%x", id[:4], id[4:6], id[6:8], id[8:10], id[10:])
}
// Btrfs mountinfo device numbers can be virtual (0:N), so query the UUID
// through the mount instead of comparing those numbers with sysfs block devs.
// Retry another path when a bind mount is inaccessible. Once resolved, reuse
// the result for that mount device to avoid opening every Docker bind mount.
func mountpointsByUUID(mountinfo string, identify func(string) string) map[string]string {
mounts := make(map[string]string)
resolved := make(map[string]bool)
for line := range strings.Lines(mountinfo) {
before, after, ok := strings.Cut(line, " - ")
fields, fs := strings.Fields(before), strings.Fields(after)
if !ok || len(fields) < 6 || len(fs) < 3 || fs[0] != "btrfs" || resolved[fields[2]] {
continue
}
path := unescapeMountPath(fields[4])
uuid := identify(path)
if uuid == "" {
continue
}
resolved[fields[2]] = true
if mounts["uuid:"+uuid] == "" {
mounts["uuid:"+uuid] = path
}
}
return mounts
}
func unescapeMountPath(path string) string {
return strings.NewReplacer(`\040`, " ", `\011`, "\t", `\012`, "\n", `\134`, `\`).Replace(path)
}
-274
View File
@@ -1,274 +0,0 @@
//go:build testing && linux
package btrfs
import (
"os"
"path/filepath"
"strconv"
"testing"
"github.com/henrygd/beszel/agent/utils"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
"golang.org/x/sys/unix"
)
func TestFilesystems(t *testing.T) {
root := t.TempDir()
oldSysfs, oldMounts := sysfsPath, mountsPath
sysfsPath, mountsPath = root, filepath.Join(root, "mounts")
t.Cleanup(func() { sysfsPath, mountsPath = oldSysfs, oldMounts })
fsDir := filepath.Join(root, "1b2c3d4e-0000-0000-0000-000000000000")
write := func(rel, content string) {
path := filepath.Join(fsDir, rel)
require.NoError(t, os.MkdirAll(filepath.Dir(path), 0o755))
require.NoError(t, os.WriteFile(path, []byte(content), 0o644))
}
require.NoError(t, os.MkdirAll(filepath.Join(root, "features"), 0o755))
oldUsage := filesystemUsage
filesystemUsage = func(string) (uint64, uint64, error) { return 0, 0, os.ErrNotExist }
t.Cleanup(func() { filesystemUsage = oldUsage })
oldDeviceSize := deviceSize
t.Cleanup(func() { deviceSize = oldDeviceSize })
deviceSize = func(_ string, devid uint64) (uint64, error) {
value, _ := utils.ReadUintFile(filepath.Join(fsDir, "recorded-size", strconv.FormatUint(devid, 10)))
return value, nil
}
// Recorded member capacities differ from the unchanged backing devices.
write("recorded-size/1", "256000\n")
write("recorded-size/2", "128000\n")
write("label", "tank\n")
write("allocation/data/disk_used", "4096\n")
write("allocation/metadata/disk_used", "2048\n")
write("allocation/system/disk_used", "1024\n")
write("devices/sda/size", "1000\n")
write("devices/sda/stat", "10 0 200 0 20 0 400 0 0 0 0\n")
write("devices/sdb/size", "1000\n")
write("devices/sdb/stat", "10 0 100 0 20 0 100 0 0 0 0\n")
write("devinfo/1/missing", "0\n")
write("devinfo/1/error_stats", "write_errs 1\nread_errs 2\nflush_errs 0\ncorruption_errs 3\ngeneration_errs 0\n")
write("devinfo/2/missing", "1\n")
filesystems, err := Filesystems()
require.NoError(t, err)
require.Len(t, filesystems, 1)
assert.Equal(t, Filesystem{
UUID: "1b2c3d4e-0000-0000-0000-000000000000", Raw: true, Name: "tank", Size: 384000, Alloc: 7168, Health: "DEGRADED", NRead: 153600, NWrite: 256000,
Devices: []Device{
{Name: "devid 1", State: "ONLINE", ReadErrs: 2, WriteErrs: 1, CorruptionErrs: 3},
{Name: "devid 2", State: "MISSING"},
},
}, filesystems[0])
// Unlabeled filesystems fall back to the first mountpoint, then the UUID.
write("label", "\n")
require.NoError(t, os.WriteFile(mountsPath, []byte(
"/dev/sdz1 /other btrfs rw 0 0\n/dev/sdb /mnt/storage btrfs rw 0 0\n/dev/sdb /mnt/storage/sub btrfs rw,subvol=/sub 0 0\n",
), 0o644))
filesystems, err = Filesystems()
require.NoError(t, err)
assert.Equal(t, "/mnt/storage", filesystems[0].Name)
require.NoError(t, os.Remove(mountsPath))
filesystems, err = Filesystems()
require.NoError(t, err)
assert.Equal(t, "1b2c3d4e-0000-0000-0000-000000000000", filesystems[0].Name)
write("devinfo/3/replace_target", "1\n")
write("recorded-size/3", "512000\n")
filesystems, err = Filesystems()
require.NoError(t, err)
assert.Equal(t, uint64(384000), filesystems[0].Size, "replacement target must not inflate capacity")
deviceSize = func(string, uint64) (uint64, error) { return 0, os.ErrPermission }
filesystems, err = Filesystems()
require.NoError(t, err)
require.Len(t, filesystems, 1)
assert.Equal(t, uint64(1024000), filesystems[0].Size)
assert.Equal(t, "DEGRADED", filesystems[0].Health)
assert.Equal(t, uint64(153600), filesystems[0].NRead)
// A partial ioctl result must not be mixed with the backing-device total.
deviceSize = func(_ string, devid uint64) (uint64, error) {
if devid == 2 {
return 0, os.ErrPermission
}
return 256000, nil
}
filesystems, err = Filesystems()
require.NoError(t, err)
assert.Equal(t, uint64(1024000), filesystems[0].Size)
// With no mount visible (e.g. Docker), the real lookup falls back too.
deviceSize = ioctlDeviceSize
filesystems, err = Filesystems()
require.NoError(t, err)
require.Len(t, filesystems, 1)
assert.Equal(t, uint64(1024000), filesystems[0].Size)
filesystemUsage = func(string) (uint64, uint64, error) { return 100, 900, nil }
filesystems, err = Filesystems()
require.NoError(t, err)
assert.Equal(t, uint64(1000), filesystems[0].Size)
assert.Equal(t, uint64(100), filesystems[0].Alloc)
assert.False(t, filesystems[0].Raw)
}
func TestFilesystemsNoBtrfs(t *testing.T) {
oldPath := sysfsPath
sysfsPath = filepath.Join(t.TempDir(), "missing")
t.Cleanup(func() { sysfsPath = oldPath })
filesystems, err := Filesystems()
require.NoError(t, err)
assert.Nil(t, filesystems)
}
func TestIoctlDeviceSizeFailure(t *testing.T) {
_, err := ioctlDeviceSize("", 1)
require.Error(t, err)
_, err = ioctlDeviceSize(t.TempDir(), 1)
require.Error(t, err)
assert.ErrorIs(t, err, unix.ENOTTY)
}
func TestMountpointsDecodeEscapes(t *testing.T) {
oldMounts := mountsPath
mountsPath = filepath.Join(t.TempDir(), "mounts")
t.Cleanup(func() { mountsPath = oldMounts })
require.NoError(t, os.WriteFile(mountsPath, []byte("/dev/test-btrfs /mnt/my\\040data btrfs rw 0 0\n"), 0o644))
assert.Equal(t, "/mnt/my data", mountpointsByDevice()["test-btrfs"])
}
func TestFilesystemWithoutDevinfo(t *testing.T) {
root := t.TempDir()
require.NoError(t, os.MkdirAll(filepath.Join(root, "devices", "sda"), 0755))
require.NoError(t, os.WriteFile(filepath.Join(root, "devices", "sda", "size"), []byte("1000"), 0644))
fs, err := readFilesystem(root, nil)
require.NoError(t, err)
assert.Equal(t, uint64(512000), fs.Size)
assert.True(t, fs.Raw)
assert.Equal(t, "UNKNOWN", fs.Health)
assert.Empty(t, fs.Devices)
require.NoError(t, os.MkdirAll(filepath.Join(root, "devinfo", "1"), 0755))
fs, err = readFilesystem(root, nil)
require.NoError(t, err)
assert.Equal(t, "UNKNOWN", fs.Health)
require.Len(t, fs.Devices, 1)
assert.Equal(t, "UNKNOWN", fs.Devices[0].State)
// Some older interfaces lack the devices directory too.
fs, err = readFilesystem(t.TempDir(), nil)
require.NoError(t, err)
assert.Equal(t, "UNKNOWN", fs.Health)
}
func TestLocalBtrfsUsage(t *testing.T) {
path := os.Getenv("BESZEL_TEST_BTRFS_MOUNT")
if path == "" {
t.Skip("set BESZEL_TEST_BTRFS_MOUNT for read-only live validation")
}
used, available, err := statfsUsage(path)
require.NoError(t, err)
filesystems, err := Filesystems()
require.NoError(t, err)
for _, fs := range filesystems {
if !fs.Raw && fs.Alloc == used && fs.Size == used+available {
t.Logf("pool=%s used=%d available=%d effective_capacity=%d", fs.Name, used, available, fs.Size)
return
}
}
t.Fatal("collector did not report the mounted filesystem's usable capacity")
}
func TestMountID(t *testing.T) {
assert.Empty(t, MountID(""))
assert.Empty(t, MountID(filepath.Join(t.TempDir(), "missing")))
path := os.Getenv("BESZEL_TEST_BTRFS_MOUNT")
if path == "" {
t.Skip("set BESZEL_TEST_BTRFS_MOUNT for live identity validation")
}
id := MountID(path)
require.NotEmpty(t, id)
assert.Equal(t, id, MountID(filepath.Join(path, ".")))
}
func TestMountinfoUUIDLookup(t *testing.T) {
info := `1 0 0:40 /@ /inaccessible ro shared:1 - btrfs /dev/mapper/unavailable rw
2 0 0:40 /@/docker/hosts /etc/hosts ro - btrfs /dev/mapper/unavailable rw
3 0 0:40 /@/docker/hostname /etc/hostname ro - btrfs /dev/mapper/unavailable rw
4 0 0:41 /subvol /extra-filesystems/my\040disk ro master:2 - btrfs /dev/missing rw
5 0 0:42 / /ext4 ro - ext4 /dev/mapper/unavailable rw
malformed
6 0 0:43 / /bad ro - btrfs
`
var calls []string
mounts := mountpointsByUUID(info, func(path string) string {
calls = append(calls, path)
switch path {
case "/etc/hosts":
return "root-uuid"
case "/extra-filesystems/my disk":
return "extra-uuid"
}
return ""
})
assert.Equal(t, map[string]string{"uuid:root-uuid": "/etc/hosts", "uuid:extra-uuid": "/extra-filesystems/my disk"}, mounts)
assert.Equal(t, []string{"/inaccessible", "/etc/hosts", "/extra-filesystems/my disk"}, calls)
}
func TestDockerFilesystemWithoutDeviceNodes(t *testing.T) {
root := t.TempDir()
oldSysfs, oldMounts, oldInfo, oldUUID, oldUsage := sysfsPath, mountsPath, mountinfoPath, mountUUID, filesystemUsage
t.Cleanup(func() {
sysfsPath, mountsPath, mountinfoPath, mountUUID, filesystemUsage = oldSysfs, oldMounts, oldInfo, oldUUID, oldUsage
})
sysfsPath = filepath.Join(root, "sysfs")
mountsPath = filepath.Join(root, "missing-mounts")
mountinfoPath = filepath.Join(root, "mountinfo")
uuid := "11111111-1111-4111-8111-111111111111"
dir := filepath.Join(sysfsPath, uuid)
for path, content := range map[string]string{"devices/dm-0/size": "1000", "devinfo/1/missing": "0"} {
target := filepath.Join(dir, path)
require.NoError(t, os.MkdirAll(filepath.Dir(target), 0755))
require.NoError(t, os.WriteFile(target, []byte(content), 0644))
}
require.NoError(t, os.WriteFile(mountinfoPath, []byte("2 1 0:40 /@/docker/hosts /etc/hosts ro - btrfs /dev/mapper/not-in-container rw\n"), 0644))
mountUUID = func(path string) string {
if path == "/etc/hosts" {
return uuid
}
return ""
}
filesystemUsage = func(path string) (uint64, uint64, error) { require.Equal(t, "/etc/hosts", path); return 100, 900, nil }
fs, err := Filesystems()
require.NoError(t, err)
require.Len(t, fs, 1)
assert.Equal(t, uuid, fs[0].MountID)
assert.Equal(t, "dm-0", fs[0].IODevice)
assert.False(t, fs[0].Raw)
assert.Equal(t, uint64(1000), fs[0].Size)
}
func TestLivePoolMountIdentity(t *testing.T) {
path := os.Getenv("BESZEL_TEST_BTRFS_MOUNT")
if path == "" {
t.Skip("set BESZEL_TEST_BTRFS_MOUNT for live validation")
}
id := MountID(path)
require.NotEmpty(t, id)
pools, err := Filesystems()
require.NoError(t, err)
for _, pool := range pools {
if pool.UUID != id {
continue
}
assert.Equal(t, id, pool.MountID)
assert.False(t, pool.Raw)
t.Logf("uuid=%s mount_identity=%s io_device=%s raw=%v", pool.UUID, pool.MountID, pool.IODevice, pool.Raw)
return
}
t.Fatal("mounted Btrfs filesystem was not discovered")
}
-11
View File
@@ -1,11 +0,0 @@
//go:build !linux
package btrfs
import "errors"
func Filesystems() ([]Filesystem, error) {
return nil, errors.ErrUnsupported
}
func MountID(string) string { return "" }
+43 -96
View File
@@ -2,7 +2,6 @@ package agent
import ( import (
"crypto/tls" "crypto/tls"
"crypto/x509"
"errors" "errors"
"fmt" "fmt"
"log/slog" "log/slog"
@@ -15,38 +14,20 @@ import (
"time" "time"
"github.com/henrygd/beszel" "github.com/henrygd/beszel"
"github.com/henrygd/beszel/agent/utils"
"github.com/henrygd/beszel/internal/common" "github.com/henrygd/beszel/internal/common"
"github.com/henrygd/beszel/internal/entities/smart"
"github.com/henrygd/beszel/internal/entities/system"
"github.com/henrygd/beszel/internal/entities/systemd"
"github.com/fxamacker/cbor/v2" "github.com/fxamacker/cbor/v2"
"github.com/lxzan/gws" "github.com/lxzan/gws"
"golang.org/x/crypto/ssh" "golang.org/x/crypto/ssh"
"golang.org/x/net/proxy"
) )
const ( const (
// Keep the connection alive long enough for a slow collection cycle to wsDeadline = 70 * time.Second
// finish before the hub considers the agent disconnected.
wsDeadline = 120 * time.Second
) )
// errNoHubURL is returned when HUB_URL is unset. This is not a failure
// condition: an agent configured with only a public key runs in SSH-only mode,
// where the hub dials the agent and no outbound WebSocket client is expected.
var errNoHubURL = errors.New("HUB_URL environment variable not set")
type caCertFileError struct {
err error
}
func (e *caCertFileError) Error() string {
return e.err.Error()
}
func (e *caCertFileError) Unwrap() error {
return e.err
}
// WebSocketClient manages the WebSocket connection between the agent and hub. // WebSocketClient manages the WebSocket connection between the agent and hub.
// It handles authentication, message routing, and connection lifecycle management. // It handles authentication, message routing, and connection lifecycle management.
type WebSocketClient struct { type WebSocketClient struct {
@@ -60,32 +41,27 @@ type WebSocketClient struct {
hubRequest *common.HubRequest[cbor.RawMessage] // Reusable request structure for message parsing hubRequest *common.HubRequest[cbor.RawMessage] // Reusable request structure for message parsing
lastConnectAttempt time.Time // Timestamp of last connection attempt lastConnectAttempt time.Time // Timestamp of last connection attempt
hubVerified bool // Whether the hub has been cryptographically verified hubVerified bool // Whether the hub has been cryptographically verified
tlsConfig *tls.Config // Optional TLS configuration with custom CA certificates
} }
// newWebSocketClient creates a new WebSocket client for the given agent. // newWebSocketClient creates a new WebSocket client for the given agent.
// It reads configuration from environment variables and validates the hub URL. // It reads configuration from environment variables and validates the hub URL.
func newWebSocketClient(agent *Agent) (client *WebSocketClient, err error) { func newWebSocketClient(agent *Agent) (client *WebSocketClient, err error) {
hubURLStr, exists := utils.GetEnv("HUB_URL") hubURLStr, exists := GetEnv("HUB_URL")
if !exists { if !exists {
return nil, errNoHubURL return nil, errors.New("HUB_URL environment variable not set")
} }
client = &WebSocketClient{} client = &WebSocketClient{}
client.hubURL, err = url.Parse(hubURLStr) client.hubURL, err = url.Parse(hubURLStr)
if err != nil || client.hubURL.Host == "" { if err != nil {
return nil, fmt.Errorf("invalid HUB_URL %q: must include scheme and host (e.g. http://hub.example.com:8090)", hubURLStr) return nil, errors.New("invalid hub URL")
} }
// get registration token // get registration token
client.token, err = getToken() client.token, err = getToken()
if err != nil { if err != nil {
return nil, err return nil, err
} }
client.tlsConfig, err = getTLSConfig()
if err != nil {
return nil, err
}
client.agent = agent client.agent = agent
client.hubRequest = &common.HubRequest[cbor.RawMessage]{} client.hubRequest = &common.HubRequest[cbor.RawMessage]{}
@@ -99,12 +75,12 @@ func newWebSocketClient(agent *Agent) (client *WebSocketClient, err error) {
// If neither is set, it returns an error. // If neither is set, it returns an error.
func getToken() (string, error) { func getToken() (string, error) {
// get token from env var // get token from env var
token, _ := utils.GetEnv("TOKEN") token, _ := GetEnv("TOKEN")
if token != "" { if token != "" {
return token, nil return token, nil
} }
// get token from file // get token from file
tokenFile, _ := utils.GetEnv("TOKEN_FILE") tokenFile, _ := GetEnv("TOKEN_FILE")
if tokenFile == "" { if tokenFile == "" {
return "", errors.New("must set TOKEN or TOKEN_FILE") return "", errors.New("must set TOKEN or TOKEN_FILE")
} }
@@ -112,52 +88,7 @@ func getToken() (string, error) {
if err != nil { if err != nil {
return "", err return "", err
} }
return parseTokenFile(string(tokenBytes), tokenFile) return strings.TrimSpace(string(tokenBytes)), nil
}
// parseTokenFile reads a single token from TOKEN_FILE.
// Blank lines and comments are ignored. Multiple tokens are rejected because
// the agent supports only one outbound hub connection.
func parseTokenFile(contents, path string) (string, error) {
var token string
for line := range strings.Lines(contents) {
line = strings.TrimSpace(line)
if len(line) == 0 || strings.HasPrefix(line, "#") {
continue
}
if token != "" {
return "", fmt.Errorf("%s must contain a single token", path)
}
token = line
}
// An empty file keeps returning an empty token, as before: the caller decides
// what to do about it.
return token, nil
}
// getTLSConfig returns a TLS configuration containing the system certificate
// pool plus any certificates configured through CA_CERT_FILE. A nil config lets
// gws use Go's default TLS configuration and system roots.
func getTLSConfig() (*tls.Config, error) {
caCertFile, _ := utils.GetEnv("CA_CERT_FILE")
if caCertFile == "" {
return nil, nil
}
caCertPEM, err := os.ReadFile(caCertFile)
if err != nil {
return nil, &caCertFileError{fmt.Errorf("read CA_CERT_FILE %q: %w", caCertFile, err)}
}
rootCAs, err := x509.SystemCertPool()
if err != nil {
return nil, &caCertFileError{fmt.Errorf("load system CA certificate pool: %w", err)}
}
if !rootCAs.AppendCertsFromPEM(caCertPEM) {
return nil, &caCertFileError{fmt.Errorf("CA_CERT_FILE %q does not contain any valid PEM certificates", caCertFile)}
}
return &tls.Config{RootCAs: rootCAs}, nil
} }
// getOptions returns the WebSocket client options, creating them if necessary. // getOptions returns the WebSocket client options, creating them if necessary.
@@ -175,22 +106,14 @@ func (client *WebSocketClient) getOptions() *gws.ClientOption {
} }
client.hubURL.Path = path.Join(client.hubURL.Path, "api/beszel/agent-connect") client.hubURL.Path = path.Join(client.hubURL.Path, "api/beszel/agent-connect")
// make sure BESZEL_AGENT_ALL_PROXY works (GWS only checks ALL_PROXY)
if val := os.Getenv("BESZEL_AGENT_ALL_PROXY"); val != "" {
os.Setenv("ALL_PROXY", val)
}
client.options = &gws.ClientOption{ client.options = &gws.ClientOption{
Addr: client.hubURL.String(), Addr: client.hubURL.String(),
TlsConfig: client.tlsConfig, TlsConfig: &tls.Config{InsecureSkipVerify: true},
RequestHeader: http.Header{ RequestHeader: http.Header{
"User-Agent": []string{getUserAgent()}, "User-Agent": []string{getUserAgent()},
"X-Token": []string{client.token}, "X-Token": []string{client.token},
"X-Beszel": []string{beszel.Version}, "X-Beszel": []string{beszel.Version},
}, },
NewDialer: func() (gws.Dialer, error) {
return proxy.FromEnvironment(), nil
},
} }
return client.options return client.options
} }
@@ -277,7 +200,7 @@ func (client *WebSocketClient) handleAuthChallenge(msg *common.HubRequest[cbor.R
} }
if authRequest.NeedSysInfo { if authRequest.NeedSysInfo {
response.Name, _ = utils.GetEnv("SYSTEM_NAME") response.Name, _ = GetEnv("SYSTEM_NAME")
response.Hostname = client.agent.systemDetails.Hostname response.Hostname = client.agent.systemDetails.Hostname
serverAddr := client.agent.connectionManager.serverOptions.Addr serverAddr := client.agent.connectionManager.serverOptions.Addr
_, response.Port, _ = net.SplitHostPort(serverAddr) _, response.Port, _ = net.SplitHostPort(serverAddr)
@@ -336,16 +259,40 @@ func (client *WebSocketClient) sendMessage(data any) error {
return err return err
} }
// sendResponse sends a response with optional request ID. // sendResponse sends a response with optional request ID for the new protocol
// For ID-based requests, we must populate legacy typed fields for backward
// compatibility with older hubs (<= 0.17) that don't read the generic Data field.
func (client *WebSocketClient) sendResponse(data any, requestID *uint32) error { func (client *WebSocketClient) sendResponse(data any, requestID *uint32) error {
if requestID != nil { if requestID != nil {
response := newAgentResponse(data, requestID) // New format with ID - use typed fields
response := common.AgentResponse{
Id: requestID,
}
// Set the appropriate typed field based on data type
switch v := data.(type) {
case *system.CombinedData:
response.SystemData = v
case *common.FingerprintResponse:
response.Fingerprint = v
case string:
response.String = &v
case map[string]smart.SmartData:
response.SmartData = v
case systemd.ServiceDetails:
response.ServiceInfo = v
// case []byte:
// response.RawBytes = v
// case string:
// response.RawBytes = []byte(v)
default:
// For any other type, convert to error
response.Error = fmt.Sprintf("unsupported response type: %T", data)
}
return client.sendMessage(response) return client.sendMessage(response)
} else {
// Legacy format - send data directly
return client.sendMessage(data)
} }
// Legacy format - send data directly
return client.sendMessage(data)
} }
// getUserAgent returns one of two User-Agent strings based on current time. // getUserAgent returns one of two User-Agent strings based on current time.
+89 -260
View File
@@ -1,22 +1,12 @@
//go:build testing //go:build testing
// +build testing
package agent package agent
import ( import (
"crypto/ed25519" "crypto/ed25519"
"crypto/rand"
"crypto/rsa"
"crypto/tls"
"crypto/x509"
"crypto/x509/pkix"
"encoding/pem"
"math/big"
"net"
"net/http"
"net/http/httptest"
"net/url" "net/url"
"os" "os"
"path/filepath"
"strings" "strings"
"testing" "testing"
"time" "time"
@@ -26,34 +16,11 @@ import (
"github.com/henrygd/beszel/internal/common" "github.com/henrygd/beszel/internal/common"
"github.com/fxamacker/cbor/v2" "github.com/fxamacker/cbor/v2"
"github.com/lxzan/gws"
"github.com/stretchr/testify/assert" "github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require" "github.com/stretchr/testify/require"
"golang.org/x/crypto/ssh" "golang.org/x/crypto/ssh"
) )
// TestNewWebSocketClientNoHubURL verifies that an unset HUB_URL returns the
// errNoHubURL sentinel rather than an opaque error. Callers rely on this to
// distinguish SSH-only mode -- a supported configuration in which the hub dials
// the agent -- from an actual misconfiguration.
func TestNewWebSocketClientNoHubURL(t *testing.T) {
agent := createTestAgent(t)
// t.Setenv registers restoration of the original value; unset afterwards so
// GetEnv's LookupEnv reports the variable as absent rather than empty.
t.Setenv("BESZEL_AGENT_HUB_URL", "")
os.Unsetenv("BESZEL_AGENT_HUB_URL")
t.Setenv("HUB_URL", "")
os.Unsetenv("HUB_URL")
t.Setenv("BESZEL_AGENT_TOKEN", "test-token")
client, err := newWebSocketClient(agent)
require.Error(t, err)
assert.Nil(t, client)
assert.ErrorIs(t, err, errNoHubURL)
}
// TestNewWebSocketClient tests WebSocket client creation // TestNewWebSocketClient tests WebSocket client creation
func TestNewWebSocketClient(t *testing.T) { func TestNewWebSocketClient(t *testing.T) {
agent := createTestAgent(t) agent := createTestAgent(t)
@@ -85,18 +52,11 @@ func TestNewWebSocketClient(t *testing.T) {
errorMsg: "HUB_URL environment variable not set", errorMsg: "HUB_URL environment variable not set",
}, },
{ {
name: "malformed URL", name: "invalid URL",
hubURL: "ht\ttp://invalid", hubURL: "ht\ttp://invalid",
token: "test-token", token: "test-token",
expectError: true, expectError: true,
errorMsg: "invalid HUB_URL", errorMsg: "invalid hub URL",
},
{
name: "URL without host",
hubURL: "http:/api",
token: "test-token",
expectError: true,
errorMsg: "invalid HUB_URL",
}, },
{ {
name: "missing token", name: "missing token",
@@ -111,11 +71,19 @@ func TestNewWebSocketClient(t *testing.T) {
t.Run(tc.name, func(t *testing.T) { t.Run(tc.name, func(t *testing.T) {
// Set up environment // Set up environment
if tc.hubURL != "" { if tc.hubURL != "" {
t.Setenv("BESZEL_AGENT_HUB_URL", tc.hubURL) os.Setenv("BESZEL_AGENT_HUB_URL", tc.hubURL)
} else {
os.Unsetenv("BESZEL_AGENT_HUB_URL")
} }
if tc.token != "" { if tc.token != "" {
t.Setenv("BESZEL_AGENT_TOKEN", tc.token) os.Setenv("BESZEL_AGENT_TOKEN", tc.token)
} else {
os.Unsetenv("BESZEL_AGENT_TOKEN")
} }
defer func() {
os.Unsetenv("BESZEL_AGENT_HUB_URL")
os.Unsetenv("BESZEL_AGENT_TOKEN")
}()
client, err := newWebSocketClient(agent) client, err := newWebSocketClient(agent)
@@ -171,8 +139,12 @@ func TestWebSocketClient_GetOptions(t *testing.T) {
for _, tc := range testCases { for _, tc := range testCases {
t.Run(tc.name, func(t *testing.T) { t.Run(tc.name, func(t *testing.T) {
// Set up environment // Set up environment
t.Setenv("BESZEL_AGENT_HUB_URL", tc.inputURL) os.Setenv("BESZEL_AGENT_HUB_URL", tc.inputURL)
t.Setenv("BESZEL_AGENT_TOKEN", "test-token") os.Setenv("BESZEL_AGENT_TOKEN", "test-token")
defer func() {
os.Unsetenv("BESZEL_AGENT_HUB_URL")
os.Unsetenv("BESZEL_AGENT_TOKEN")
}()
client, err := newWebSocketClient(agent) client, err := newWebSocketClient(agent)
require.NoError(t, err) require.NoError(t, err)
@@ -198,155 +170,6 @@ func TestWebSocketClient_GetOptions(t *testing.T) {
} }
} }
func TestWebSocketClient_TLSVerification(t *testing.T) {
agent := createTestAgent(t)
serverCert, serverCertPEM := newSelfSignedServerCertificate(t)
upgrader := gws.NewUpgrader(&gws.BuiltinEventHandler{}, nil)
server := httptest.NewUnstartedServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
conn, err := upgrader.Upgrade(w, r)
if err == nil {
go conn.ReadLoop()
}
}))
server.TLS = &tls.Config{Certificates: []tls.Certificate{serverCert}}
server.StartTLS()
t.Cleanup(server.Close)
caCertFile := filepath.Join(t.TempDir(), "hub-ca.crt")
require.NoError(t, os.WriteFile(caCertFile, serverCertPEM, 0600))
newClient := func(t *testing.T, caCertFile string) *WebSocketClient {
t.Helper()
t.Setenv("BESZEL_AGENT_HUB_URL", server.URL)
t.Setenv("BESZEL_AGENT_TOKEN", "test-token")
t.Setenv("BESZEL_AGENT_CA_CERT_FILE", caCertFile)
client, err := newWebSocketClient(agent)
require.NoError(t, err)
return client
}
t.Run("system roots are used by default", func(t *testing.T) {
client := newClient(t, "")
assert.Nil(t, client.getOptions().TlsConfig)
_, _, err := gws.NewClient(&gws.BuiltinEventHandler{}, client.getOptions())
require.Error(t, err)
})
t.Run("custom CA trusts self-signed certificate", func(t *testing.T) {
systemRoots, err := x509.SystemCertPool()
require.NoError(t, err)
client := newClient(t, caCertFile)
assert.Greater(t, len(client.getOptions().TlsConfig.RootCAs.Subjects()), len(systemRoots.Subjects()))
conn, _, err := gws.NewClient(&gws.BuiltinEventHandler{}, client.getOptions())
require.NoError(t, err)
require.NoError(t, conn.NetConn().Close())
})
t.Run("custom CA does not bypass hostname verification", func(t *testing.T) {
client := newClient(t, caCertFile)
client.getOptions().TlsConfig.ServerName = "wrong.example.com"
_, _, err := gws.NewClient(&gws.BuiltinEventHandler{}, client.getOptions())
require.Error(t, err)
})
}
func TestWebSocketClient_NonTLSConnection(t *testing.T) {
agent := createTestAgent(t)
upgrader := gws.NewUpgrader(&gws.BuiltinEventHandler{}, nil)
server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
conn, err := upgrader.Upgrade(w, r)
if err == nil {
go conn.ReadLoop()
}
}))
t.Cleanup(server.Close)
t.Setenv("BESZEL_AGENT_HUB_URL", server.URL)
t.Setenv("BESZEL_AGENT_TOKEN", "test-token")
t.Setenv("BESZEL_AGENT_CA_CERT_FILE", "")
client, err := newWebSocketClient(agent)
require.NoError(t, err)
assert.Nil(t, client.getOptions().TlsConfig)
conn, _, err := gws.NewClient(&gws.BuiltinEventHandler{}, client.getOptions())
require.NoError(t, err)
require.NoError(t, conn.NetConn().Close())
}
func TestGetTLSConfigErrors(t *testing.T) {
tempDir := t.TempDir()
testCases := []struct {
name string
path string
contents []byte
errorMatch string
}{
{
name: "missing file",
path: filepath.Join(tempDir, "missing.pem"),
errorMatch: "read CA_CERT_FILE",
},
{
name: "unreadable path",
path: tempDir,
errorMatch: "read CA_CERT_FILE",
},
{
name: "empty file",
path: filepath.Join(tempDir, "empty.pem"),
contents: []byte{},
errorMatch: "does not contain any valid PEM certificates",
},
{
name: "malformed file",
path: filepath.Join(tempDir, "malformed.pem"),
contents: []byte("not a PEM certificate"),
errorMatch: "does not contain any valid PEM certificates",
},
}
for _, tc := range testCases {
t.Run(tc.name, func(t *testing.T) {
if tc.contents != nil {
require.NoError(t, os.WriteFile(tc.path, tc.contents, 0600))
}
t.Setenv("BESZEL_AGENT_CA_CERT_FILE", tc.path)
tlsConfig, err := getTLSConfig()
require.Error(t, err)
assert.Nil(t, tlsConfig)
assert.Contains(t, err.Error(), tc.errorMatch)
assert.Contains(t, err.Error(), tc.path)
})
}
}
func newSelfSignedServerCertificate(t *testing.T) (tls.Certificate, []byte) {
t.Helper()
privateKey, err := rsa.GenerateKey(rand.Reader, 2048)
require.NoError(t, err)
template := &x509.Certificate{
SerialNumber: big.NewInt(1),
Subject: pkix.Name{CommonName: "127.0.0.1"},
NotBefore: time.Now().Add(-time.Hour),
NotAfter: time.Now().Add(time.Hour),
IPAddresses: []net.IP{net.ParseIP("127.0.0.1")},
KeyUsage: x509.KeyUsageDigitalSignature | x509.KeyUsageKeyEncipherment | x509.KeyUsageCertSign,
ExtKeyUsage: []x509.ExtKeyUsage{x509.ExtKeyUsageServerAuth},
BasicConstraintsValid: true,
IsCA: true,
}
certDER, err := x509.CreateCertificate(rand.Reader, template, template, &privateKey.PublicKey, privateKey)
require.NoError(t, err)
certPEM := pem.EncodeToMemory(&pem.Block{Type: "CERTIFICATE", Bytes: certDER})
keyPEM := pem.EncodeToMemory(&pem.Block{Type: "RSA PRIVATE KEY", Bytes: x509.MarshalPKCS1PrivateKey(privateKey)})
certificate, err := tls.X509KeyPair(certPEM, keyPEM)
require.NoError(t, err)
return certificate, certPEM
}
// TestWebSocketClient_VerifySignature tests signature verification // TestWebSocketClient_VerifySignature tests signature verification
func TestWebSocketClient_VerifySignature(t *testing.T) { func TestWebSocketClient_VerifySignature(t *testing.T) {
agent := createTestAgent(t) agent := createTestAgent(t)
@@ -363,8 +186,12 @@ func TestWebSocketClient_VerifySignature(t *testing.T) {
require.NoError(t, err) require.NoError(t, err)
// Set up environment // Set up environment
t.Setenv("BESZEL_AGENT_HUB_URL", "http://localhost:8080") os.Setenv("BESZEL_AGENT_HUB_URL", "http://localhost:8080")
t.Setenv("BESZEL_AGENT_TOKEN", "test-token") os.Setenv("BESZEL_AGENT_TOKEN", "test-token")
defer func() {
os.Unsetenv("BESZEL_AGENT_HUB_URL")
os.Unsetenv("BESZEL_AGENT_TOKEN")
}()
client, err := newWebSocketClient(agent) client, err := newWebSocketClient(agent)
require.NoError(t, err) require.NoError(t, err)
@@ -432,8 +259,12 @@ func TestWebSocketClient_HandleHubRequest(t *testing.T) {
agent := createTestAgent(t) agent := createTestAgent(t)
// Set up environment // Set up environment
t.Setenv("BESZEL_AGENT_HUB_URL", "http://localhost:8080") os.Setenv("BESZEL_AGENT_HUB_URL", "http://localhost:8080")
t.Setenv("BESZEL_AGENT_TOKEN", "test-token") os.Setenv("BESZEL_AGENT_TOKEN", "test-token")
defer func() {
os.Unsetenv("BESZEL_AGENT_HUB_URL")
os.Unsetenv("BESZEL_AGENT_TOKEN")
}()
client, err := newWebSocketClient(agent) client, err := newWebSocketClient(agent)
require.NoError(t, err) require.NoError(t, err)
@@ -520,8 +351,13 @@ func TestGetUserAgent(t *testing.T) {
func TestWebSocketClient_Close(t *testing.T) { func TestWebSocketClient_Close(t *testing.T) {
agent := createTestAgent(t) agent := createTestAgent(t)
t.Setenv("BESZEL_AGENT_HUB_URL", "http://localhost:8080") // Set up environment
t.Setenv("BESZEL_AGENT_TOKEN", "test-token") os.Setenv("BESZEL_AGENT_HUB_URL", "http://localhost:8080")
os.Setenv("BESZEL_AGENT_TOKEN", "test-token")
defer func() {
os.Unsetenv("BESZEL_AGENT_HUB_URL")
os.Unsetenv("BESZEL_AGENT_TOKEN")
}()
client, err := newWebSocketClient(agent) client, err := newWebSocketClient(agent)
require.NoError(t, err) require.NoError(t, err)
@@ -536,8 +372,13 @@ func TestWebSocketClient_Close(t *testing.T) {
func TestWebSocketClient_ConnectRateLimit(t *testing.T) { func TestWebSocketClient_ConnectRateLimit(t *testing.T) {
agent := createTestAgent(t) agent := createTestAgent(t)
t.Setenv("BESZEL_AGENT_HUB_URL", "http://localhost:8080") // Set up environment
t.Setenv("BESZEL_AGENT_TOKEN", "test-token") os.Setenv("BESZEL_AGENT_HUB_URL", "http://localhost:8080")
os.Setenv("BESZEL_AGENT_TOKEN", "test-token")
defer func() {
os.Unsetenv("BESZEL_AGENT_HUB_URL")
os.Unsetenv("BESZEL_AGENT_TOKEN")
}()
client, err := newWebSocketClient(agent) client, err := newWebSocketClient(agent)
require.NoError(t, err) require.NoError(t, err)
@@ -553,10 +394,20 @@ func TestWebSocketClient_ConnectRateLimit(t *testing.T) {
// TestGetToken tests the getToken function with various scenarios // TestGetToken tests the getToken function with various scenarios
func TestGetToken(t *testing.T) { func TestGetToken(t *testing.T) {
unsetEnvVars := func() {
os.Unsetenv("BESZEL_AGENT_TOKEN")
os.Unsetenv("TOKEN")
os.Unsetenv("BESZEL_AGENT_TOKEN_FILE")
os.Unsetenv("TOKEN_FILE")
}
t.Run("token from TOKEN environment variable", func(t *testing.T) { t.Run("token from TOKEN environment variable", func(t *testing.T) {
unsetEnvVars()
// Set TOKEN env var // Set TOKEN env var
expectedToken := "test-token-from-env" expectedToken := "test-token-from-env"
t.Setenv("TOKEN", expectedToken) os.Setenv("TOKEN", expectedToken)
defer os.Unsetenv("TOKEN")
token, err := getToken() token, err := getToken()
assert.NoError(t, err) assert.NoError(t, err)
@@ -564,9 +415,12 @@ func TestGetToken(t *testing.T) {
}) })
t.Run("token from BESZEL_AGENT_TOKEN environment variable", func(t *testing.T) { t.Run("token from BESZEL_AGENT_TOKEN environment variable", func(t *testing.T) {
unsetEnvVars()
// Set BESZEL_AGENT_TOKEN env var (should take precedence) // Set BESZEL_AGENT_TOKEN env var (should take precedence)
expectedToken := "test-token-from-beszel-env" expectedToken := "test-token-from-beszel-env"
t.Setenv("BESZEL_AGENT_TOKEN", expectedToken) os.Setenv("BESZEL_AGENT_TOKEN", expectedToken)
defer os.Unsetenv("BESZEL_AGENT_TOKEN")
token, err := getToken() token, err := getToken()
assert.NoError(t, err) assert.NoError(t, err)
@@ -574,6 +428,8 @@ func TestGetToken(t *testing.T) {
}) })
t.Run("token from TOKEN_FILE", func(t *testing.T) { t.Run("token from TOKEN_FILE", func(t *testing.T) {
unsetEnvVars()
// Create a temporary token file // Create a temporary token file
expectedToken := "test-token-from-file" expectedToken := "test-token-from-file"
tokenFile, err := os.CreateTemp("", "token-test-*.txt") tokenFile, err := os.CreateTemp("", "token-test-*.txt")
@@ -585,49 +441,17 @@ func TestGetToken(t *testing.T) {
tokenFile.Close() tokenFile.Close()
// Set TOKEN_FILE env var // Set TOKEN_FILE env var
t.Setenv("TOKEN_FILE", tokenFile.Name()) os.Setenv("TOKEN_FILE", tokenFile.Name())
defer os.Unsetenv("TOKEN_FILE")
token, err := getToken() token, err := getToken()
assert.NoError(t, err) assert.NoError(t, err)
assert.Equal(t, expectedToken, token) assert.Equal(t, expectedToken, token)
}) })
t.Run("TOKEN_FILE with surrounding blank lines and comments", func(t *testing.T) {
expectedToken := "test-token-with-noise"
tokenFile := filepath.Join(t.TempDir(), "token")
require.NoError(t, os.WriteFile(tokenFile, []byte("# hub token\n\n"+expectedToken+"\n\n"), 0o600))
t.Setenv("TOKEN_FILE", tokenFile)
token, err := getToken()
assert.NoError(t, err)
assert.Equal(t, expectedToken, token)
})
t.Run("TOKEN_FILE with multiple tokens is rejected", func(t *testing.T) {
tokenFile := filepath.Join(t.TempDir(), "token")
require.NoError(t, os.WriteFile(tokenFile, []byte("11111111-1111-1111-1111-111111111111\n22222222-2222-2222-2222-222222222222\n"), 0o600))
t.Setenv("TOKEN_FILE", tokenFile)
token, err := getToken()
require.Error(t, err)
assert.Empty(t, token)
assert.Contains(t, err.Error(), "must contain a single token")
})
t.Run("TOKEN_FILE holding only comments behaves like an empty file", func(t *testing.T) {
tokenFile := filepath.Join(t.TempDir(), "token")
require.NoError(t, os.WriteFile(tokenFile, []byte("\n# only a comment\n"), 0o600))
t.Setenv("TOKEN_FILE", tokenFile)
token, err := getToken()
assert.NoError(t, err)
assert.Equal(t, "", token)
})
t.Run("token from BESZEL_AGENT_TOKEN_FILE", func(t *testing.T) { t.Run("token from BESZEL_AGENT_TOKEN_FILE", func(t *testing.T) {
unsetEnvVars()
// Create a temporary token file // Create a temporary token file
expectedToken := "test-token-from-beszel-file" expectedToken := "test-token-from-beszel-file"
tokenFile, err := os.CreateTemp("", "token-test-*.txt") tokenFile, err := os.CreateTemp("", "token-test-*.txt")
@@ -639,7 +463,8 @@ func TestGetToken(t *testing.T) {
tokenFile.Close() tokenFile.Close()
// Set BESZEL_AGENT_TOKEN_FILE env var (should take precedence) // Set BESZEL_AGENT_TOKEN_FILE env var (should take precedence)
t.Setenv("BESZEL_AGENT_TOKEN_FILE", tokenFile.Name()) os.Setenv("BESZEL_AGENT_TOKEN_FILE", tokenFile.Name())
defer os.Unsetenv("BESZEL_AGENT_TOKEN_FILE")
token, err := getToken() token, err := getToken()
assert.NoError(t, err) assert.NoError(t, err)
@@ -647,6 +472,8 @@ func TestGetToken(t *testing.T) {
}) })
t.Run("TOKEN takes precedence over TOKEN_FILE", func(t *testing.T) { t.Run("TOKEN takes precedence over TOKEN_FILE", func(t *testing.T) {
unsetEnvVars()
// Create a temporary token file // Create a temporary token file
fileToken := "token-from-file" fileToken := "token-from-file"
tokenFile, err := os.CreateTemp("", "token-test-*.txt") tokenFile, err := os.CreateTemp("", "token-test-*.txt")
@@ -659,8 +486,12 @@ func TestGetToken(t *testing.T) {
// Set both TOKEN and TOKEN_FILE // Set both TOKEN and TOKEN_FILE
envToken := "token-from-env" envToken := "token-from-env"
t.Setenv("TOKEN", envToken) os.Setenv("TOKEN", envToken)
t.Setenv("TOKEN_FILE", tokenFile.Name()) os.Setenv("TOKEN_FILE", tokenFile.Name())
defer func() {
os.Unsetenv("TOKEN")
os.Unsetenv("TOKEN_FILE")
}()
token, err := getToken() token, err := getToken()
assert.NoError(t, err) assert.NoError(t, err)
@@ -668,10 +499,7 @@ func TestGetToken(t *testing.T) {
}) })
t.Run("error when neither TOKEN nor TOKEN_FILE is set", func(t *testing.T) { t.Run("error when neither TOKEN nor TOKEN_FILE is set", func(t *testing.T) {
t.Setenv("BESZEL_AGENT_TOKEN", "") unsetEnvVars()
t.Setenv("TOKEN", "")
t.Setenv("BESZEL_AGENT_TOKEN_FILE", "")
t.Setenv("TOKEN_FILE", "")
token, err := getToken() token, err := getToken()
assert.Error(t, err) assert.Error(t, err)
@@ -680,8 +508,11 @@ func TestGetToken(t *testing.T) {
}) })
t.Run("error when TOKEN_FILE points to non-existent file", func(t *testing.T) { t.Run("error when TOKEN_FILE points to non-existent file", func(t *testing.T) {
unsetEnvVars()
// Set TOKEN_FILE to a non-existent file // Set TOKEN_FILE to a non-existent file
t.Setenv("TOKEN_FILE", "/non/existent/file.txt") os.Setenv("TOKEN_FILE", "/non/existent/file.txt")
defer os.Unsetenv("TOKEN_FILE")
token, err := getToken() token, err := getToken()
assert.Error(t, err) assert.Error(t, err)
@@ -690,6 +521,8 @@ func TestGetToken(t *testing.T) {
}) })
t.Run("handles empty token file", func(t *testing.T) { t.Run("handles empty token file", func(t *testing.T) {
unsetEnvVars()
// Create an empty token file // Create an empty token file
tokenFile, err := os.CreateTemp("", "token-test-*.txt") tokenFile, err := os.CreateTemp("", "token-test-*.txt")
require.NoError(t, err) require.NoError(t, err)
@@ -697,7 +530,8 @@ func TestGetToken(t *testing.T) {
tokenFile.Close() tokenFile.Close()
// Set TOKEN_FILE env var // Set TOKEN_FILE env var
t.Setenv("TOKEN_FILE", tokenFile.Name()) os.Setenv("TOKEN_FILE", tokenFile.Name())
defer os.Unsetenv("TOKEN_FILE")
token, err := getToken() token, err := getToken()
assert.NoError(t, err) assert.NoError(t, err)
@@ -705,6 +539,8 @@ func TestGetToken(t *testing.T) {
}) })
t.Run("strips whitespace from TOKEN_FILE", func(t *testing.T) { t.Run("strips whitespace from TOKEN_FILE", func(t *testing.T) {
unsetEnvVars()
tokenWithWhitespace := " test-token-with-whitespace \n\t" tokenWithWhitespace := " test-token-with-whitespace \n\t"
expectedToken := "test-token-with-whitespace" expectedToken := "test-token-with-whitespace"
tokenFile, err := os.CreateTemp("", "token-test-*.txt") tokenFile, err := os.CreateTemp("", "token-test-*.txt")
@@ -715,18 +551,11 @@ func TestGetToken(t *testing.T) {
require.NoError(t, err) require.NoError(t, err)
tokenFile.Close() tokenFile.Close()
t.Setenv("TOKEN_FILE", tokenFile.Name()) os.Setenv("TOKEN_FILE", tokenFile.Name())
defer os.Unsetenv("TOKEN_FILE")
token, err := getToken() token, err := getToken()
assert.NoError(t, err) assert.NoError(t, err)
assert.Equal(t, expectedToken, token, "Whitespace should be stripped from token file content") assert.Equal(t, expectedToken, token, "Whitespace should be stripped from token file content")
}) })
} }
func TestWebSocketDeadlineCoversSlowCollection(t *testing.T) {
const minimumDeadline = 120 * time.Second
if wsDeadline < minimumDeadline {
t.Fatalf("WebSocket deadline %s is shorter than the slow-collection window of %s", wsDeadline, minimumDeadline)
}
}
+12 -72
View File
@@ -1,18 +1,14 @@
package agent package agent
import ( import (
"context"
"errors" "errors"
"log/slog" "log/slog"
"net"
"os" "os"
"os/signal" "os/signal"
"strings"
"syscall" "syscall"
"time" "time"
"github.com/henrygd/beszel/agent/health" "github.com/henrygd/beszel/agent/health"
"github.com/henrygd/beszel/agent/utils"
"github.com/henrygd/beszel/internal/entities/system" "github.com/henrygd/beszel/internal/entities/system"
) )
@@ -87,19 +83,7 @@ func (c *ConnectionManager) Start(serverOptions ServerOptions) error {
wsClient, err := newWebSocketClient(c.agent) wsClient, err := newWebSocketClient(c.agent)
if err != nil { if err != nil {
var caCertErr *caCertFileError slog.Warn("Error creating WebSocket client", "err", err)
if errors.As(err, &caCertErr) {
return err
}
disableSSH, _ := utils.GetEnv("DISABLE_SSH")
if errors.Is(err, errNoHubURL) && disableSSH != "true" {
// SSH-only mode: the hub dials the agent, so there is nothing to warn
// about. With SSH also disabled there is no connection method at all,
// so that case still warns.
slog.Debug("WebSocket client not configured", "err", err)
} else {
slog.Warn("Error creating WebSocket client", "err", err)
}
} }
c.wsClient = wsClient c.wsClient = wsClient
@@ -107,8 +91,8 @@ func (c *ConnectionManager) Start(serverOptions ServerOptions) error {
c.eventChan = make(chan ConnectionEvent, 1) c.eventChan = make(chan ConnectionEvent, 1)
// signal handling for shutdown // signal handling for shutdown
sigCtx, stopSignals := signal.NotifyContext(context.Background(), syscall.SIGINT, syscall.SIGTERM) sigChan := make(chan os.Signal, 1)
defer stopSignals() signal.Notify(sigChan, syscall.SIGINT, syscall.SIGTERM)
c.startWsTicker() c.startWsTicker()
c.connect() c.connect()
@@ -125,48 +109,22 @@ func (c *ConnectionManager) Start(serverOptions ServerOptions) error {
_ = c.startWebSocketConnection() _ = c.startWebSocketConnection()
case <-healthTicker: case <-healthTicker:
_ = health.Update() _ = health.Update()
case <-sigCtx.Done(): case <-sigChan:
slog.Info("Shutting down", "cause", context.Cause(sigCtx)) slog.Info("Shutting down")
return c.stop() _ = c.agent.StopServer()
c.closeWebSocket()
return health.CleanUp()
} }
} }
} }
// stop does not stop the connection manager itself, just any active connections. The manager will attempt to reconnect after stopping, so this should only be called immediately before shutting down the entire agent.
//
// If we need or want to expose a graceful Stop method in the future, do something like this to actually stop the manager:
//
// func (c *ConnectionManager) Start(serverOptions ServerOptions) error {
// ctx, cancel := context.WithCancel(context.Background())
// c.cancel = cancel
//
// for {
// select {
// case <-ctx.Done():
// return c.stop()
// }
// }
// }
//
// func (c *ConnectionManager) Stop() {
// c.cancel()
// }
func (c *ConnectionManager) stop() error {
_ = c.agent.StopServer()
c.agent.monitorManager.Stop()
c.closeWebSocket()
return health.CleanUp()
}
// handleEvent processes connection events and updates the connection state accordingly. // handleEvent processes connection events and updates the connection state accordingly.
func (c *ConnectionManager) handleEvent(event ConnectionEvent) { func (c *ConnectionManager) handleEvent(event ConnectionEvent) {
switch event { switch event {
case WebSocketConnect: case WebSocketConnect:
c.handleStateChange(WebSocketConnected) c.handleStateChange(WebSocketConnected)
case SSHConnect: case SSHConnect:
if c.State == Disconnected { c.handleStateChange(SSHConnected)
c.handleStateChange(SSHConnected)
}
case WebSocketDisconnect: case WebSocketDisconnect:
if c.State == WebSocketConnected { if c.State == WebSocketConnected {
c.handleStateChange(Disconnected) c.handleStateChange(Disconnected)
@@ -227,16 +185,9 @@ func (c *ConnectionManager) connect() {
// Try WebSocket first, if it fails, start SSH server // Try WebSocket first, if it fails, start SSH server
err := c.startWebSocketConnection() err := c.startWebSocketConnection()
if err != nil { if err != nil && c.State == Disconnected {
if shouldExitOnErr(err) { c.startSSHServer()
time.Sleep(2 * time.Second) // prevent tight restart loop c.startWsTicker()
_ = c.stop()
os.Exit(1)
}
if c.State == Disconnected {
c.startSSHServer()
c.startWsTicker()
}
} }
} }
@@ -273,14 +224,3 @@ func (c *ConnectionManager) closeWebSocket() {
c.wsClient.Close() c.wsClient.Close()
} }
} }
// shouldExitOnErr checks if the error is a DNS resolution failure and if the
// EXIT_ON_DNS_ERROR env var is set. https://github.com/henrygd/beszel/issues/1924.
func shouldExitOnErr(err error) bool {
if val, _ := utils.GetEnv("EXIT_ON_DNS_ERROR"); val == "true" {
if opErr, ok := errors.AsType[*net.OpError](err); ok {
return strings.Contains(opErr.Err.Error(), "lookup")
}
}
return false
}
+28 -95
View File
@@ -1,13 +1,14 @@
//go:build testing //go:build testing
// +build testing
package agent package agent
import ( import (
"crypto/ed25519" "crypto/ed25519"
"errors"
"fmt" "fmt"
"net" "net"
"net/url" "net/url"
"os"
"testing" "testing"
"time" "time"
@@ -114,12 +115,6 @@ func TestConnectionManager_EventHandling(t *testing.T) {
event: SSHConnect, event: SSHConnect,
expectedState: SSHConnected, expectedState: SSHConnected,
}, },
{
name: "SSH connect from WebSocket connected (no change)",
initialState: WebSocketConnected,
event: SSHConnect,
expectedState: WebSocketConnected,
},
{ {
name: "WebSocket disconnect from connected", name: "WebSocket disconnect from connected",
initialState: WebSocketConnected, initialState: WebSocketConnected,
@@ -189,6 +184,10 @@ func TestConnectionManager_TickerManagement(t *testing.T) {
// TestConnectionManager_WebSocketConnectionFlow tests WebSocket connection logic // TestConnectionManager_WebSocketConnectionFlow tests WebSocket connection logic
func TestConnectionManager_WebSocketConnectionFlow(t *testing.T) { func TestConnectionManager_WebSocketConnectionFlow(t *testing.T) {
if testing.Short() {
t.Skip("Skipping WebSocket connection test in short mode")
}
agent := createTestAgent(t) agent := createTestAgent(t)
cm := agent.connectionManager cm := agent.connectionManager
@@ -198,18 +197,19 @@ func TestConnectionManager_WebSocketConnectionFlow(t *testing.T) {
assert.Equal(t, Disconnected, cm.State, "State should remain Disconnected after failed connection") assert.Equal(t, Disconnected, cm.State, "State should remain Disconnected after failed connection")
// Test with invalid URL // Test with invalid URL
t.Setenv("BESZEL_AGENT_HUB_URL", "1,33%") os.Setenv("BESZEL_AGENT_HUB_URL", "invalid-url")
t.Setenv("BESZEL_AGENT_TOKEN", "test-token") os.Setenv("BESZEL_AGENT_TOKEN", "test-token")
defer func() {
_, err2 := newWebSocketClient(agent) os.Unsetenv("BESZEL_AGENT_HUB_URL")
assert.Error(t, err2, "WebSocket client creation should fail with invalid URL") os.Unsetenv("BESZEL_AGENT_TOKEN")
}()
// Test with missing token // Test with missing token
t.Setenv("BESZEL_AGENT_HUB_URL", "http://localhost:8080") os.Setenv("BESZEL_AGENT_HUB_URL", "http://localhost:8080")
t.Setenv("BESZEL_AGENT_TOKEN", "") os.Unsetenv("BESZEL_AGENT_TOKEN")
_, err3 := newWebSocketClient(agent) _, err2 := newWebSocketClient(agent)
assert.Error(t, err3, "WebSocket client creation should fail without token") assert.Error(t, err2, "WebSocket client creation should fail without token")
} }
// TestConnectionManager_ReconnectionLogic tests reconnection prevention logic // TestConnectionManager_ReconnectionLogic tests reconnection prevention logic
@@ -235,8 +235,12 @@ func TestConnectionManager_ConnectWithRateLimit(t *testing.T) {
cm := agent.connectionManager cm := agent.connectionManager
// Set up environment for WebSocket client creation // Set up environment for WebSocket client creation
t.Setenv("BESZEL_AGENT_HUB_URL", "ws://localhost:8080") os.Setenv("BESZEL_AGENT_HUB_URL", "ws://localhost:8080")
t.Setenv("BESZEL_AGENT_TOKEN", "test-token") os.Setenv("BESZEL_AGENT_TOKEN", "test-token")
defer func() {
os.Unsetenv("BESZEL_AGENT_HUB_URL")
os.Unsetenv("BESZEL_AGENT_TOKEN")
}()
// Create WebSocket client // Create WebSocket client
wsClient, err := newWebSocketClient(agent) wsClient, err := newWebSocketClient(agent)
@@ -271,19 +275,6 @@ func TestConnectionManager_StartWithInvalidConfig(t *testing.T) {
assert.Error(t, err, "Should error when starting already started connection manager") assert.Error(t, err, "Should error when starting already started connection manager")
} }
func TestConnectionManager_StartRejectsInvalidCACertFile(t *testing.T) {
agent := createTestAgent(t)
cm := agent.connectionManager
t.Setenv("BESZEL_AGENT_HUB_URL", "https://hub.example.com")
t.Setenv("BESZEL_AGENT_TOKEN", "test-token")
t.Setenv("BESZEL_AGENT_CA_CERT_FILE", t.TempDir())
err := cm.Start(ServerOptions{})
require.Error(t, err)
assert.Contains(t, err.Error(), "read CA_CERT_FILE")
assert.Nil(t, cm.eventChan)
}
// TestConnectionManager_CloseWebSocket tests WebSocket closing // TestConnectionManager_CloseWebSocket tests WebSocket closing
func TestConnectionManager_CloseWebSocket(t *testing.T) { func TestConnectionManager_CloseWebSocket(t *testing.T) {
agent := createTestAgent(t) agent := createTestAgent(t)
@@ -295,8 +286,12 @@ func TestConnectionManager_CloseWebSocket(t *testing.T) {
}, "Should not panic when closing nil WebSocket client") }, "Should not panic when closing nil WebSocket client")
// Set up environment and create WebSocket client // Set up environment and create WebSocket client
t.Setenv("BESZEL_AGENT_HUB_URL", "ws://localhost:8080") os.Setenv("BESZEL_AGENT_HUB_URL", "ws://localhost:8080")
t.Setenv("BESZEL_AGENT_TOKEN", "test-token") os.Setenv("BESZEL_AGENT_TOKEN", "test-token")
defer func() {
os.Unsetenv("BESZEL_AGENT_HUB_URL")
os.Unsetenv("BESZEL_AGENT_TOKEN")
}()
wsClient, err := newWebSocketClient(agent) wsClient, err := newWebSocketClient(agent)
require.NoError(t, err) require.NoError(t, err)
@@ -318,65 +313,3 @@ func TestConnectionManager_ConnectFlow(t *testing.T) {
cm.connect() cm.connect()
}, "Connect should not panic without WebSocket client") }, "Connect should not panic without WebSocket client")
} }
func TestShouldExitOnErr(t *testing.T) {
createDialErr := func(msg string) error {
return &net.OpError{
Op: "dial",
Net: "tcp",
Err: errors.New(msg),
}
}
tests := []struct {
name string
err error
envValue string
expected bool
}{
{
name: "no env var",
err: createDialErr("lookup lkahsdfasdf: no such host"),
envValue: "",
expected: false,
},
{
name: "env var false",
err: createDialErr("lookup lkahsdfasdf: no such host"),
envValue: "false",
expected: false,
},
{
name: "env var true, matching error",
err: createDialErr("lookup lkahsdfasdf: no such host"),
envValue: "true",
expected: true,
},
{
name: "env var true, matching error with extra context",
err: createDialErr("lookup beszel.server.lan on [::1]:53: read udp [::1]:44557->[::1]:53: read: connection refused"),
envValue: "true",
expected: true,
},
{
name: "env var true, non-matching error",
err: errors.New("connection refused"),
envValue: "true",
expected: false,
},
{
name: "env var true, dial but not lookup",
err: createDialErr("connection timeout"),
envValue: "true",
expected: false,
},
}
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
t.Setenv("EXIT_ON_DNS_ERROR", tt.envValue)
result := shouldExitOnErr(tt.err)
assert.Equal(t, tt.expected, result)
})
}
}
+6 -3
View File
@@ -14,10 +14,10 @@ var lastPerCoreCpuTimes = make(map[uint16][]cpu.TimesStat)
// init initializes the CPU monitoring by storing the initial CPU times // init initializes the CPU monitoring by storing the initial CPU times
// for the default 60-second cache interval. // for the default 60-second cache interval.
func init() { func init() {
if times, err := cpu.Times(false); err == nil && len(times) > 0 { if times, err := cpu.Times(false); err == nil {
lastCpuTimes[60000] = times[0] lastCpuTimes[60000] = times[0]
} }
if perCoreTimes, err := cpu.Times(true); err == nil && len(perCoreTimes) > 0 { if perCoreTimes, err := cpu.Times(true); err == nil {
lastPerCoreCpuTimes[60000] = perCoreTimes lastPerCoreCpuTimes[60000] = perCoreTimes
} }
} }
@@ -89,7 +89,10 @@ func getPerCoreCpuUsage(cacheTimeMs uint16) (system.Uint8Slice, error) {
lastTimes := lastPerCoreCpuTimes[cacheTimeMs] lastTimes := lastPerCoreCpuTimes[cacheTimeMs]
// Limit to the number of cores available in both samples // Limit to the number of cores available in both samples
length := min(len(lastTimes), len(perCoreTimes)) length := len(perCoreTimes)
if len(lastTimes) < length {
length = len(lastTimes)
}
usage := make([]uint8, length) usage := make([]uint8, length)
for i := 0; i < length; i++ { for i := 0; i < length; i++ {
+3 -5
View File
@@ -6,19 +6,17 @@ import (
"os" "os"
"path/filepath" "path/filepath"
"runtime" "runtime"
"github.com/henrygd/beszel/agent/utils"
) )
// GetDataDir returns the path to the data directory for the agent and an error // getDataDir returns the path to the data directory for the agent and an error
// if the directory is not valid. Attempts to find the optimal data directory if // if the directory is not valid. Attempts to find the optimal data directory if
// no data directories are provided. // no data directories are provided.
func GetDataDir(dataDirs ...string) (string, error) { func getDataDir(dataDirs ...string) (string, error) {
if len(dataDirs) > 0 { if len(dataDirs) > 0 {
return testDataDirs(dataDirs) return testDataDirs(dataDirs)
} }
dataDir, _ := utils.GetEnv("DATA_DIR") dataDir, _ := GetEnv("DATA_DIR")
if dataDir != "" { if dataDir != "" {
dataDirs = append(dataDirs, dataDir) dataDirs = append(dataDirs, dataDir)
} }
+32 -18
View File
@@ -1,4 +1,5 @@
//go:build testing //go:build testing
// +build testing
package agent package agent
@@ -12,19 +13,11 @@ import (
"github.com/stretchr/testify/require" "github.com/stretchr/testify/require"
) )
func invalidDataDir(t *testing.T) string {
t.Helper()
filePath := filepath.Join(t.TempDir(), "file")
require.NoError(t, os.WriteFile(filePath, nil, 0644))
return filepath.Join(filePath, "data")
}
func TestGetDataDir(t *testing.T) { func TestGetDataDir(t *testing.T) {
// Test with explicit dataDir parameter // Test with explicit dataDir parameter
t.Run("explicit data dir", func(t *testing.T) { t.Run("explicit data dir", func(t *testing.T) {
tempDir := t.TempDir() tempDir := t.TempDir()
result, err := GetDataDir(tempDir) result, err := getDataDir(tempDir)
require.NoError(t, err) require.NoError(t, err)
assert.Equal(t, tempDir, result) assert.Equal(t, tempDir, result)
}) })
@@ -33,7 +26,7 @@ func TestGetDataDir(t *testing.T) {
t.Run("explicit data dir - create new", func(t *testing.T) { t.Run("explicit data dir - create new", func(t *testing.T) {
tempDir := t.TempDir() tempDir := t.TempDir()
newDir := filepath.Join(tempDir, "new-data-dir") newDir := filepath.Join(tempDir, "new-data-dir")
result, err := GetDataDir(newDir) result, err := getDataDir(newDir)
require.NoError(t, err) require.NoError(t, err)
assert.Equal(t, newDir, result) assert.Equal(t, newDir, result)
@@ -47,25 +40,46 @@ func TestGetDataDir(t *testing.T) {
t.Run("DATA_DIR environment variable", func(t *testing.T) { t.Run("DATA_DIR environment variable", func(t *testing.T) {
tempDir := t.TempDir() tempDir := t.TempDir()
t.Setenv("BESZEL_AGENT_DATA_DIR", tempDir) // Set environment variable
oldValue := os.Getenv("DATA_DIR")
defer func() {
if oldValue == "" {
os.Unsetenv("BESZEL_AGENT_DATA_DIR")
} else {
os.Setenv("BESZEL_AGENT_DATA_DIR", oldValue)
}
}()
result, err := GetDataDir() os.Setenv("BESZEL_AGENT_DATA_DIR", tempDir)
result, err := getDataDir()
require.NoError(t, err) require.NoError(t, err)
assert.Equal(t, tempDir, result) assert.Equal(t, tempDir, result)
}) })
// Test with invalid explicit dataDir // Test with invalid explicit dataDir
t.Run("invalid explicit data dir", func(t *testing.T) { t.Run("invalid explicit data dir", func(t *testing.T) {
invalidPath := invalidDataDir(t) invalidPath := "/invalid/path/that/cannot/be/created"
_, err := GetDataDir(invalidPath) _, err := getDataDir(invalidPath)
assert.Error(t, err) assert.Error(t, err)
}) })
// Test fallback behavior (empty dataDir, no env var) // Test fallback behavior (empty dataDir, no env var)
t.Run("fallback to default directories", func(t *testing.T) { t.Run("fallback to default directories", func(t *testing.T) {
// Clear DATA_DIR environment variable
oldValue := os.Getenv("DATA_DIR")
defer func() {
if oldValue == "" {
os.Unsetenv("DATA_DIR")
} else {
os.Setenv("DATA_DIR", oldValue)
}
}()
os.Unsetenv("DATA_DIR")
// This will try platform-specific defaults, which may or may not work // This will try platform-specific defaults, which may or may not work
// We're mainly testing that it doesn't panic and returns some result // We're mainly testing that it doesn't panic and returns some result
result, err := GetDataDir() result, err := getDataDir()
// We don't assert success/failure here since it depends on system permissions // We don't assert success/failure here since it depends on system permissions
// Just verify we get a string result if no error // Just verify we get a string result if no error
if err == nil { if err == nil {
@@ -86,7 +100,7 @@ func TestTestDataDirs(t *testing.T) {
// Test with multiple directories, first one valid // Test with multiple directories, first one valid
t.Run("multiple dirs - first valid", func(t *testing.T) { t.Run("multiple dirs - first valid", func(t *testing.T) {
tempDir := t.TempDir() tempDir := t.TempDir()
invalidDir := invalidDataDir(t) invalidDir := "/invalid/path"
result, err := testDataDirs([]string{tempDir, invalidDir}) result, err := testDataDirs([]string{tempDir, invalidDir})
require.NoError(t, err) require.NoError(t, err)
assert.Equal(t, tempDir, result) assert.Equal(t, tempDir, result)
@@ -95,7 +109,7 @@ func TestTestDataDirs(t *testing.T) {
// Test with multiple directories, second one valid // Test with multiple directories, second one valid
t.Run("multiple dirs - second valid", func(t *testing.T) { t.Run("multiple dirs - second valid", func(t *testing.T) {
tempDir := t.TempDir() tempDir := t.TempDir()
invalidDir := invalidDataDir(t) invalidDir := "/invalid/path"
result, err := testDataDirs([]string{invalidDir, tempDir}) result, err := testDataDirs([]string{invalidDir, tempDir})
require.NoError(t, err) require.NoError(t, err)
assert.Equal(t, tempDir, result) assert.Equal(t, tempDir, result)
@@ -117,7 +131,7 @@ func TestTestDataDirs(t *testing.T) {
// Test with no valid directories // Test with no valid directories
t.Run("no valid directories", func(t *testing.T) { t.Run("no valid directories", func(t *testing.T) {
invalidPaths := []string{invalidDataDir(t), invalidDataDir(t)} invalidPaths := []string{"/invalid/path1", "/invalid/path2"}
_, err := testDataDirs(invalidPaths) _, err := testDataDirs(invalidPaths)
assert.Error(t, err) assert.Error(t, err)
assert.Contains(t, err.Error(), "data directory not found") assert.Contains(t, err.Error(), "data directory not found")
+143 -559
View File
@@ -1,7 +1,6 @@
package agent package agent
import ( import (
"context"
"log/slog" "log/slog"
"os" "os"
"path/filepath" "path/filepath"
@@ -9,60 +8,11 @@ import (
"strings" "strings"
"time" "time"
"github.com/henrygd/beszel/agent/utils"
"github.com/henrygd/beszel/internal/entities/system" "github.com/henrygd/beszel/internal/entities/system"
"github.com/shirou/gopsutil/v4/disk" "github.com/shirou/gopsutil/v4/disk"
) )
// fsRegistrationContext holds the shared lookup state needed to resolve a
// filesystem into the tracked fsStats key and metadata.
type fsRegistrationContext struct {
filesystem string // device part of optional FILESYSTEM env var
filesystemName string // optional custom name from FILESYSTEM=device__name
isWindows bool
efPath string // path to extra filesystems (default "/extra-filesystems")
diskIoCounters map[string]disk.IOCountersStat
}
// diskDiscovery groups the transient state for a single initializeDiskInfo run so
// helper methods can share the same partitions, mount paths, and lookup functions
type diskDiscovery struct {
agent *Agent
rootMountPoint string
partitions []disk.PartitionStat
usageFn func(string) (*disk.UsageStat, error)
ctx fsRegistrationContext
}
// prevDisk stores previous per-device disk counters for a given cache interval
type prevDisk struct {
readBytes uint64
writeBytes uint64
readTime uint64 // cumulative ms spent on reads (from ReadTime)
writeTime uint64 // cumulative ms spent on writes (from WriteTime)
ioTime uint64 // cumulative ms spent doing I/O (from IoTime)
weightedIO uint64 // cumulative weighted ms (queue-depth × ms, from WeightedIO)
readCount uint64 // cumulative read operation count
writeCount uint64 // cumulative write operation count
at time.Time
}
// prevDiskFromCounter creates a prevDisk snapshot from a disk.IOCountersStat at time t.
func prevDiskFromCounter(d disk.IOCountersStat, t time.Time) prevDisk {
return prevDisk{
readBytes: d.ReadBytes,
writeBytes: d.WriteBytes,
readTime: d.ReadTime,
writeTime: d.WriteTime,
ioTime: d.IoTime,
weightedIO: d.WeightedIO,
readCount: d.ReadCount,
writeCount: d.WriteCount,
at: t,
}
}
// parseFilesystemEntry parses a filesystem entry in the format "device__customname" // parseFilesystemEntry parses a filesystem entry in the format "device__customname"
// Returns the device/filesystem part and the custom name part // Returns the device/filesystem part and the custom name part
func parseFilesystemEntry(entry string) (device, customName string) { func parseFilesystemEntry(entry string) (device, customName string) {
@@ -76,237 +26,14 @@ func parseFilesystemEntry(entry string) (device, customName string) {
return device, customName return device, customName
} }
// extraFilesystemPartitionInfo derives the I/O device and optional display name
// for a mounted /extra-filesystems partition. Prefer the partition device reported
// by the system and only use the folder name for custom naming metadata.
func extraFilesystemPartitionInfo(p disk.PartitionStat) (device, customName string) {
device = strings.TrimSpace(p.Device)
folderDevice, customName := parseFilesystemEntry(filepath.Base(p.Mountpoint))
if device == "" {
device = folderDevice
}
return device, customName
}
func isDockerSpecialMountpoint(mountpoint string) bool {
switch mountpoint {
case "/etc/hosts", "/etc/resolv.conf", "/etc/hostname":
return true
}
return false
}
// registerFilesystemStats resolves the tracked key and stats payload for a
// filesystem before it is inserted into fsStats.
func registerFilesystemStats(existing map[string]*system.FsStats, device, mountpoint string, root bool, customName string, ctx fsRegistrationContext) (string, *system.FsStats, bool) {
key := device
if !ctx.isWindows {
key = filepath.Base(device)
}
if root {
// Try to map root device to a diskIoCounters entry. First checks for an
// exact key match, then uses findIoDevice for normalized / prefix-based
// matching (e.g. nda0p2 -> nda0), and finally falls back to FILESYSTEM.
if _, ioMatch := ctx.diskIoCounters[key]; !ioMatch {
if matchedKey, match := findIoDevice(key, ctx.diskIoCounters); match {
key = matchedKey
} else if ctx.filesystem != "" {
if matchedKey, match := findIoDevice(ctx.filesystem, ctx.diskIoCounters); match {
key = matchedKey
}
}
if _, ioMatch = ctx.diskIoCounters[key]; !ioMatch {
slog.Warn("Root I/O unmapped; set FILESYSTEM", "device", device, "mountpoint", mountpoint)
}
}
} else {
// Check if non-root has diskstats and prefer the folder device for
// /extra-filesystems mounts when the discovered partition device is a
// mapper path (e.g. luks UUID) that obscures the underlying block device.
if _, ioMatch := ctx.diskIoCounters[key]; !ioMatch {
if strings.HasPrefix(mountpoint, ctx.efPath) {
folderDevice, _ := parseFilesystemEntry(filepath.Base(mountpoint))
if folderDevice != "" {
if matchedKey, match := findIoDevice(folderDevice, ctx.diskIoCounters); match {
key = matchedKey
}
}
}
if _, ioMatch = ctx.diskIoCounters[key]; !ioMatch {
if matchedKey, match := findIoDevice(key, ctx.diskIoCounters); match {
key = matchedKey
}
}
}
}
if _, exists := existing[key]; exists {
return "", nil, false
}
fsStats := &system.FsStats{Root: root, Mountpoint: mountpoint}
if customName != "" {
fsStats.Name = customName
}
return key, fsStats, true
}
// addFsStat inserts a discovered filesystem if it resolves to a new tracking
// key. The key selection itself lives in buildFsStatRegistration so that logic
// can stay directly unit-tested.
func (d *diskDiscovery) addFsStat(device, mountpoint string, root bool, customName string) {
key, fsStats, ok := registerFilesystemStats(d.agent.fsStats, device, mountpoint, root, customName, d.ctx)
if !ok {
return
}
d.agent.fsStats[key] = fsStats
name := key
if customName != "" {
name = customName
}
slog.Info("Detected disk", "name", name, "device", device, "mount", mountpoint, "io", key, "root", root)
}
// addConfiguredRootFs resolves FILESYSTEM against partitions first, then falls
// back to direct diskstats matching for setups like ZFS where partitions do not
// expose the physical device name.
func (d *diskDiscovery) addConfiguredRootFs() bool {
if d.ctx.filesystem == "" {
return false
}
for _, p := range d.partitions {
if filesystemMatchesPartitionSetting(d.ctx.filesystem, p) {
d.addFsStat(p.Device, p.Mountpoint, true, d.ctx.filesystemName)
return true
}
}
// FILESYSTEM may name a physical disk absent from partitions (e.g. ZFS lists
// dataset paths like zroot/ROOT/default, not block devices).
if ioKey, match := findIoDevice(d.ctx.filesystem, d.ctx.diskIoCounters); match {
d.agent.fsStats[ioKey] = &system.FsStats{Root: true, Mountpoint: d.rootMountPoint, Name: d.ctx.filesystemName}
return true
}
slog.Warn("Partition details not found", "filesystem", d.ctx.filesystem)
return false
}
func isRootFallbackPartition(p disk.PartitionStat, rootMountPoint string) bool {
return p.Mountpoint == rootMountPoint ||
(isDockerSpecialMountpoint(p.Mountpoint) && strings.HasPrefix(p.Device, "/dev"))
}
// addPartitionRootFs handles the non-configured root fallback path when a
// partition looks like the active root mount but still needs translating to an
// I/O device key.
func (d *diskDiscovery) addPartitionRootFs(device, mountpoint string) bool {
fs, match := findIoDevice(filepath.Base(device), d.ctx.diskIoCounters)
if !match {
return false
}
// The resolved I/O device is already known here, so use it directly to avoid
// a second fallback search inside buildFsStatRegistration.
d.addFsStat(fs, mountpoint, true, "")
return true
}
// addLastResortRootFs is only used when neither FILESYSTEM nor partition-based
// heuristics can identify root, so it picks the busiest I/O device as a final
// fallback and preserves the root mountpoint for usage collection.
func (d *diskDiscovery) addLastResortRootFs() {
rootKey := mostActiveIoDevice(d.ctx.diskIoCounters)
if rootKey != "" {
slog.Warn("Using most active device for root I/O; set FILESYSTEM to override", "device", rootKey)
} else {
rootKey = filepath.Base(d.rootMountPoint)
if _, exists := d.agent.fsStats[rootKey]; exists {
rootKey = "root"
}
slog.Warn("Root I/O device not detected; set FILESYSTEM to override")
}
d.agent.fsStats[rootKey] = &system.FsStats{Root: true, Mountpoint: d.rootMountPoint}
}
// findPartitionByFilesystemSetting matches an EXTRA_FILESYSTEMS entry against a
// discovered partition either by mountpoint or by device suffix.
func findPartitionByFilesystemSetting(filesystem string, partitions []disk.PartitionStat) (disk.PartitionStat, bool) {
for _, p := range partitions {
if strings.HasSuffix(p.Device, filesystem) || p.Mountpoint == filesystem {
return p, true
}
}
return disk.PartitionStat{}, false
}
// addConfiguredExtraFsEntry resolves one EXTRA_FILESYSTEMS entry, preferring a
// discovered partition and falling back to any path that disk.Usage accepts.
func (d *diskDiscovery) addConfiguredExtraFsEntry(filesystem, customName string) {
if p, found := findPartitionByFilesystemSetting(filesystem, d.partitions); found {
d.addFsStat(p.Device, p.Mountpoint, false, customName)
return
}
if _, err := d.usageFn(filesystem); err == nil {
d.addFsStat(filepath.Base(filesystem), filesystem, false, customName)
return
} else {
slog.Error("Invalid filesystem", "name", filesystem, "err", err)
}
}
// addConfiguredExtraFilesystems parses and registers the comma-separated
// EXTRA_FILESYSTEMS env var entries.
func (d *diskDiscovery) addConfiguredExtraFilesystems(extraFilesystems string) {
for fsEntry := range strings.SplitSeq(extraFilesystems, ",") {
filesystem, customName := parseFilesystemEntry(fsEntry)
d.addConfiguredExtraFsEntry(filesystem, customName)
}
}
// addPartitionExtraFs registers partitions mounted under /extra-filesystems so
// their display names can come from the folder name while their I/O keys still
// prefer the underlying partition device. Only direct children are matched to
// avoid registering nested virtual mounts (e.g. /proc, /sys) that are returned by
// disk.Partitions(true) when the host root is bind-mounted in /extra-filesystems.
func (d *diskDiscovery) addPartitionExtraFs(p disk.PartitionStat) {
if filepath.Dir(p.Mountpoint) != d.ctx.efPath {
return
}
device, customName := extraFilesystemPartitionInfo(p)
d.addFsStat(device, p.Mountpoint, false, customName)
}
// addExtraFilesystemFolders handles bare directories under /extra-filesystems
// that may not appear in partition discovery, while skipping mountpoints that
// were already registered from higher-fidelity sources.
func (d *diskDiscovery) addExtraFilesystemFolders(folderNames []string) {
existingMountpoints := make(map[string]bool, len(d.agent.fsStats))
for _, stats := range d.agent.fsStats {
existingMountpoints[stats.Mountpoint] = true
}
for _, folderName := range folderNames {
mountpoint := filepath.Join(d.ctx.efPath, folderName)
slog.Debug("/extra-filesystems", "mountpoint", mountpoint)
if existingMountpoints[mountpoint] {
continue
}
device, customName := parseFilesystemEntry(folderName)
d.addFsStat(device, mountpoint, false, customName)
}
}
// Sets up the filesystems to monitor for disk usage and I/O. // Sets up the filesystems to monitor for disk usage and I/O.
func (a *Agent) initializeDiskInfo() { func (a *Agent) initializeDiskInfo() {
filesystemRaw, _ := utils.GetEnv("FILESYSTEM") filesystem, _ := GetEnv("FILESYSTEM")
filesystem, filesystemName := parseFilesystemEntry(filesystemRaw) efPath := "/extra-filesystems"
hasRoot := false hasRoot := false
isWindows := runtime.GOOS == "windows" isWindows := runtime.GOOS == "windows"
partitions, err := disk.PartitionsWithContext(context.Background(), true) partitions, err := disk.Partitions(false)
if err != nil { if err != nil {
slog.Error("Error getting disk partitions", "err", err) slog.Error("Error getting disk partitions", "err", err)
} }
@@ -319,234 +46,168 @@ func (a *Agent) initializeDiskInfo() {
} }
} }
// ioContext := context.WithValue(a.sensorsContext,
// common.EnvKey, common.EnvMap{common.HostProcEnvKey: "/tmp/testproc"},
// )
// diskIoCounters, err := disk.IOCountersWithContext(ioContext)
diskIoCounters, err := disk.IOCounters() diskIoCounters, err := disk.IOCounters()
if err != nil { if err != nil {
slog.Error("Error getting diskstats", "err", err) slog.Error("Error getting diskstats", "err", err)
} }
slog.Debug("Disk I/O", "diskstats", diskIoCounters) slog.Debug("Disk I/O", "diskstats", diskIoCounters)
ctx := fsRegistrationContext{
filesystem: filesystem, // Helper function to add a filesystem to fsStats if it doesn't exist
filesystemName: filesystemName, addFsStat := func(device, mountpoint string, root bool, customName ...string) {
isWindows: isWindows, var key string
diskIoCounters: diskIoCounters, if isWindows {
efPath: "/extra-filesystems", key = device
} else {
key = filepath.Base(device)
}
var ioMatch bool
if _, exists := a.fsStats[key]; !exists {
if root {
slog.Info("Detected root device", "name", key)
// Check if root device is in /proc/diskstats, use fallback if not
if _, ioMatch = diskIoCounters[key]; !ioMatch {
key, ioMatch = findIoDevice(filesystem, diskIoCounters, a.fsStats)
if !ioMatch {
slog.Info("Using I/O fallback", "device", device, "mountpoint", mountpoint, "fallback", key)
}
}
} else {
// Check if non-root has diskstats and fall back to folder name if not
// Scenario: device is encrypted and named luks-2bcb02be-999d-4417-8d18-5c61e660fb6e - not in /proc/diskstats.
// However, the device can be specified by mounting folder from luks device at /extra-filesystems/sda1
if _, ioMatch = diskIoCounters[key]; !ioMatch {
efBase := filepath.Base(mountpoint)
if _, ioMatch = diskIoCounters[efBase]; ioMatch {
key = efBase
}
}
}
fsStats := &system.FsStats{Root: root, Mountpoint: mountpoint}
if len(customName) > 0 && customName[0] != "" {
fsStats.Name = customName[0]
}
a.fsStats[key] = fsStats
}
} }
// Get the appropriate root mount point for this system // Get the appropriate root mount point for this system
discovery := diskDiscovery{ rootMountPoint := a.getRootMountPoint()
agent: a,
rootMountPoint: a.getRootMountPoint(), // Use FILESYSTEM env var to find root filesystem
partitions: partitions, if filesystem != "" {
usageFn: disk.Usage, for _, p := range partitions {
ctx: ctx, if strings.HasSuffix(p.Device, filesystem) || p.Mountpoint == filesystem {
addFsStat(p.Device, p.Mountpoint, true)
hasRoot = true
break
}
}
if !hasRoot {
slog.Warn("Partition details not found", "filesystem", filesystem)
}
} }
hasRoot = discovery.addConfiguredRootFs()
// Add EXTRA_FILESYSTEMS env var values to fsStats // Add EXTRA_FILESYSTEMS env var values to fsStats
if extraFilesystems, exists := utils.GetEnv("EXTRA_FILESYSTEMS"); exists { if extraFilesystems, exists := GetEnv("EXTRA_FILESYSTEMS"); exists {
discovery.addConfiguredExtraFilesystems(extraFilesystems) for _, fsEntry := range strings.Split(extraFilesystems, ",") {
// Parse custom name from format: device__customname
fs, customName := parseFilesystemEntry(fsEntry)
found := false
for _, p := range partitions {
if strings.HasSuffix(p.Device, fs) || p.Mountpoint == fs {
addFsStat(p.Device, p.Mountpoint, false, customName)
found = true
break
}
}
// if not in partitions, test if we can get disk usage
if !found {
if _, err := disk.Usage(fs); err == nil {
addFsStat(filepath.Base(fs), fs, false, customName)
} else {
slog.Error("Invalid filesystem", "name", fs, "err", err)
}
}
}
} }
// Process partitions for various mount points // Process partitions for various mount points
for _, p := range partitions { for _, p := range partitions {
if !hasRoot && isRootFallbackPartition(p, discovery.rootMountPoint) { // fmt.Println(p.Device, p.Mountpoint)
hasRoot = discovery.addPartitionRootFs(p.Device, p.Mountpoint) // Binary root fallback or docker root fallback
if !hasRoot && (p.Mountpoint == rootMountPoint || (p.Mountpoint == "/etc/hosts" && strings.HasPrefix(p.Device, "/dev"))) {
fs, match := findIoDevice(filepath.Base(p.Device), diskIoCounters, a.fsStats)
if match {
addFsStat(fs, p.Mountpoint, true)
hasRoot = true
}
}
// Check if device is in /extra-filesystems
if strings.HasPrefix(p.Mountpoint, efPath) {
device, customName := parseFilesystemEntry(p.Mountpoint)
addFsStat(device, p.Mountpoint, false, customName)
} }
discovery.addPartitionExtraFs(p)
} }
// Check all folders in /extra-filesystems and add them if not already present // Check all folders in /extra-filesystems and add them if not already present
if folders, err := os.ReadDir(discovery.ctx.efPath); err == nil { if folders, err := os.ReadDir(efPath); err == nil {
folderNames := make([]string, 0, len(folders)) existingMountpoints := make(map[string]bool)
for _, stats := range a.fsStats {
existingMountpoints[stats.Mountpoint] = true
}
for _, folder := range folders { for _, folder := range folders {
if folder.IsDir() { if folder.IsDir() {
folderNames = append(folderNames, folder.Name()) mountpoint := filepath.Join(efPath, folder.Name())
slog.Debug("/extra-filesystems", "mountpoint", mountpoint)
if !existingMountpoints[mountpoint] {
device, customName := parseFilesystemEntry(folder.Name())
addFsStat(device, mountpoint, false, customName)
}
} }
} }
discovery.addExtraFilesystemFolders(folderNames)
} }
// If no root filesystem set, try the most active I/O device as a last // If no root filesystem set, use fallback
// resort (e.g. ZFS where dataset names are unrelated to disk names).
if !hasRoot { if !hasRoot {
discovery.addLastResortRootFs() rootDevice, _ := findIoDevice(filepath.Base(filesystem), diskIoCounters, a.fsStats)
slog.Info("Root disk", "mountpoint", rootMountPoint, "io", rootDevice)
a.fsStats[rootDevice] = &system.FsStats{Root: true, Mountpoint: rootMountPoint}
} }
a.pruneDuplicateRootExtraFilesystems()
a.initializeDiskIoStats(diskIoCounters) a.initializeDiskIoStats(diskIoCounters)
} }
// Removes extra filesystems that mirror root usage (https://github.com/henrygd/beszel/issues/1428). // Returns matching device from /proc/diskstats,
func (a *Agent) pruneDuplicateRootExtraFilesystems() { // or the device with the most reads if no match is found.
var rootMountpoint string // bool is true if a match was found.
for _, stats := range a.fsStats { func findIoDevice(filesystem string, diskIoCounters map[string]disk.IOCountersStat, fsStats map[string]*system.FsStats) (string, bool) {
if stats != nil && stats.Root { var maxReadBytes uint64
rootMountpoint = stats.Mountpoint maxReadDevice := "/"
break
}
}
if rootMountpoint == "" {
return
}
rootUsage, err := disk.Usage(rootMountpoint)
if err != nil {
return
}
for name, stats := range a.fsStats {
if stats == nil || stats.Root {
continue
}
extraUsage, err := disk.Usage(stats.Mountpoint)
if err != nil {
continue
}
if hasSameDiskUsage(rootUsage, extraUsage) {
slog.Info("Ignoring duplicate FS", "name", name, "mount", stats.Mountpoint)
delete(a.fsStats, name)
}
}
}
// hasSameDiskUsage compares root/extra usage with a small byte tolerance.
func hasSameDiskUsage(a, b *disk.UsageStat) bool {
if a == nil || b == nil || a.Total == 0 || b.Total == 0 {
return false
}
// Allow minor drift between sequential disk usage calls.
const toleranceBytes uint64 = 16 * 1024 * 1024
return withinUsageTolerance(a.Total, b.Total, toleranceBytes) &&
withinUsageTolerance(a.Used, b.Used, toleranceBytes)
}
// withinUsageTolerance reports whether two byte values differ by at most tolerance.
func withinUsageTolerance(a, b, tolerance uint64) bool {
if a >= b {
return a-b <= tolerance
}
return b-a <= tolerance
}
type ioMatchCandidate struct {
name string
bytes uint64
ops uint64
}
// findIoDevice prefers exact device/label matches, then falls back to a
// prefix-related candidate with the highest recent activity.
func findIoDevice(filesystem string, diskIoCounters map[string]disk.IOCountersStat) (string, bool) {
filesystem = normalizeDeviceName(filesystem)
if filesystem == "" {
return "", false
}
candidates := []ioMatchCandidate{}
for _, d := range diskIoCounters { for _, d := range diskIoCounters {
if normalizeDeviceName(d.Name) == filesystem || (d.Label != "" && normalizeDeviceName(d.Label) == filesystem) { if d.Name == filesystem || (d.Label != "" && d.Label == filesystem) {
return d.Name, true return d.Name, true
} }
if prefixRelated(normalizeDeviceName(d.Name), filesystem) || if d.ReadBytes > maxReadBytes {
(d.Label != "" && prefixRelated(normalizeDeviceName(d.Label), filesystem)) { // don't use if device already exists in fsStats
candidates = append(candidates, ioMatchCandidate{ if _, exists := fsStats[d.Name]; !exists {
name: d.Name, maxReadBytes = d.ReadBytes
bytes: d.ReadBytes + d.WriteBytes, maxReadDevice = d.Name
ops: d.ReadCount + d.WriteCount, }
})
} }
} }
return maxReadDevice, false
if len(candidates) == 0 {
return "", false
}
best := candidates[0]
for _, c := range candidates[1:] {
if c.bytes > best.bytes ||
(c.bytes == best.bytes && c.ops > best.ops) ||
(c.bytes == best.bytes && c.ops == best.ops && c.name < best.name) {
best = c
}
}
slog.Info("Using disk I/O fallback", "requested", filesystem, "selected", best.name)
return best.name, true
}
// mostActiveIoDevice returns the device with the highest I/O activity,
// or "" if diskIoCounters is empty.
func mostActiveIoDevice(diskIoCounters map[string]disk.IOCountersStat) string {
var best ioMatchCandidate
for _, d := range diskIoCounters {
c := ioMatchCandidate{
name: d.Name,
bytes: d.ReadBytes + d.WriteBytes,
ops: d.ReadCount + d.WriteCount,
}
if best.name == "" || c.bytes > best.bytes ||
(c.bytes == best.bytes && c.ops > best.ops) ||
(c.bytes == best.bytes && c.ops == best.ops && c.name < best.name) {
best = c
}
}
return best.name
}
// prefixRelated reports whether either identifier is a prefix of the other.
func prefixRelated(a, b string) bool {
if a == "" || b == "" || a == b {
return false
}
return strings.HasPrefix(a, b) || strings.HasPrefix(b, a)
}
// filesystemMatchesPartitionSetting checks whether a FILESYSTEM env var value
// matches a partition by mountpoint, exact device name, or prefix relationship
// (e.g. FILESYSTEM=ada0 matches partition /dev/ada0p2).
func filesystemMatchesPartitionSetting(filesystem string, p disk.PartitionStat) bool {
filesystem = strings.TrimSpace(filesystem)
if filesystem == "" {
return false
}
if p.Mountpoint == filesystem {
return true
}
fsName := normalizeDeviceName(filesystem)
partName := normalizeDeviceName(p.Device)
if fsName == "" || partName == "" {
return false
}
if fsName == partName {
return true
}
return prefixRelated(partName, fsName)
}
// normalizeDeviceName canonicalizes device strings for comparisons.
func normalizeDeviceName(value string) string {
name := filepath.Base(strings.TrimSpace(value))
if name == "." {
return ""
}
return name
} }
// Sets start values for disk I/O stats. // Sets start values for disk I/O stats.
func (a *Agent) initializeDiskIoStats(diskIoCounters map[string]disk.IOCountersStat) { func (a *Agent) initializeDiskIoStats(diskIoCounters map[string]disk.IOCountersStat) {
a.fsNames = a.fsNames[:0]
now := time.Now()
// ZFS datasets have no /proc/diskstats entry, so they are excluded from
// I/O tracking instead of warning about a missing device (#1541).
var zfsMountpoints map[string]bool
if a.storagePoolManager != nil {
zfsMountpoints = a.storagePoolManager.ZfsMountpoints()
}
for device, stats := range a.fsStats { for device, stats := range a.fsStats {
if zfsMountpoints[stats.Mountpoint] {
continue
}
// skip if not in diskIoCounters // skip if not in diskIoCounters
d, exists := diskIoCounters[device] d, exists := diskIoCounters[device]
if !exists { if !exists {
@@ -554,7 +215,7 @@ func (a *Agent) initializeDiskIoStats(diskIoCounters map[string]disk.IOCountersS
continue continue
} }
// populate initial values // populate initial values
stats.Time = now stats.Time = time.Now()
stats.TotalRead = d.ReadBytes stats.TotalRead = d.ReadBytes
stats.TotalWrite = d.WriteBytes stats.TotalWrite = d.WriteBytes
// add to list of valid io device names // add to list of valid io device names
@@ -571,31 +232,20 @@ func (a *Agent) updateDiskUsage(systemStats *system.Stats) {
!a.lastDiskUsageUpdate.IsZero() && !a.lastDiskUsageUpdate.IsZero() &&
time.Since(a.lastDiskUsageUpdate) < a.diskUsageCacheDuration time.Since(a.lastDiskUsageUpdate) < a.diskUsageCacheDuration
// ZFS dataset mountpoints use `zfs list` values because statfs(2) reports
// dataset-level usage that excludes child datasets (#1541).
var zfsUsage map[string]zfsDatasetUsage
if a.storagePoolManager != nil {
zfsUsage = a.storagePoolManager.DatasetUsage()
}
// disk usage // disk usage
for _, stats := range a.fsStats { for _, stats := range a.fsStats {
// Skip non-root filesystems if caching is active // Skip non-root filesystems if caching is active
if cacheExtraFs && !stats.Root { if cacheExtraFs && !stats.Root {
continue continue
} }
var total, used uint64 if d, err := disk.Usage(stats.Mountpoint); err == nil {
var usedPct float64 stats.DiskTotal = bytesToGigabytes(d.Total)
if u, ok := zfsUsage[stats.Mountpoint]; ok { stats.DiskUsed = bytesToGigabytes(d.Used)
total = u.used + u.avail if stats.Root {
used = u.used systemStats.DiskTotal = bytesToGigabytes(d.Total)
if total > 0 { systemStats.DiskUsed = bytesToGigabytes(d.Used)
usedPct = float64(used) / float64(total) * 100 systemStats.DiskPct = twoDecimals(d.UsedPercent)
} }
} else if d, err := disk.Usage(stats.Mountpoint); err == nil {
total = d.Total
used = d.Used
usedPct = d.UsedPercent
} else { } else {
// reset stats if error (likely unmounted) // reset stats if error (likely unmounted)
slog.Error("Error getting disk stats", "name", stats.Mountpoint, "err", err) slog.Error("Error getting disk stats", "name", stats.Mountpoint, "err", err)
@@ -603,14 +253,6 @@ func (a *Agent) updateDiskUsage(systemStats *system.Stats) {
stats.DiskUsed = 0 stats.DiskUsed = 0
stats.TotalRead = 0 stats.TotalRead = 0
stats.TotalWrite = 0 stats.TotalWrite = 0
continue
}
stats.DiskTotal = utils.BytesToGigabytes(total)
stats.DiskUsed = utils.BytesToGigabytes(used)
if stats.Root {
systemStats.DiskTotal = stats.DiskTotal
systemStats.DiskUsed = stats.DiskUsed
systemStats.DiskPct = utils.TwoDecimals(usedPct)
} }
} }
@@ -640,72 +282,36 @@ func (a *Agent) updateDiskIo(cacheTimeMs uint16, systemStats *system.Stats) {
prev, hasPrev := a.diskPrev[cacheTimeMs][name] prev, hasPrev := a.diskPrev[cacheTimeMs][name]
if !hasPrev { if !hasPrev {
// Seed from agent-level fsStats if present, else seed from current // Seed from agent-level fsStats if present, else seed from current
prev = prevDisk{ prev = prevDisk{readBytes: stats.TotalRead, writeBytes: stats.TotalWrite, at: stats.Time}
readBytes: stats.TotalRead,
writeBytes: stats.TotalWrite,
readTime: d.ReadTime,
writeTime: d.WriteTime,
ioTime: d.IoTime,
weightedIO: d.WeightedIO,
readCount: d.ReadCount,
writeCount: d.WriteCount,
at: stats.Time,
}
if prev.at.IsZero() { if prev.at.IsZero() {
prev = prevDiskFromCounter(d, now) prev = prevDisk{readBytes: d.ReadBytes, writeBytes: d.WriteBytes, at: now}
} }
} }
msElapsed := uint64(now.Sub(prev.at).Milliseconds()) msElapsed := uint64(now.Sub(prev.at).Milliseconds())
// Update per-interval snapshot
a.diskPrev[cacheTimeMs][name] = prevDiskFromCounter(d, now)
// Avoid division by zero or clock issues
if msElapsed < 100 { if msElapsed < 100 {
// Avoid division by zero or clock issues; update snapshot and continue
a.diskPrev[cacheTimeMs][name] = prevDisk{readBytes: d.ReadBytes, writeBytes: d.WriteBytes, at: now}
continue continue
} }
diskIORead := (d.ReadBytes - prev.readBytes) * 1000 / msElapsed diskIORead := (d.ReadBytes - prev.readBytes) * 1000 / msElapsed
diskIOWrite := (d.WriteBytes - prev.writeBytes) * 1000 / msElapsed diskIOWrite := (d.WriteBytes - prev.writeBytes) * 1000 / msElapsed
readMbPerSecond := utils.BytesToMegabytes(float64(diskIORead)) readMbPerSecond := bytesToMegabytes(float64(diskIORead))
writeMbPerSecond := utils.BytesToMegabytes(float64(diskIOWrite)) writeMbPerSecond := bytesToMegabytes(float64(diskIOWrite))
// validate values // validate values
if readMbPerSecond > 50_000 || writeMbPerSecond > 50_000 { if readMbPerSecond > 50_000 || writeMbPerSecond > 50_000 {
slog.Warn("Invalid disk I/O. Resetting.", "name", d.Name, "read", readMbPerSecond, "write", writeMbPerSecond) slog.Warn("Invalid disk I/O. Resetting.", "name", d.Name, "read", readMbPerSecond, "write", writeMbPerSecond)
// Reset interval snapshot and seed from current
a.diskPrev[cacheTimeMs][name] = prevDisk{readBytes: d.ReadBytes, writeBytes: d.WriteBytes, at: now}
// also refresh agent baseline to avoid future negatives // also refresh agent baseline to avoid future negatives
a.initializeDiskIoStats(ioCounters) a.initializeDiskIoStats(ioCounters)
continue continue
} }
// These properties are calculated differently on different platforms, // Update per-interval snapshot
// but generally represent cumulative time spent doing reads/writes on the device. a.diskPrev[cacheTimeMs][name] = prevDisk{readBytes: d.ReadBytes, writeBytes: d.WriteBytes, at: now}
// This can surpass 100% if there are multiple concurrent I/O operations.
// Linux kernel docs:
// This is the total number of milliseconds spent by all reads (as
// measured from __make_request() to end_that_request_last()).
// https://www.kernel.org/doc/Documentation/iostats.txt (fields 4, 8)
diskReadTime := utils.TwoDecimals(float64(d.ReadTime-prev.readTime) / float64(msElapsed) * 100)
diskWriteTime := utils.TwoDecimals(float64(d.WriteTime-prev.writeTime) / float64(msElapsed) * 100)
// I/O utilization %: fraction of wall time the device had any I/O in progress (0-100).
diskIoUtilPct := utils.TwoDecimals(float64(d.IoTime-prev.ioTime) / float64(msElapsed) * 100)
// Weighted I/O: queue-depth weighted I/O time, normalized to interval (can exceed 100%).
// Linux kernel field 11: incremented by iops_in_progress × ms_since_last_update.
// Used to display queue depth. Multipled by 100 to increase accuracy of digit truncation (divided by 100 in UI).
diskWeightedIO := utils.TwoDecimals(float64(d.WeightedIO-prev.weightedIO) / float64(msElapsed) * 100)
// r_await / w_await: average time per read/write operation in milliseconds.
// Equivalent to r_await and w_await in iostat.
var rAwait, wAwait float64
if deltaReadCount := d.ReadCount - prev.readCount; deltaReadCount > 0 {
rAwait = utils.TwoDecimals(float64(d.ReadTime-prev.readTime) / float64(deltaReadCount))
}
if deltaWriteCount := d.WriteCount - prev.writeCount; deltaWriteCount > 0 {
wAwait = utils.TwoDecimals(float64(d.WriteTime-prev.writeTime) / float64(deltaWriteCount))
}
// Update global fsStats baseline for cross-interval correctness // Update global fsStats baseline for cross-interval correctness
stats.Time = now stats.Time = now
@@ -715,42 +321,20 @@ func (a *Agent) updateDiskIo(cacheTimeMs uint16, systemStats *system.Stats) {
stats.DiskWritePs = writeMbPerSecond stats.DiskWritePs = writeMbPerSecond
stats.DiskReadBytes = diskIORead stats.DiskReadBytes = diskIORead
stats.DiskWriteBytes = diskIOWrite stats.DiskWriteBytes = diskIOWrite
stats.DiskIoStats[0] = diskReadTime
stats.DiskIoStats[1] = diskWriteTime
stats.DiskIoStats[2] = diskIoUtilPct
stats.DiskIoStats[3] = rAwait
stats.DiskIoStats[4] = wAwait
stats.DiskIoStats[5] = diskWeightedIO
if stats.Root { if stats.Root {
systemStats.DiskReadPs = stats.DiskReadPs systemStats.DiskReadPs = stats.DiskReadPs
systemStats.DiskWritePs = stats.DiskWritePs systemStats.DiskWritePs = stats.DiskWritePs
systemStats.DiskIO[0] = diskIORead systemStats.DiskIO[0] = diskIORead
systemStats.DiskIO[1] = diskIOWrite systemStats.DiskIO[1] = diskIOWrite
systemStats.DiskIOTotal[0] = d.ReadBytes
systemStats.DiskIOTotal[1] = d.WriteBytes
systemStats.DiskIoStats[0] = diskReadTime
systemStats.DiskIoStats[1] = diskWriteTime
systemStats.DiskIoStats[2] = diskIoUtilPct
systemStats.DiskIoStats[3] = rAwait
systemStats.DiskIoStats[4] = wAwait
systemStats.DiskIoStats[5] = diskWeightedIO
} }
} }
} }
} }
// getRootMountPoint returns the appropriate root mount point for the system. // getRootMountPoint returns the appropriate root mount point for the system
// On Windows it returns the system drive (e.g. "C:").
// For immutable systems like Fedora Silverblue, it returns /sysroot instead of / // For immutable systems like Fedora Silverblue, it returns /sysroot instead of /
func (a *Agent) getRootMountPoint() string { func (a *Agent) getRootMountPoint() string {
if runtime.GOOS == "windows" {
if sd := os.Getenv("SystemDrive"); sd != "" {
return sd
}
return "C:"
}
// 1. Check if /etc/os-release contains indicators of an immutable system // 1. Check if /etc/os-release contains indicators of an immutable system
if osReleaseContent, err := os.ReadFile("/etc/os-release"); err == nil { if osReleaseContent, err := os.ReadFile("/etc/os-release"); err == nil {
content := string(osReleaseContent) content := string(osReleaseContent)
+22 -746
View File
@@ -1,4 +1,5 @@
//go:build testing //go:build testing
// +build testing
package agent package agent
@@ -78,7 +79,14 @@ func TestParseFilesystemEntry(t *testing.T) {
for _, tt := range tests { for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) { t.Run(tt.name, func(t *testing.T) {
fs, customName := parseFilesystemEntry(tt.input) fsEntry := strings.TrimSpace(tt.input)
var fs, customName string
if parts := strings.SplitN(fsEntry, "__", 2); len(parts) == 2 {
fs = strings.TrimSpace(parts[0])
customName = strings.TrimSpace(parts[1])
} else {
fs = fsEntry
}
assert.Equal(t, tt.expectedFs, fs) assert.Equal(t, tt.expectedFs, fs)
assert.Equal(t, tt.expectedName, customName) assert.Equal(t, tt.expectedName, customName)
@@ -86,687 +94,19 @@ func TestParseFilesystemEntry(t *testing.T) {
} }
} }
func TestExtraFilesystemPartitionInfo(t *testing.T) {
t.Run("uses partition device for label-only mountpoint", func(t *testing.T) {
device, customName := extraFilesystemPartitionInfo(disk.PartitionStat{
Device: "/dev/sdc",
Mountpoint: "/extra-filesystems/Share",
})
assert.Equal(t, "/dev/sdc", device)
assert.Equal(t, "", customName)
})
t.Run("uses custom name from mountpoint suffix", func(t *testing.T) {
device, customName := extraFilesystemPartitionInfo(disk.PartitionStat{
Device: "/dev/sdc",
Mountpoint: "/extra-filesystems/sdc__Share",
})
assert.Equal(t, "/dev/sdc", device)
assert.Equal(t, "Share", customName)
})
t.Run("falls back to folder device when partition device is unavailable", func(t *testing.T) {
device, customName := extraFilesystemPartitionInfo(disk.PartitionStat{
Mountpoint: "/extra-filesystems/sdc__Share",
})
assert.Equal(t, "sdc", device)
assert.Equal(t, "Share", customName)
})
t.Run("supports custom name without folder device prefix", func(t *testing.T) {
device, customName := extraFilesystemPartitionInfo(disk.PartitionStat{
Device: "/dev/sdc",
Mountpoint: "/extra-filesystems/__Share",
})
assert.Equal(t, "/dev/sdc", device)
assert.Equal(t, "Share", customName)
})
}
func TestBuildFsStatRegistration(t *testing.T) {
t.Run("uses basename for non-windows exact io match", func(t *testing.T) {
key, stats, ok := registerFilesystemStats(
map[string]*system.FsStats{},
"/dev/sda1",
"/mnt/data",
false,
"archive",
fsRegistrationContext{
isWindows: false,
diskIoCounters: map[string]disk.IOCountersStat{
"sda1": {Name: "sda1"},
},
},
)
assert.True(t, ok)
assert.Equal(t, "sda1", key)
assert.Equal(t, "/mnt/data", stats.Mountpoint)
assert.Equal(t, "archive", stats.Name)
assert.False(t, stats.Root)
})
t.Run("maps root partition to io device by prefix", func(t *testing.T) {
key, stats, ok := registerFilesystemStats(
map[string]*system.FsStats{},
"/dev/ada0p2",
"/",
true,
"",
fsRegistrationContext{
isWindows: false,
diskIoCounters: map[string]disk.IOCountersStat{
"ada0": {Name: "ada0", ReadBytes: 1000, WriteBytes: 1000},
},
},
)
assert.True(t, ok)
assert.Equal(t, "ada0", key)
assert.True(t, stats.Root)
assert.Equal(t, "/", stats.Mountpoint)
})
t.Run("uses filesystem setting as root fallback", func(t *testing.T) {
key, _, ok := registerFilesystemStats(
map[string]*system.FsStats{},
"overlay",
"/",
true,
"",
fsRegistrationContext{
filesystem: "nvme0n1p2",
isWindows: false,
diskIoCounters: map[string]disk.IOCountersStat{
"nvme0n1": {Name: "nvme0n1", ReadBytes: 1000, WriteBytes: 1000},
},
},
)
assert.True(t, ok)
assert.Equal(t, "nvme0n1", key)
})
t.Run("prefers parsed extra-filesystems device over mapper device", func(t *testing.T) {
key, stats, ok := registerFilesystemStats(
map[string]*system.FsStats{},
"/dev/mapper/luks-2bcb02be-999d-4417-8d18-5c61e660fb6e",
"/extra-filesystems/nvme0n1p2__Archive",
false,
"Archive",
fsRegistrationContext{
isWindows: false,
diskIoCounters: map[string]disk.IOCountersStat{
"dm-1": {Name: "dm-1", Label: "luks-2bcb02be-999d-4417-8d18-5c61e660fb6e"},
"nvme0n1p2": {Name: "nvme0n1p2"},
},
},
)
assert.True(t, ok)
assert.Equal(t, "nvme0n1p2", key)
assert.Equal(t, "Archive", stats.Name)
})
t.Run("falls back to mapper io device when folder device cannot be resolved", func(t *testing.T) {
key, stats, ok := registerFilesystemStats(
map[string]*system.FsStats{},
"/dev/mapper/luks-2bcb02be-999d-4417-8d18-5c61e660fb6e",
"/extra-filesystems/Archive",
false,
"Archive",
fsRegistrationContext{
isWindows: false,
diskIoCounters: map[string]disk.IOCountersStat{
"dm-1": {Name: "dm-1", Label: "luks-2bcb02be-999d-4417-8d18-5c61e660fb6e"},
},
},
)
assert.True(t, ok)
assert.Equal(t, "dm-1", key)
assert.Equal(t, "Archive", stats.Name)
})
t.Run("uses full device name on windows", func(t *testing.T) {
key, _, ok := registerFilesystemStats(
map[string]*system.FsStats{},
`C:`,
`C:\\`,
false,
"",
fsRegistrationContext{
isWindows: true,
diskIoCounters: map[string]disk.IOCountersStat{
`C:`: {Name: `C:`},
},
},
)
assert.True(t, ok)
assert.Equal(t, `C:`, key)
})
t.Run("skips existing key", func(t *testing.T) {
key, stats, ok := registerFilesystemStats(
map[string]*system.FsStats{"sda1": {Mountpoint: "/existing"}},
"/dev/sda1",
"/mnt/data",
false,
"",
fsRegistrationContext{
isWindows: false,
diskIoCounters: map[string]disk.IOCountersStat{
"sda1": {Name: "sda1"},
},
},
)
assert.False(t, ok)
assert.Empty(t, key)
assert.Nil(t, stats)
})
}
func TestAddConfiguredRootFs(t *testing.T) {
t.Run("adds root from matching partition", func(t *testing.T) {
agent := &Agent{fsStats: make(map[string]*system.FsStats)}
discovery := diskDiscovery{
agent: agent,
rootMountPoint: "/",
partitions: []disk.PartitionStat{{Device: "/dev/ada0p2", Mountpoint: "/"}},
ctx: fsRegistrationContext{
filesystem: "/dev/ada0p2",
filesystemName: "root disk",
isWindows: false,
diskIoCounters: map[string]disk.IOCountersStat{
"ada0": {Name: "ada0", ReadBytes: 1000, WriteBytes: 1000},
},
},
}
ok := discovery.addConfiguredRootFs()
assert.True(t, ok)
stats, exists := agent.fsStats["ada0"]
assert.True(t, exists)
assert.True(t, stats.Root)
assert.Equal(t, "/", stats.Mountpoint)
assert.Equal(t, "root disk", stats.Name)
})
t.Run("adds root from io device when partition is missing", func(t *testing.T) {
agent := &Agent{fsStats: make(map[string]*system.FsStats)}
discovery := diskDiscovery{
agent: agent,
rootMountPoint: "/sysroot",
ctx: fsRegistrationContext{
filesystem: "zroot",
filesystemName: "root pool",
isWindows: false,
diskIoCounters: map[string]disk.IOCountersStat{
"nda0": {Name: "nda0", Label: "zroot", ReadBytes: 1000, WriteBytes: 1000},
},
},
}
ok := discovery.addConfiguredRootFs()
assert.True(t, ok)
stats, exists := agent.fsStats["nda0"]
assert.True(t, exists)
assert.True(t, stats.Root)
assert.Equal(t, "/sysroot", stats.Mountpoint)
assert.Equal(t, "root pool", stats.Name)
})
t.Run("returns false when filesystem cannot be resolved", func(t *testing.T) {
agent := &Agent{fsStats: make(map[string]*system.FsStats)}
discovery := diskDiscovery{
agent: agent,
rootMountPoint: "/",
ctx: fsRegistrationContext{
filesystem: "missing-disk",
isWindows: false,
diskIoCounters: map[string]disk.IOCountersStat{},
},
}
ok := discovery.addConfiguredRootFs()
assert.False(t, ok)
assert.Empty(t, agent.fsStats)
})
}
func TestAddPartitionRootFs(t *testing.T) {
t.Run("adds root from fallback partition candidate", func(t *testing.T) {
agent := &Agent{fsStats: make(map[string]*system.FsStats)}
discovery := diskDiscovery{
agent: agent,
ctx: fsRegistrationContext{
isWindows: false,
diskIoCounters: map[string]disk.IOCountersStat{
"nvme0n1": {Name: "nvme0n1", ReadBytes: 1000, WriteBytes: 1000},
},
},
}
ok := discovery.addPartitionRootFs("/dev/nvme0n1p2", "/")
assert.True(t, ok)
stats, exists := agent.fsStats["nvme0n1"]
assert.True(t, exists)
assert.True(t, stats.Root)
assert.Equal(t, "/", stats.Mountpoint)
})
t.Run("returns false when no io device matches", func(t *testing.T) {
agent := &Agent{fsStats: make(map[string]*system.FsStats)}
discovery := diskDiscovery{agent: agent, ctx: fsRegistrationContext{diskIoCounters: map[string]disk.IOCountersStat{}}}
ok := discovery.addPartitionRootFs("/dev/mapper/root", "/")
assert.False(t, ok)
assert.Empty(t, agent.fsStats)
})
}
func TestAddLastResortRootFs(t *testing.T) {
t.Run("uses most active io device when available", func(t *testing.T) {
agent := &Agent{fsStats: make(map[string]*system.FsStats)}
discovery := diskDiscovery{agent: agent, rootMountPoint: "/", ctx: fsRegistrationContext{diskIoCounters: map[string]disk.IOCountersStat{
"sda": {Name: "sda", ReadBytes: 5000, WriteBytes: 5000},
"sdb": {Name: "sdb", ReadBytes: 1000, WriteBytes: 1000},
}}}
discovery.addLastResortRootFs()
stats, exists := agent.fsStats["sda"]
assert.True(t, exists)
assert.True(t, stats.Root)
})
t.Run("falls back to root key when mountpoint basename collides", func(t *testing.T) {
agent := &Agent{fsStats: map[string]*system.FsStats{
"sysroot": {Mountpoint: "/extra-filesystems/sysroot"},
}}
discovery := diskDiscovery{agent: agent, rootMountPoint: "/sysroot", ctx: fsRegistrationContext{diskIoCounters: map[string]disk.IOCountersStat{}}}
discovery.addLastResortRootFs()
stats, exists := agent.fsStats["root"]
assert.True(t, exists)
assert.True(t, stats.Root)
assert.Equal(t, "/sysroot", stats.Mountpoint)
})
}
func TestAddConfiguredExtraFsEntry(t *testing.T) {
t.Run("uses matching partition when present", func(t *testing.T) {
agent := &Agent{fsStats: make(map[string]*system.FsStats)}
discovery := diskDiscovery{
agent: agent,
partitions: []disk.PartitionStat{{Device: "/dev/sdb1", Mountpoint: "/mnt/backup"}},
usageFn: func(string) (*disk.UsageStat, error) {
t.Fatal("usage fallback should not be called when partition matches")
return nil, nil
},
ctx: fsRegistrationContext{
isWindows: false,
diskIoCounters: map[string]disk.IOCountersStat{
"sdb1": {Name: "sdb1"},
},
},
}
discovery.addConfiguredExtraFsEntry("sdb1", "backup")
stats, exists := agent.fsStats["sdb1"]
assert.True(t, exists)
assert.Equal(t, "/mnt/backup", stats.Mountpoint)
assert.Equal(t, "backup", stats.Name)
})
t.Run("falls back to usage-validated path", func(t *testing.T) {
agent := &Agent{fsStats: make(map[string]*system.FsStats)}
discovery := diskDiscovery{
agent: agent,
usageFn: func(path string) (*disk.UsageStat, error) {
assert.Equal(t, "/srv/archive", path)
return &disk.UsageStat{}, nil
},
ctx: fsRegistrationContext{
isWindows: false,
diskIoCounters: map[string]disk.IOCountersStat{
"archive": {Name: "archive"},
},
},
}
discovery.addConfiguredExtraFsEntry("/srv/archive", "archive")
stats, exists := agent.fsStats["archive"]
assert.True(t, exists)
assert.Equal(t, "/srv/archive", stats.Mountpoint)
assert.Equal(t, "archive", stats.Name)
})
t.Run("ignores invalid filesystem entry", func(t *testing.T) {
agent := &Agent{fsStats: make(map[string]*system.FsStats)}
discovery := diskDiscovery{
agent: agent,
usageFn: func(string) (*disk.UsageStat, error) {
return nil, os.ErrNotExist
},
}
discovery.addConfiguredExtraFsEntry("/missing/archive", "")
assert.Empty(t, agent.fsStats)
})
}
func TestAddConfiguredExtraFilesystems(t *testing.T) {
t.Run("parses and registers multiple configured filesystems", func(t *testing.T) {
agent := &Agent{fsStats: make(map[string]*system.FsStats)}
discovery := diskDiscovery{
agent: agent,
partitions: []disk.PartitionStat{{Device: "/dev/sda1", Mountpoint: "/mnt/fast"}},
usageFn: func(path string) (*disk.UsageStat, error) {
if path == "/srv/archive" {
return &disk.UsageStat{}, nil
}
return nil, os.ErrNotExist
},
ctx: fsRegistrationContext{
isWindows: false,
diskIoCounters: map[string]disk.IOCountersStat{
"sda1": {Name: "sda1"},
"archive": {Name: "archive"},
},
},
}
discovery.addConfiguredExtraFilesystems("sda1__fast,/srv/archive__cold")
assert.Contains(t, agent.fsStats, "sda1")
assert.Equal(t, "fast", agent.fsStats["sda1"].Name)
assert.Contains(t, agent.fsStats, "archive")
assert.Equal(t, "cold", agent.fsStats["archive"].Name)
})
}
func TestAddExtraFilesystemFolders(t *testing.T) {
t.Run("adds missing folders and skips existing mountpoints", func(t *testing.T) {
agent := &Agent{fsStats: map[string]*system.FsStats{
"existing": {Mountpoint: "/extra-filesystems/existing"},
}}
discovery := diskDiscovery{
agent: agent,
ctx: fsRegistrationContext{
isWindows: false,
efPath: "/extra-filesystems",
diskIoCounters: map[string]disk.IOCountersStat{
"newdisk": {Name: "newdisk"},
},
},
}
discovery.addExtraFilesystemFolders([]string{"existing", "newdisk__Archive"})
assert.Len(t, agent.fsStats, 2)
stats, exists := agent.fsStats["newdisk"]
assert.True(t, exists)
assert.Equal(t, "/extra-filesystems/newdisk__Archive", stats.Mountpoint)
assert.Equal(t, "Archive", stats.Name)
})
}
func TestAddPartitionExtraFs(t *testing.T) {
makeDiscovery := func(agent *Agent) diskDiscovery {
return diskDiscovery{
agent: agent,
ctx: fsRegistrationContext{
isWindows: false,
efPath: "/extra-filesystems",
diskIoCounters: map[string]disk.IOCountersStat{
"nvme0n1p1": {Name: "nvme0n1p1"},
"nvme1n1": {Name: "nvme1n1"},
},
},
}
}
t.Run("registers direct child of extra-filesystems", func(t *testing.T) {
agent := &Agent{fsStats: make(map[string]*system.FsStats)}
d := makeDiscovery(agent)
d.addPartitionExtraFs(disk.PartitionStat{
Device: "/dev/nvme0n1p1",
Mountpoint: "/extra-filesystems/nvme0n1p1__caddy1-root",
})
stats, exists := agent.fsStats["nvme0n1p1"]
assert.True(t, exists)
assert.Equal(t, "/extra-filesystems/nvme0n1p1__caddy1-root", stats.Mountpoint)
assert.Equal(t, "caddy1-root", stats.Name)
})
t.Run("skips nested mount under extra-filesystem bind mount", func(t *testing.T) {
agent := &Agent{fsStats: make(map[string]*system.FsStats)}
d := makeDiscovery(agent)
// These simulate the virtual mounts that appear when host / is bind-mounted
// with disk.Partitions(all=true) — e.g. /proc, /sys, /dev visible under the mount.
for _, nested := range []string{
"/extra-filesystems/nvme0n1p1__caddy1-root/proc",
"/extra-filesystems/nvme0n1p1__caddy1-root/sys",
"/extra-filesystems/nvme0n1p1__caddy1-root/dev",
"/extra-filesystems/nvme0n1p1__caddy1-root/run",
} {
d.addPartitionExtraFs(disk.PartitionStat{Device: "tmpfs", Mountpoint: nested})
}
assert.Empty(t, agent.fsStats)
})
t.Run("registers both direct children, skips their nested mounts", func(t *testing.T) {
agent := &Agent{fsStats: make(map[string]*system.FsStats)}
d := makeDiscovery(agent)
partitions := []disk.PartitionStat{
{Device: "/dev/nvme0n1p1", Mountpoint: "/extra-filesystems/nvme0n1p1__caddy1-root"},
{Device: "/dev/nvme1n1", Mountpoint: "/extra-filesystems/nvme1n1__caddy1-docker"},
{Device: "proc", Mountpoint: "/extra-filesystems/nvme0n1p1__caddy1-root/proc"},
{Device: "sysfs", Mountpoint: "/extra-filesystems/nvme0n1p1__caddy1-root/sys"},
{Device: "overlay", Mountpoint: "/extra-filesystems/nvme0n1p1__caddy1-root/var/lib/docker"},
}
for _, p := range partitions {
d.addPartitionExtraFs(p)
}
assert.Len(t, agent.fsStats, 2)
assert.Equal(t, "caddy1-root", agent.fsStats["nvme0n1p1"].Name)
assert.Equal(t, "caddy1-docker", agent.fsStats["nvme1n1"].Name)
})
t.Run("skips partition not under extra-filesystems", func(t *testing.T) {
agent := &Agent{fsStats: make(map[string]*system.FsStats)}
d := makeDiscovery(agent)
d.addPartitionExtraFs(disk.PartitionStat{
Device: "/dev/nvme0n1p1",
Mountpoint: "/",
})
assert.Empty(t, agent.fsStats)
})
}
func TestFindIoDevice(t *testing.T) {
t.Run("matches by device name", func(t *testing.T) {
ioCounters := map[string]disk.IOCountersStat{
"sda": {Name: "sda"},
"sdb": {Name: "sdb"},
}
device, ok := findIoDevice("sdb", ioCounters)
assert.True(t, ok)
assert.Equal(t, "sdb", device)
})
t.Run("matches by device label", func(t *testing.T) {
ioCounters := map[string]disk.IOCountersStat{
"sda": {Name: "sda", Label: "rootfs"},
"sdb": {Name: "sdb"},
}
device, ok := findIoDevice("rootfs", ioCounters)
assert.True(t, ok)
assert.Equal(t, "sda", device)
})
t.Run("returns no match when not found", func(t *testing.T) {
ioCounters := map[string]disk.IOCountersStat{
"sda": {Name: "sda"},
"sdb": {Name: "sdb"},
}
device, ok := findIoDevice("nvme0n1p1", ioCounters)
assert.False(t, ok)
assert.Equal(t, "", device)
})
t.Run("uses uncertain unique prefix fallback", func(t *testing.T) {
ioCounters := map[string]disk.IOCountersStat{
"nvme0n1": {Name: "nvme0n1"},
"sda": {Name: "sda"},
}
device, ok := findIoDevice("nvme0n1p2", ioCounters)
assert.True(t, ok)
assert.Equal(t, "nvme0n1", device)
})
t.Run("uses dominant activity when prefix matches are ambiguous", func(t *testing.T) {
ioCounters := map[string]disk.IOCountersStat{
"sda": {Name: "sda", ReadBytes: 5000, WriteBytes: 5000, ReadCount: 100, WriteCount: 100},
"sdb": {Name: "sdb", ReadBytes: 1000, WriteBytes: 1000, ReadCount: 50, WriteCount: 50},
}
device, ok := findIoDevice("sd", ioCounters)
assert.True(t, ok)
assert.Equal(t, "sda", device)
})
t.Run("uses highest activity when ambiguous without dominance", func(t *testing.T) {
ioCounters := map[string]disk.IOCountersStat{
"sda": {Name: "sda", ReadBytes: 3000, WriteBytes: 3000, ReadCount: 50, WriteCount: 50},
"sdb": {Name: "sdb", ReadBytes: 2500, WriteBytes: 2500, ReadCount: 40, WriteCount: 40},
}
device, ok := findIoDevice("sd", ioCounters)
assert.True(t, ok)
assert.Equal(t, "sda", device)
})
t.Run("matches /dev/-prefixed partition to parent disk", func(t *testing.T) {
ioCounters := map[string]disk.IOCountersStat{
"nda0": {Name: "nda0", ReadBytes: 1000, WriteBytes: 1000},
}
device, ok := findIoDevice("/dev/nda0p2", ioCounters)
assert.True(t, ok)
assert.Equal(t, "nda0", device)
})
t.Run("uses deterministic name tie-breaker", func(t *testing.T) {
ioCounters := map[string]disk.IOCountersStat{
"sdb": {Name: "sdb", ReadBytes: 2000, WriteBytes: 2000, ReadCount: 10, WriteCount: 10},
"sda": {Name: "sda", ReadBytes: 2000, WriteBytes: 2000, ReadCount: 10, WriteCount: 10},
}
device, ok := findIoDevice("sd", ioCounters)
assert.True(t, ok)
assert.Equal(t, "sda", device)
})
}
func TestFilesystemMatchesPartitionSetting(t *testing.T) {
p := disk.PartitionStat{Device: "/dev/ada0p2", Mountpoint: "/"}
t.Run("matches mountpoint setting", func(t *testing.T) {
assert.True(t, filesystemMatchesPartitionSetting("/", p))
})
t.Run("matches exact partition setting", func(t *testing.T) {
assert.True(t, filesystemMatchesPartitionSetting("ada0p2", p))
assert.True(t, filesystemMatchesPartitionSetting("/dev/ada0p2", p))
})
t.Run("matches prefix-style parent setting", func(t *testing.T) {
assert.True(t, filesystemMatchesPartitionSetting("ada0", p))
assert.True(t, filesystemMatchesPartitionSetting("/dev/ada0", p))
})
t.Run("does not match unrelated device", func(t *testing.T) {
assert.False(t, filesystemMatchesPartitionSetting("sda", p))
assert.False(t, filesystemMatchesPartitionSetting("nvme0n1", p))
assert.False(t, filesystemMatchesPartitionSetting("", p))
})
}
func TestMostActiveIoDevice(t *testing.T) {
t.Run("returns most active device", func(t *testing.T) {
ioCounters := map[string]disk.IOCountersStat{
"nda0": {Name: "nda0", ReadBytes: 5000, WriteBytes: 5000, ReadCount: 100, WriteCount: 100},
"nda1": {Name: "nda1", ReadBytes: 1000, WriteBytes: 1000, ReadCount: 50, WriteCount: 50},
}
assert.Equal(t, "nda0", mostActiveIoDevice(ioCounters))
})
t.Run("uses deterministic tie-breaker", func(t *testing.T) {
ioCounters := map[string]disk.IOCountersStat{
"sdb": {Name: "sdb", ReadBytes: 1000, WriteBytes: 1000, ReadCount: 10, WriteCount: 10},
"sda": {Name: "sda", ReadBytes: 1000, WriteBytes: 1000, ReadCount: 10, WriteCount: 10},
}
assert.Equal(t, "sda", mostActiveIoDevice(ioCounters))
})
t.Run("returns empty for empty map", func(t *testing.T) {
assert.Equal(t, "", mostActiveIoDevice(map[string]disk.IOCountersStat{}))
})
}
func TestIsDockerSpecialMountpoint(t *testing.T) {
testCases := []struct {
name string
mountpoint string
expected bool
}{
{name: "hosts", mountpoint: "/etc/hosts", expected: true},
{name: "resolv", mountpoint: "/etc/resolv.conf", expected: true},
{name: "hostname", mountpoint: "/etc/hostname", expected: true},
{name: "root", mountpoint: "/", expected: false},
{name: "passwd", mountpoint: "/etc/passwd", expected: false},
{name: "extra-filesystem", mountpoint: "/extra-filesystems/sda1", expected: false},
}
for _, tc := range testCases {
t.Run(tc.name, func(t *testing.T) {
assert.Equal(t, tc.expected, isDockerSpecialMountpoint(tc.mountpoint))
})
}
}
func TestInitializeDiskInfoWithCustomNames(t *testing.T) { func TestInitializeDiskInfoWithCustomNames(t *testing.T) {
// Set up environment variables
oldEnv := os.Getenv("EXTRA_FILESYSTEMS")
defer func() {
if oldEnv != "" {
os.Setenv("EXTRA_FILESYSTEMS", oldEnv)
} else {
os.Unsetenv("EXTRA_FILESYSTEMS")
}
}()
// Test with custom names // Test with custom names
t.Setenv("EXTRA_FILESYSTEMS", "sda1__my-storage,/dev/sdb1__backup-drive,nvme0n1p2") os.Setenv("EXTRA_FILESYSTEMS", "sda1__my-storage,/dev/sdb1__backup-drive,nvme0n1p2")
// Mock disk partitions (we'll just test the parsing logic) // Mock disk partitions (we'll just test the parsing logic)
// Since the actual disk operations are system-dependent, we'll focus on the parsing // Since the actual disk operations are system-dependent, we'll focus on the parsing
@@ -794,7 +134,7 @@ func TestInitializeDiskInfoWithCustomNames(t *testing.T) {
for _, tc := range testCases { for _, tc := range testCases {
t.Run("env_"+tc.envValue, func(t *testing.T) { t.Run("env_"+tc.envValue, func(t *testing.T) {
t.Setenv("EXTRA_FILESYSTEMS", tc.envValue) os.Setenv("EXTRA_FILESYSTEMS", tc.envValue)
// Create mock partitions that would match our test cases // Create mock partitions that would match our test cases
partitions := []disk.PartitionStat{} partitions := []disk.PartitionStat{}
@@ -815,7 +155,7 @@ func TestInitializeDiskInfoWithCustomNames(t *testing.T) {
// Test the parsing logic by calling the relevant part // Test the parsing logic by calling the relevant part
// We'll create a simplified version to test just the parsing // We'll create a simplified version to test just the parsing
extraFilesystems := tc.envValue extraFilesystems := tc.envValue
for fsEntry := range strings.SplitSeq(extraFilesystems, ",") { for _, fsEntry := range strings.Split(extraFilesystems, ",") {
// Parse the entry // Parse the entry
fsEntry = strings.TrimSpace(fsEntry) fsEntry = strings.TrimSpace(fsEntry)
var fs, customName string var fs, customName string
@@ -977,67 +317,3 @@ func TestDiskUsageCaching(t *testing.T) {
"lastDiskUsageUpdate should be refreshed when cache expires") "lastDiskUsageUpdate should be refreshed when cache expires")
}) })
} }
func TestHasSameDiskUsage(t *testing.T) {
const toleranceBytes uint64 = 16 * 1024 * 1024
t.Run("returns true when totals and usage are equal", func(t *testing.T) {
a := &disk.UsageStat{Total: 100 * 1024 * 1024 * 1024, Used: 42 * 1024 * 1024 * 1024}
b := &disk.UsageStat{Total: 100 * 1024 * 1024 * 1024, Used: 42 * 1024 * 1024 * 1024}
assert.True(t, hasSameDiskUsage(a, b))
})
t.Run("returns true within tolerance", func(t *testing.T) {
a := &disk.UsageStat{Total: 100 * 1024 * 1024 * 1024, Used: 42 * 1024 * 1024 * 1024}
b := &disk.UsageStat{
Total: a.Total + toleranceBytes - 1,
Used: a.Used - toleranceBytes + 1,
}
assert.True(t, hasSameDiskUsage(a, b))
})
t.Run("returns false when total exceeds tolerance", func(t *testing.T) {
a := &disk.UsageStat{Total: 100 * 1024 * 1024 * 1024, Used: 42 * 1024 * 1024 * 1024}
b := &disk.UsageStat{
Total: a.Total + toleranceBytes + 1,
Used: a.Used,
}
assert.False(t, hasSameDiskUsage(a, b))
})
t.Run("returns false for nil or zero total", func(t *testing.T) {
assert.False(t, hasSameDiskUsage(nil, &disk.UsageStat{Total: 1, Used: 1}))
assert.False(t, hasSameDiskUsage(&disk.UsageStat{Total: 1, Used: 1}, nil))
assert.False(t, hasSameDiskUsage(&disk.UsageStat{Total: 0, Used: 0}, &disk.UsageStat{Total: 1, Used: 1}))
})
}
func TestInitializeDiskIoStatsResetsTrackedDevices(t *testing.T) {
agent := &Agent{
fsStats: map[string]*system.FsStats{
"sda": {},
"sdb": {},
},
fsNames: []string{"stale", "sda"},
}
agent.initializeDiskIoStats(map[string]disk.IOCountersStat{
"sda": {Name: "sda", ReadBytes: 10, WriteBytes: 20},
"sdb": {Name: "sdb", ReadBytes: 30, WriteBytes: 40},
})
assert.ElementsMatch(t, []string{"sda", "sdb"}, agent.fsNames)
assert.Len(t, agent.fsNames, 2)
assert.Equal(t, uint64(10), agent.fsStats["sda"].TotalRead)
assert.Equal(t, uint64(20), agent.fsStats["sda"].TotalWrite)
assert.False(t, agent.fsStats["sda"].Time.IsZero())
assert.False(t, agent.fsStats["sdb"].Time.IsZero())
agent.initializeDiskIoStats(map[string]disk.IOCountersStat{
"sdb": {Name: "sdb", ReadBytes: 50, WriteBytes: 60},
})
assert.Equal(t, []string{"sdb"}, agent.fsNames)
assert.Equal(t, uint64(50), agent.fsStats["sdb"].TotalRead)
assert.Equal(t, uint64(60), agent.fsStats["sdb"].TotalWrite)
}
-110
View File
@@ -1,110 +0,0 @@
//go:build testing
package agent
import (
"testing"
"time"
"github.com/henrygd/beszel/agent/zfs"
"github.com/henrygd/beszel/internal/entities/system"
"github.com/shirou/gopsutil/v4/disk"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
// TestUpdateDiskUsageZfsMountpoint verifies that a filesystem whose mountpoint
// is a ZFS dataset reports `zfs list` usage (which includes child datasets)
// instead of the dataset-scoped statfs values (#1541).
func TestUpdateDiskUsageZfsMountpoint(t *testing.T) {
zm := &StoragePoolManager{detailInterval: time.Hour, backends: []*poolBackend{{name: "zfs"}}}
zm.backends[0].datasetsFn = func() ([]zfs.Dataset, error) {
return []zfs.Dataset{
{Name: "tank", Used: 12000000000000, Avail: 11999000000000, Mountpoint: "/tank"},
}, nil
}
agent := &Agent{
fsStats: map[string]*system.FsStats{
"tank": {Root: false, Mountpoint: "/tank"},
},
storagePoolManager: zm,
}
var stats system.Stats
agent.updateDiskUsage(&stats)
fs := agent.fsStats["tank"]
require.NotNil(t, fs)
assert.Equal(t, 22350.81, fs.DiskTotal) // (used + avail) in GiB
assert.Equal(t, 11175.87, fs.DiskUsed)
// Non-root filesystems do not populate system-level stats.
assert.Equal(t, float64(0), stats.DiskTotal)
}
// TestUpdateDiskUsageZfsRootPopulatesSystemStats verifies the root disk values
// are derived from ZFS usage when the root mountpoint is a ZFS dataset.
func TestUpdateDiskUsageZfsRootPopulatesSystemStats(t *testing.T) {
zm := &StoragePoolManager{detailInterval: time.Hour, backends: []*poolBackend{{name: "zfs"}}}
zm.backends[0].datasetsFn = func() ([]zfs.Dataset, error) {
return []zfs.Dataset{
{Name: "rpool/ROOT/pve-1", Used: 900000000000, Avail: 300000000000, Mountpoint: "/"},
}, nil
}
agent := &Agent{
fsStats: map[string]*system.FsStats{
"rpool/ROOT/pve-1": {Root: true, Mountpoint: "/"},
},
storagePoolManager: zm,
}
var stats system.Stats
agent.updateDiskUsage(&stats)
assert.Equal(t, 1117.59, agent.fsStats["rpool/ROOT/pve-1"].DiskTotal)
assert.Equal(t, 838.19, agent.fsStats["rpool/ROOT/pve-1"].DiskUsed)
assert.Equal(t, 75.0, stats.DiskPct)
assert.Equal(t, 1117.59, stats.DiskTotal)
assert.Equal(t, 838.19, stats.DiskUsed)
}
// TestUpdateDiskUsageWithoutZfsManager falls back to statfs when no manager is
// present (e.g. tests constructing bare Agent values).
func TestUpdateDiskUsageWithoutZfsManager(t *testing.T) {
agent := &Agent{
fsStats: map[string]*system.FsStats{
"root": {Root: true, Mountpoint: "/"},
},
}
var stats system.Stats
agent.updateDiskUsage(&stats)
assert.True(t, agent.fsStats["root"].DiskTotal > 0, "root usage should come from statfs")
assert.True(t, stats.DiskTotal > 0)
}
// TestInitializeDiskIoStatsSkipsZfsMountpoints verifies ZFS filesystems are
// excluded from diskstats I/O tracking instead of warning about a missing device.
func TestInitializeDiskIoStatsSkipsZfsMountpoints(t *testing.T) {
zm := &StoragePoolManager{detailInterval: time.Hour, backends: []*poolBackend{{name: "zfs"}}}
zm.backends[0].datasetsFn = func() ([]zfs.Dataset, error) {
return []zfs.Dataset{{Name: "tank", Mountpoint: "/tank"}}, nil
}
agent := &Agent{
fsStats: map[string]*system.FsStats{
"tank": {Root: false, Mountpoint: "/tank"},
"sda1": {Root: false, Mountpoint: "/mnt/data"},
},
storagePoolManager: zm,
diskPrev: make(map[uint16]map[string]prevDisk),
}
agent.initializeDiskIoStats(map[string]disk.IOCountersStat{
"sda1": {Name: "sda1", ReadBytes: 100, WriteBytes: 100},
})
assert.Equal(t, []string{"sda1"}, agent.fsNames)
assert.Equal(t, uint64(100), agent.fsStats["sda1"].TotalRead)
// ZFS entry is present but untouched by diskstats initialization.
assert.Equal(t, uint64(0), agent.fsStats["tank"].TotalRead)
}
+83 -350
View File
@@ -1,7 +1,6 @@
package agent package agent
import ( import (
"bufio"
"bytes" "bytes"
"context" "context"
"encoding/binary" "encoding/binary"
@@ -16,16 +15,12 @@ import (
"os" "os"
"path" "path"
"regexp" "regexp"
"sort"
"strconv"
"strings" "strings"
"sync" "sync"
"time" "time"
"github.com/henrygd/beszel/agent/deltatracker" "github.com/henrygd/beszel/agent/deltatracker"
"github.com/henrygd/beszel/agent/utils"
"github.com/henrygd/beszel/internal/entities/container" "github.com/henrygd/beszel/internal/entities/container"
"github.com/henrygd/beszel/internal/entities/system"
"github.com/blang/semver" "github.com/blang/semver"
) )
@@ -33,7 +28,6 @@ import (
// ansiEscapePattern matches ANSI escape sequences (colors, cursor movement, etc.) // ansiEscapePattern matches ANSI escape sequences (colors, cursor movement, etc.)
// This includes CSI sequences like \x1b[...m and simple escapes like \x1b[K // This includes CSI sequences like \x1b[...m and simple escapes like \x1b[K
var ansiEscapePattern = regexp.MustCompile(`\x1b\[[0-9;]*[a-zA-Z]|\x1b\][^\x07]*\x07|\x1b[@-Z\\-_]`) var ansiEscapePattern = regexp.MustCompile(`\x1b\[[0-9;]*[a-zA-Z]|\x1b\][^\x07]*\x07|\x1b[@-Z\\-_]`)
var dockerContainerIDPattern = regexp.MustCompile(`^[a-fA-F0-9]{12,64}$`)
const ( const (
// Docker API timeout in milliseconds // Docker API timeout in milliseconds
@@ -53,25 +47,20 @@ const (
) )
type dockerManager struct { type dockerManager struct {
agent *Agent // Used to propagate system detail changes back to the agent client *http.Client // Client to query Docker API
client *http.Client // Client to query Docker API wg sync.WaitGroup // WaitGroup to wait for all goroutines to finish
wg sync.WaitGroup // WaitGroup to wait for all goroutines to finish sem chan struct{} // Semaphore to limit concurrent container requests
sem chan struct{} // Semaphore to limit concurrent container requests containerStatsMutex sync.RWMutex // Mutex to prevent concurrent access to containerStatsMap
containerStatsMutex sync.RWMutex // Mutex to prevent concurrent access to containerStatsMap apiContainerList []*container.ApiInfo // List of containers from Docker API
apiContainerList []*container.ApiInfo // List of containers from Docker API containerStatsMap map[string]*container.Stats // Keeps track of container stats
containerStatsMap map[string]*container.Stats // Keeps track of container stats validIds map[string]struct{} // Map of valid container ids, used to prune invalid containers from containerStatsMap
validIds map[string]struct{} // Map of valid container ids, used to prune invalid containers from containerStatsMap goodDockerVersion bool // Whether docker version is at least 25.0.0 (one-shot works correctly)
goodDockerVersion bool // Whether docker version is at least 25.0.0 (one-shot works correctly) isWindows bool // Whether the Docker Engine API is running on Windows
dockerVersionChecked bool // Whether a version probe has completed successfully buf *bytes.Buffer // Buffer to store and read response bodies
isWindows bool // Whether the Docker Engine API is running on Windows decoder *json.Decoder // Reusable JSON decoder that reads from buf
buf *bytes.Buffer // Buffer to store and read response bodies apiStats *container.ApiStats // Reusable API stats object
excludeContainers []string // Patterns to exclude containers by name excludeContainers []string // Patterns to exclude containers by name
usingPodman bool // Whether the Docker Engine API is running on Podman usingPodman bool // Whether the Docker Engine API is running on Podman
registryClient *http.Client // Client for registry requests; nil uses a client with a 10-second timeout
imageUpdatesMutex sync.RWMutex // Protects imageUpdates, its entries, and imageUpdatesRunning
imageUpdates map[string]*imageUpdateStatus // Shared update status keyed by normalized image reference
imageUpdatesRunning bool // Whether a background image-update batch is in progress
// Cache-time-aware tracking for CPU stats (similar to cpu.go) // Cache-time-aware tracking for CPU stats (similar to cpu.go)
// Maps cache time intervals to container-specific CPU usage tracking // Maps cache time intervals to container-specific CPU usage tracking
@@ -83,7 +72,6 @@ type dockerManager struct {
// cacheTimeMs -> DeltaTracker for network bytes sent/received // cacheTimeMs -> DeltaTracker for network bytes sent/received
networkSentTrackers map[uint16]*deltatracker.DeltaTracker[string, uint64] networkSentTrackers map[uint16]*deltatracker.DeltaTracker[string, uint64]
networkRecvTrackers map[uint16]*deltatracker.DeltaTracker[string, uint64] networkRecvTrackers map[uint16]*deltatracker.DeltaTracker[string, uint64]
lastNetworkReadTime map[uint16]map[string]time.Time // cacheTimeMs -> containerId -> last network read time
} }
// userAgentRoundTripper is a custom http.RoundTripper that adds a User-Agent header to all requests // userAgentRoundTripper is a custom http.RoundTripper that adds a User-Agent header to all requests
@@ -92,14 +80,6 @@ type userAgentRoundTripper struct {
userAgent string userAgent string
} }
// dockerVersionResponse contains the /version fields used for engine checks.
type dockerVersionResponse struct {
Version string `json:"Version"`
Components []struct {
Name string `json:"Name"`
} `json:"Components"`
}
// RoundTrip implements the http.RoundTripper interface // RoundTrip implements the http.RoundTripper interface
func (u *userAgentRoundTripper) RoundTrip(req *http.Request) (*http.Response, error) { func (u *userAgentRoundTripper) RoundTrip(req *http.Request) (*http.Response, error) {
req.Header.Set("User-Agent", u.userAgent) req.Header.Set("User-Agent", u.userAgent)
@@ -147,14 +127,7 @@ func (dm *dockerManager) getDockerStats(cacheTimeMs uint16) ([]*container.Stats,
return nil, err return nil, err
} }
// Detect Podman and Windows from Server header dm.isWindows = strings.Contains(resp.Header.Get("Server"), "windows")
serverHeader := resp.Header.Get("Server")
if !dm.usingPodman && detectPodmanFromHeader(serverHeader) {
dm.setIsPodman()
}
dm.isWindows = strings.Contains(serverHeader, "windows")
dm.ensureDockerVersionChecked()
containersLength := len(dm.apiContainerList) containersLength := len(dm.apiContainerList)
@@ -165,9 +138,6 @@ func (dm *dockerManager) getDockerStats(cacheTimeMs uint16) ([]*container.Stats,
clear(dm.validIds) clear(dm.validIds)
} }
// Only schedule auxiliary work here; metrics never wait for image discovery.
dm.refreshImageUpdates(dm.apiContainerList, time.Now())
var failedContainers []*container.ApiInfo var failedContainers []*container.ApiInfo
for _, ctr := range dm.apiContainerList { for _, ctr := range dm.apiContainerList {
@@ -309,7 +279,7 @@ func (dm *dockerManager) cycleNetworkDeltasForCacheTime(cacheTimeMs uint16) {
} }
// calculateNetworkStats calculates network sent/receive deltas using DeltaTracker // calculateNetworkStats calculates network sent/receive deltas using DeltaTracker
func (dm *dockerManager) calculateNetworkStats(ctr *container.ApiInfo, apiStats *container.ApiStats, name string, cacheTimeMs uint16) (uint64, uint64) { func (dm *dockerManager) calculateNetworkStats(ctr *container.ApiInfo, apiStats *container.ApiStats, stats *container.Stats, initialized bool, name string, cacheTimeMs uint16) (uint64, uint64) {
var total_sent, total_recv uint64 var total_sent, total_recv uint64
for _, v := range apiStats.Networks { for _, v := range apiStats.Networks {
total_sent += v.TxBytes total_sent += v.TxBytes
@@ -328,11 +298,10 @@ func (dm *dockerManager) calculateNetworkStats(ctr *container.ApiInfo, apiStats
sent_delta_raw := sentTracker.Delta(ctr.IdShort) sent_delta_raw := sentTracker.Delta(ctr.IdShort)
recv_delta_raw := recvTracker.Delta(ctr.IdShort) recv_delta_raw := recvTracker.Delta(ctr.IdShort)
// Calculate bytes per second using per-cache-time read time to avoid // Calculate bytes per second independently for Tx and Rx if we have previous data
// interference between different cache intervals (e.g. 1000ms vs 60000ms)
var sent_delta, recv_delta uint64 var sent_delta, recv_delta uint64
if prevReadTime, ok := dm.lastNetworkReadTime[cacheTimeMs][ctr.IdShort]; ok { if initialized {
millisecondsElapsed := uint64(time.Since(prevReadTime).Milliseconds()) millisecondsElapsed := uint64(time.Since(stats.PrevReadTime).Milliseconds())
if millisecondsElapsed > 0 { if millisecondsElapsed > 0 {
if sent_delta_raw > 0 { if sent_delta_raw > 0 {
sent_delta = sent_delta_raw * 1000 / millisecondsElapsed sent_delta = sent_delta_raw * 1000 / millisecondsElapsed
@@ -364,58 +333,13 @@ func validateCpuPercentage(cpuPct float64, containerName string) error {
// updateContainerStatsValues updates the final stats values // updateContainerStatsValues updates the final stats values
func updateContainerStatsValues(stats *container.Stats, cpuPct float64, usedMemory uint64, sent_delta, recv_delta uint64, readTime time.Time) { func updateContainerStatsValues(stats *container.Stats, cpuPct float64, usedMemory uint64, sent_delta, recv_delta uint64, readTime time.Time) {
stats.Cpu = utils.TwoDecimals(cpuPct) stats.Cpu = twoDecimals(cpuPct)
stats.Mem = utils.BytesToMegabytes(float64(usedMemory)) stats.Mem = bytesToMegabytes(float64(usedMemory))
stats.Bandwidth = [2]uint64{sent_delta, recv_delta} stats.NetworkSent = bytesToMegabytes(float64(sent_delta))
// TODO(0.19+): stop populating NetworkSent/NetworkRecv (deprecated in 0.18.3) stats.NetworkRecv = bytesToMegabytes(float64(recv_delta))
stats.NetworkSent = utils.BytesToMegabytes(float64(sent_delta))
stats.NetworkRecv = utils.BytesToMegabytes(float64(recv_delta))
stats.PrevReadTime = readTime stats.PrevReadTime = readTime
} }
// convertContainerPortsToString formats the ports of a container into a sorted, deduplicated string.
// ctr.Ports is nilled out after processing so the slice is not accidentally reused.
func convertContainerPortsToString(ctr *container.ApiInfo) string {
if len(ctr.Ports) == 0 {
return ""
}
sort.Slice(ctr.Ports, func(i, j int) bool {
if ctr.Ports[i].PublicPort != ctr.Ports[j].PublicPort {
return ctr.Ports[i].PublicPort < ctr.Ports[j].PublicPort
}
return ctr.Ports[i].IP < ctr.Ports[j].IP
})
var builder strings.Builder
seen := make(map[string]struct{})
for _, p := range ctr.Ports {
if p.PublicPort == 0 {
continue
}
keyIP := p.IP
if keyIP == "0.0.0.0" || keyIP == "::" {
keyIP = ""
}
key := keyIP + ":" + strconv.Itoa(int(p.PublicPort))
if _, ok := seen[key]; ok {
continue
}
seen[key] = struct{}{}
if builder.Len() > 0 {
builder.WriteString(", ")
}
switch p.IP {
case "0.0.0.0", "::":
default:
builder.WriteString(p.IP)
builder.WriteByte(':')
}
builder.WriteString(strconv.Itoa(int(p.PublicPort)))
}
// clear ports slice so it doesn't get reused and blend into next response
ctr.Ports = nil
return builder.String()
}
func parseDockerStatus(status string) (string, container.DockerHealth) { func parseDockerStatus(status string) (string, container.DockerHealth) {
trimmed := strings.TrimSpace(status) trimmed := strings.TrimSpace(status)
if trimmed == "" { if trimmed == "" {
@@ -435,60 +359,22 @@ func parseDockerStatus(status string) (string, container.DockerHealth) {
statusText = trimmed statusText = trimmed
} }
healthText := strings.TrimSpace(strings.TrimSuffix(trimmed[openIdx+1:], ")")) healthText := strings.ToLower(strings.TrimSpace(strings.TrimSuffix(trimmed[openIdx+1:], ")")))
// Some Docker statuses include a "health:" prefix inside the parentheses. // Some Docker statuses include a "health:" prefix inside the parentheses.
// Strip it so it maps correctly to the known health states. // Strip it so it maps correctly to the known health states.
if colonIdx := strings.IndexRune(healthText, ':'); colonIdx != -1 { if colonIdx := strings.IndexRune(healthText, ':'); colonIdx != -1 {
prefix := strings.ToLower(strings.TrimSpace(healthText[:colonIdx])) prefix := strings.TrimSpace(healthText[:colonIdx])
if prefix == "health" || prefix == "health status" { if prefix == "health" || prefix == "health status" {
healthText = strings.TrimSpace(healthText[colonIdx+1:]) healthText = strings.TrimSpace(healthText[colonIdx+1:])
} }
} }
if health, ok := parseDockerHealthStatus(healthText); ok { if health, ok := container.DockerHealthStrings[healthText]; ok {
return statusText, health return statusText, health
} }
return trimmed, container.DockerHealthNone return trimmed, container.DockerHealthNone
} }
// parseDockerHealthStatus maps Docker health status strings to container.DockerHealth values
func parseDockerHealthStatus(status string) (container.DockerHealth, bool) {
health, ok := container.DockerHealthStrings[strings.ToLower(strings.TrimSpace(status))]
return health, ok
}
// getPodmanContainerHealth fetches container health status from the container inspect endpoint.
// Used for Podman which doesn't provide health status in the /containers/json endpoint as of March 2026.
// https://github.com/containers/podman/issues/27786
func (dm *dockerManager) getPodmanContainerHealth(containerID string) (container.DockerHealth, error) {
resp, err := dm.client.Get(fmt.Sprintf("http://localhost/containers/%s/json", url.PathEscape(containerID)))
if err != nil {
return container.DockerHealthNone, err
}
defer resp.Body.Close()
if resp.StatusCode != http.StatusOK {
return container.DockerHealthNone, fmt.Errorf("container inspect request failed: %s", resp.Status)
}
var inspectInfo struct {
State struct {
Health struct {
Status string
}
}
}
if err := json.NewDecoder(resp.Body).Decode(&inspectInfo); err != nil {
return container.DockerHealthNone, err
}
if health, ok := parseDockerHealthStatus(inspectInfo.State.Health.Status); ok {
return health, nil
}
return container.DockerHealthNone, nil
}
// Updates stats for individual container with cache-time-aware delta tracking // Updates stats for individual container with cache-time-aware delta tracking
func (dm *dockerManager) updateContainerStats(ctr *container.ApiInfo, cacheTimeMs uint16) error { func (dm *dockerManager) updateContainerStats(ctr *container.ApiInfo, cacheTimeMs uint16) error {
name := ctr.Names[0][1:] name := ctr.Names[0][1:]
@@ -498,32 +384,6 @@ func (dm *dockerManager) updateContainerStats(ctr *container.ApiInfo, cacheTimeM
return err return err
} }
statusText, health := parseDockerStatus(ctr.Status)
// Docker exposes Health.Status on /containers/json in API 1.52+.
// Podman currently requires falling back to the inspect endpoint as of March 2026.
// https://github.com/containers/podman/issues/27786
if ctr.Health.Status != "" {
if h, ok := parseDockerHealthStatus(ctr.Health.Status); ok {
health = h
}
} else if dm.usingPodman {
if podmanHealth, err := dm.getPodmanContainerHealth(ctr.IdShort); err == nil {
health = podmanHealth
}
}
// Read and decode the response before locking shared stats to avoid blocking
defer resp.Body.Close()
if resp.StatusCode != http.StatusOK {
return fmt.Errorf("container stats request failed: %s", resp.Status)
}
res := &container.ApiStats{}
if err := json.NewDecoder(resp.Body).Decode(res); err != nil {
return err
}
updateAvailable := dm.cachedImageUpdate(ctr.Image)
dm.containerStatsMutex.Lock() dm.containerStatsMutex.Lock()
defer dm.containerStatsMutex.Unlock() defer dm.containerStatsMutex.Unlock()
@@ -535,42 +395,34 @@ func (dm *dockerManager) updateContainerStats(ctr *container.ApiInfo, cacheTimeM
} }
stats.Id = ctr.IdShort stats.Id = ctr.IdShort
statusText, health := parseDockerStatus(ctr.Status)
stats.Status = statusText stats.Status = statusText
stats.Health = health stats.Health = health
stats.Image = ctr.Image
stats.UpdateAvailable = updateAvailable
if len(ctr.Ports) > 0 {
stats.Ports = convertContainerPortsToString(ctr)
}
// reset current stats // reset current stats
stats.Cpu = 0 stats.Cpu = 0
stats.Mem = 0 stats.Mem = 0
stats.Bandwidth = [2]uint64{0, 0}
// TODO(0.19+): stop populating NetworkSent/NetworkRecv (deprecated in 0.18.3)
stats.NetworkSent = 0 stats.NetworkSent = 0
stats.NetworkRecv = 0 stats.NetworkRecv = 0
res := dm.apiStats
res.Networks = nil
if err := dm.decode(resp, res); err != nil {
return err
}
// Initialize CPU tracking for this cache time interval // Initialize CPU tracking for this cache time interval
dm.initializeCpuTracking(cacheTimeMs) dm.initializeCpuTracking(cacheTimeMs)
// Get previous CPU values // Get previous CPU values
prevCpuContainer, prevCpuSystem := dm.getCpuPreviousValues(cacheTimeMs, ctr.IdShort) prevCpuContainer, prevCpuSystem := dm.getCpuPreviousValues(cacheTimeMs, ctr.IdShort)
// Calculate CPU percentage based on platform. // Calculate CPU percentage based on platform
// Podman reports system_cpu_usage from cgroup cpu.stat (not /proc/stat), so it reflects
// only cgroup-tracked activity rather than total host capacity. Use a time-based method
// instead so the result is comparable to host CPU utilization. See:
// https://github.com/henrygd/beszel/issues/2049
var cpuPct float64 var cpuPct float64
if dm.isWindows { if dm.isWindows {
prevRead := dm.lastCpuReadTime[cacheTimeMs][ctr.IdShort] prevRead := dm.lastCpuReadTime[cacheTimeMs][ctr.IdShort]
cpuPct = res.CalculateCpuPercentWindows(prevCpuContainer, prevRead) cpuPct = res.CalculateCpuPercentWindows(prevCpuContainer, prevRead)
} else if dm.usingPodman && res.CPUStats.OnlineCPUs > 0 {
prevRead := dm.lastCpuReadTime[cacheTimeMs][ctr.IdShort]
cpuPct = res.CalculateCpuPercentPodman(prevCpuContainer, prevRead)
} else { } else {
cpuPct = res.CalculateCpuPercentLinux(prevCpuContainer, prevCpuSystem) cpuPct = res.CalculateCpuPercentLinux(prevCpuContainer, prevCpuSystem)
} }
@@ -592,13 +444,7 @@ func (dm *dockerManager) updateContainerStats(ctr *container.ApiInfo, cacheTimeM
} }
// Calculate network stats using DeltaTracker // Calculate network stats using DeltaTracker
sent_delta, recv_delta := dm.calculateNetworkStats(ctr, res, name, cacheTimeMs) sent_delta, recv_delta := dm.calculateNetworkStats(ctr, res, stats, initialized, name, cacheTimeMs)
// Store per-cache-time network read time for next rate calculation
if dm.lastNetworkReadTime[cacheTimeMs] == nil {
dm.lastNetworkReadTime[cacheTimeMs] = make(map[string]time.Time)
}
dm.lastNetworkReadTime[cacheTimeMs][ctr.IdShort] = time.Now()
// Store current network values for legacy compatibility // Store current network values for legacy compatibility
var total_sent, total_recv uint64 var total_sent, total_recv uint64
@@ -630,14 +476,11 @@ func (dm *dockerManager) deleteContainerStatsSync(id string) {
for ct := range dm.lastCpuReadTime { for ct := range dm.lastCpuReadTime {
delete(dm.lastCpuReadTime[ct], id) delete(dm.lastCpuReadTime[ct], id)
} }
for ct := range dm.lastNetworkReadTime {
delete(dm.lastNetworkReadTime[ct], id)
}
} }
// Creates a new http client for Docker or Podman API // Creates a new http client for Docker or Podman API
func newDockerManager(agent *Agent) *dockerManager { func newDockerManager() *dockerManager {
dockerHost, exists := utils.GetEnv("DOCKER_HOST") dockerHost, exists := GetEnv("DOCKER_HOST")
if exists { if exists {
// return nil if set to empty string // return nil if set to empty string
if dockerHost == "" { if dockerHost == "" {
@@ -673,7 +516,7 @@ func newDockerManager(agent *Agent) *dockerManager {
// configurable timeout // configurable timeout
timeout := time.Millisecond * time.Duration(dockerTimeoutMs) timeout := time.Millisecond * time.Duration(dockerTimeoutMs)
if t, set := utils.GetEnv("DOCKER_TIMEOUT"); set { if t, set := GetEnv("DOCKER_TIMEOUT"); set {
timeout, err = time.ParseDuration(t) timeout, err = time.ParseDuration(t)
if err != nil { if err != nil {
slog.Error(err.Error()) slog.Error(err.Error())
@@ -690,7 +533,7 @@ func newDockerManager(agent *Agent) *dockerManager {
// Read container exclusion patterns from environment variable // Read container exclusion patterns from environment variable
var excludeContainers []string var excludeContainers []string
if excludeStr, set := utils.GetEnv("EXCLUDE_CONTAINERS"); set && excludeStr != "" { if excludeStr, set := GetEnv("EXCLUDE_CONTAINERS"); set && excludeStr != "" {
parts := strings.SplitSeq(excludeStr, ",") parts := strings.SplitSeq(excludeStr, ",")
for part := range parts { for part := range parts {
trimmed := strings.TrimSpace(part) trimmed := strings.TrimSpace(part)
@@ -702,7 +545,6 @@ func newDockerManager(agent *Agent) *dockerManager {
} }
manager := &dockerManager{ manager := &dockerManager{
agent: agent,
client: &http.Client{ client: &http.Client{
Timeout: timeout, Timeout: timeout,
Transport: userAgentTransport, Transport: userAgentTransport,
@@ -710,6 +552,7 @@ func newDockerManager(agent *Agent) *dockerManager {
containerStatsMap: make(map[string]*container.Stats), containerStatsMap: make(map[string]*container.Stats),
sem: make(chan struct{}, 5), sem: make(chan struct{}, 5),
apiContainerList: []*container.ApiInfo{}, apiContainerList: []*container.ApiInfo{},
apiStats: &container.ApiStats{},
excludeContainers: excludeContainers, excludeContainers: excludeContainers,
// Initialize cache-time-aware tracking structures // Initialize cache-time-aware tracking structures
@@ -718,55 +561,50 @@ func newDockerManager(agent *Agent) *dockerManager {
lastCpuReadTime: make(map[uint16]map[string]time.Time), lastCpuReadTime: make(map[uint16]map[string]time.Time),
networkSentTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]), networkSentTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
networkRecvTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]), networkRecvTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
lastNetworkReadTime: make(map[uint16]map[string]time.Time),
} }
// Best-effort startup probe. If the engine is not ready yet, getDockerStats will // If using podman, return client
// retry after the first successful /containers/json request. if strings.Contains(dockerHost, "podman") {
_, _ = manager.checkDockerVersion() manager.usingPodman = true
manager.goodDockerVersion = true
return manager
}
// this can take up to 5 seconds with retry, so run in goroutine
go manager.checkDockerVersion()
// give version check a chance to complete before returning
time.Sleep(50 * time.Millisecond)
return manager return manager
} }
// checkDockerVersion checks Docker version and sets goodDockerVersion if at least 25.0.0. // checkDockerVersion checks Docker version and sets goodDockerVersion if at least 25.0.0.
// Versions before 25.0.0 have a bug with one-shot which requires all requests to be made in one batch. // Versions before 25.0.0 have a bug with one-shot which requires all requests to be made in one batch.
func (dm *dockerManager) checkDockerVersion() (bool, error) { func (dm *dockerManager) checkDockerVersion() {
resp, err := dm.client.Get("http://localhost/version") var err error
var resp *http.Response
var versionInfo struct {
Version string `json:"Version"`
}
const versionMaxTries = 2
for i := 1; i <= versionMaxTries; i++ {
resp, err = dm.client.Get("http://localhost/version")
if err == nil {
break
}
if resp != nil {
resp.Body.Close()
}
if i < versionMaxTries {
slog.Debug("Failed to get Docker version; retrying", "attempt", i, "error", err)
time.Sleep(5 * time.Second)
}
}
if err != nil { if err != nil {
return false, err
}
if resp.StatusCode != http.StatusOK {
status := resp.Status
resp.Body.Close()
return false, fmt.Errorf("docker version request failed: %s", status)
}
var versionInfo dockerVersionResponse
serverHeader := resp.Header.Get("Server")
if err := dm.decode(resp, &versionInfo); err != nil {
return false, err
}
dm.applyDockerVersionInfo(serverHeader, &versionInfo)
dm.dockerVersionChecked = true
return true, nil
}
// ensureDockerVersionChecked retries the version probe after a successful
// container list request.
func (dm *dockerManager) ensureDockerVersionChecked() {
if dm.dockerVersionChecked {
return return
} }
if _, err := dm.checkDockerVersion(); err != nil { if err := dm.decode(resp, &versionInfo); err != nil {
slog.Debug("Failed to get Docker version", "err", err)
}
}
// applyDockerVersionInfo updates version-dependent behavior from engine metadata.
func (dm *dockerManager) applyDockerVersionInfo(serverHeader string, versionInfo *dockerVersionResponse) {
if detectPodmanEngine(serverHeader, versionInfo) {
dm.setIsPodman()
return return
} }
// if version > 24, one-shot works correctly and we can limit concurrent operations // if version > 24, one-shot works correctly and we can limit concurrent operations
@@ -777,18 +615,20 @@ func (dm *dockerManager) applyDockerVersionInfo(serverHeader string, versionInfo
} }
} }
// Decodes a Docker API JSON response using a reusable buffer. Not thread safe. // Decodes Docker API JSON response using a reusable buffer and decoder. Not thread safe.
func (dm *dockerManager) decode(resp *http.Response, d any) error { func (dm *dockerManager) decode(resp *http.Response, d any) error {
if dm.buf == nil { if dm.buf == nil {
// initialize buffer with 256kb starting size // initialize buffer with 256kb starting size
dm.buf = bytes.NewBuffer(make([]byte, 0, 1024*256)) dm.buf = bytes.NewBuffer(make([]byte, 0, 1024*256))
dm.decoder = json.NewDecoder(dm.buf)
} }
defer resp.Body.Close() defer resp.Body.Close()
defer dm.buf.Reset() defer dm.buf.Reset()
if _, err := dm.buf.ReadFrom(resp.Body); err != nil { _, err := dm.buf.ReadFrom(resp.Body)
if err != nil {
return err return err
} }
return json.Unmarshal(dm.buf.Bytes(), d) return dm.decoder.Decode(d)
} }
// Test docker / podman sockets and return if one exists // Test docker / podman sockets and return if one exists
@@ -803,34 +643,9 @@ func getDockerHost() string {
return scheme + socks[0] return scheme + socks[0]
} }
func validateContainerID(containerID string) error {
if !dockerContainerIDPattern.MatchString(containerID) {
return fmt.Errorf("invalid container id")
}
return nil
}
func buildDockerContainerEndpoint(containerID, action string, query url.Values) (string, error) {
if err := validateContainerID(containerID); err != nil {
return "", err
}
u := &url.URL{
Scheme: "http",
Host: "localhost",
Path: fmt.Sprintf("/containers/%s/%s", url.PathEscape(containerID), action),
}
if len(query) > 0 {
u.RawQuery = query.Encode()
}
return u.String(), nil
}
// getContainerInfo fetches the inspection data for a container // getContainerInfo fetches the inspection data for a container
func (dm *dockerManager) getContainerInfo(ctx context.Context, containerID string) ([]byte, error) { func (dm *dockerManager) getContainerInfo(ctx context.Context, containerID string) ([]byte, error) {
endpoint, err := buildDockerContainerEndpoint(containerID, "json", nil) endpoint := fmt.Sprintf("http://localhost/containers/%s/json", containerID)
if err != nil {
return nil, err
}
req, err := http.NewRequestWithContext(ctx, http.MethodGet, endpoint, nil) req, err := http.NewRequestWithContext(ctx, http.MethodGet, endpoint, nil)
if err != nil { if err != nil {
return nil, err return nil, err
@@ -861,15 +676,7 @@ func (dm *dockerManager) getContainerInfo(ctx context.Context, containerID strin
// getLogs fetches the logs for a container // getLogs fetches the logs for a container
func (dm *dockerManager) getLogs(ctx context.Context, containerID string) (string, error) { func (dm *dockerManager) getLogs(ctx context.Context, containerID string) (string, error) {
query := url.Values{ endpoint := fmt.Sprintf("http://localhost/containers/%s/logs?stdout=1&stderr=1&tail=%d", containerID, dockerLogsTail)
"stdout": []string{"1"},
"stderr": []string{"1"},
"tail": []string{fmt.Sprintf("%d", dockerLogsTail)},
}
endpoint, err := buildDockerContainerEndpoint(containerID, "logs", query)
if err != nil {
return "", err
}
req, err := http.NewRequestWithContext(ctx, http.MethodGet, endpoint, nil) req, err := http.NewRequestWithContext(ctx, http.MethodGet, endpoint, nil)
if err != nil { if err != nil {
return "", err return "", err
@@ -887,17 +694,7 @@ func (dm *dockerManager) getLogs(ctx context.Context, containerID string) (strin
} }
var builder strings.Builder var builder strings.Builder
contentType := resp.Header.Get("Content-Type") if err := decodeDockerLogStream(resp.Body, &builder); err != nil {
multiplexed := strings.HasSuffix(contentType, "multiplexed-stream")
logReader := io.Reader(resp.Body)
if !multiplexed {
// Podman may return multiplexed logs without Content-Type. Sniff the first frame header
// with a small buffered reader only when the header check fails.
bufferedReader := bufio.NewReaderSize(resp.Body, 8)
multiplexed = detectDockerMultiplexedStream(bufferedReader)
logReader = bufferedReader
}
if err := decodeDockerLogStream(logReader, &builder, multiplexed); err != nil {
return "", err return "", err
} }
@@ -909,28 +706,7 @@ func (dm *dockerManager) getLogs(ctx context.Context, containerID string) (strin
return logs, nil return logs, nil
} }
func detectDockerMultiplexedStream(reader *bufio.Reader) bool { func decodeDockerLogStream(reader io.Reader, builder *strings.Builder) error {
const headerSize = 8
header, err := reader.Peek(headerSize)
if err != nil {
return false
}
if header[0] != 0x01 && header[0] != 0x02 {
return false
}
// Docker's stream framing header reserves bytes 1-3 as zero.
if header[1] != 0 || header[2] != 0 || header[3] != 0 {
return false
}
frameLen := binary.BigEndian.Uint32(header[4:])
return frameLen <= maxLogFrameSize
}
func decodeDockerLogStream(reader io.Reader, builder *strings.Builder, multiplexed bool) error {
if !multiplexed {
_, err := io.Copy(builder, io.LimitReader(reader, maxTotalLogSize))
return err
}
const headerSize = 8 const headerSize = 8
var header [headerSize]byte var header [headerSize]byte
totalBytesRead := 0 totalBytesRead := 0
@@ -990,46 +766,3 @@ func (dm *dockerManager) GetHostInfo() (info container.HostInfo, err error) {
func (dm *dockerManager) IsPodman() bool { func (dm *dockerManager) IsPodman() bool {
return dm.usingPodman return dm.usingPodman
} }
// setIsPodman sets the manager to Podman mode and updates system details accordingly.
func (dm *dockerManager) setIsPodman() {
if dm.usingPodman {
return
}
dm.usingPodman = true
dm.goodDockerVersion = true
dm.dockerVersionChecked = true
// keep system details updated - this may be detected late if server isn't ready when
// agent starts, so make sure we notify the hub if this happens later.
if dm.agent != nil {
dm.agent.updateSystemDetails(func(details *system.Details) {
details.Podman = true
})
}
}
// detectPodmanFromHeader identifies Podman from the Docker API server header.
func detectPodmanFromHeader(server string) bool {
return strings.HasPrefix(server, "Libpod")
}
// detectPodmanFromVersion identifies Podman from the version payload.
func detectPodmanFromVersion(versionInfo *dockerVersionResponse) bool {
if versionInfo == nil {
return false
}
for _, component := range versionInfo.Components {
if strings.HasPrefix(component.Name, "Podman") {
return true
}
}
return false
}
// detectPodmanEngine checks both header and version metadata for Podman.
func detectPodmanEngine(serverHeader string, versionInfo *dockerVersionResponse) bool {
if detectPodmanFromHeader(serverHeader) {
return true
}
return detectPodmanFromVersion(versionInfo)
}
-105
View File
@@ -1,105 +0,0 @@
package agent
import (
"log/slog"
"sync"
"time"
"github.com/distribution/reference"
"github.com/henrygd/beszel/internal/entities/container"
)
const imageUpdateInterval = time.Hour
type imageUpdateStatus struct {
available bool
checkedAt time.Time
}
func normalizedImageReference(image string) string {
named, err := reference.ParseNormalizedNamed(image)
if err != nil {
return ""
}
// Digest-pinned references cannot move to a new version.
if _, pinned := named.(reference.Digested); pinned {
return ""
}
return reference.TagNameOnly(named).String()
}
// refreshImageUpdates starts at most one background batch. Neither its network
// work nor its completion is part of the container metrics wait group.
func (dm *dockerManager) refreshImageUpdates(containers []*container.ApiInfo, now time.Time) {
dm.imageUpdatesMutex.Lock()
defer dm.imageUpdatesMutex.Unlock()
if dm.imageUpdatesRunning {
return
}
if dm.imageUpdates == nil {
dm.imageUpdates = make(map[string]*imageUpdateStatus)
}
active := make(map[string]struct{}, len(containers))
pending := make(map[string]*imageUpdateStatus)
for _, ctr := range containers {
if len(ctr.Names) > 0 && dm.shouldExcludeContainer(ctr.Names[0][1:]) {
continue
}
key := normalizedImageReference(ctr.Image)
if key == "" {
continue
}
active[key] = struct{}{}
entry := dm.imageUpdates[key]
if entry == nil {
entry = &imageUpdateStatus{}
dm.imageUpdates[key] = entry
}
if entry.checkedAt.IsZero() || now.Sub(entry.checkedAt) >= imageUpdateInterval {
pending[key] = entry
}
}
for key := range dm.imageUpdates {
if _, ok := active[key]; !ok {
delete(dm.imageUpdates, key)
}
}
if len(pending) == 0 {
return
}
dm.imageUpdatesRunning = true
go func() {
// Limit auxiliary requests even on hosts running many different images.
sem := make(chan struct{}, 2)
var wg sync.WaitGroup
for key, entry := range pending {
sem <- struct{}{}
wg.Add(1)
go func() {
defer wg.Done()
defer func() { <-sem }()
available, err := dm.checkImageUpdate(key)
if err != nil {
available = false
slog.Debug("Image update check failed", "image", key, "err", err)
}
dm.imageUpdatesMutex.Lock()
entry.available = available
entry.checkedAt = time.Now()
dm.imageUpdatesMutex.Unlock()
}()
}
wg.Wait()
dm.imageUpdatesMutex.Lock()
dm.imageUpdatesRunning = false
dm.imageUpdatesMutex.Unlock()
}()
}
func (dm *dockerManager) cachedImageUpdate(image string) bool {
key := normalizedImageReference(image)
dm.imageUpdatesMutex.RLock()
defer dm.imageUpdatesMutex.RUnlock()
entry := dm.imageUpdates[key]
return entry != nil && entry.available
}
-225
View File
@@ -1,225 +0,0 @@
//go:build testing
package agent
import (
"encoding/json"
"fmt"
"github.com/fxamacker/cbor/v2"
"io"
"net/http"
"net/http/httptest"
"strings"
"sync/atomic"
"testing"
"time"
"github.com/henrygd/beszel/internal/entities/container"
"github.com/stretchr/testify/require"
)
func waitForImageUpdates(t *testing.T, dm *dockerManager) {
t.Helper()
require.Eventually(t, func() bool {
dm.imageUpdatesMutex.RLock()
defer dm.imageUpdatesMutex.RUnlock()
return !dm.imageUpdatesRunning
}, time.Second*3, time.Millisecond)
}
func TestImageUpdateCacheAndStats(t *testing.T) {
local := "sha256:" + strings.Repeat("a", 64)
remote := "sha256:" + strings.Repeat("b", 64)
var inspections, lookups atomic.Int32
var fail atomic.Bool
var upToDate atomic.Bool
server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
switch {
case strings.HasPrefix(r.URL.Path, "/images/"):
inspections.Add(1)
fmt.Fprintf(w, `{"RepoDigests":["docker.io/library/nginx@%s"]}`, local)
case r.URL.Path == "/containers/json":
fmt.Fprint(w, `[{"Id":"aaaaaaaaaaaa","Names":["/one"],"Image":"nginx","Status":"Up 2 hours"},{"Id":"bbbbbbbbbbbb","Names":["/two"],"Image":"docker.io/library/nginx:latest","Status":"Up 2 hours"}]`)
case strings.Contains(r.URL.Path, "/stats"):
fmt.Fprint(w, `{"memory_stats":{"usage":1048576},"cpu_stats":{},"networks":{}}`)
default:
http.NotFound(w, r)
}
}))
defer server.Close()
dm := newDockerManagerForVersionTest(server)
dm.dockerVersionChecked = true
dm.registryClient = &http.Client{Timeout: time.Second, Transport: roundTripFunc(func(r *http.Request) (*http.Response, error) {
if fail.Load() {
return nil, fmt.Errorf("registry unavailable")
}
response := &http.Response{StatusCode: 200, Header: make(http.Header), Body: io.NopCloser(strings.NewReader(`{"token":"test"}`))}
if r.Method == http.MethodHead {
lookups.Add(1)
digest := remote
if upToDate.Load() {
digest = local
}
response.Header.Set("Docker-Content-Digest", digest)
}
return response, nil
})}
stats, err := dm.getDockerStats(defaultCacheTimeMs)
require.NoError(t, err)
require.Len(t, stats, 2)
waitForImageUpdates(t, dm)
require.EqualValues(t, 1, lookups.Load())
require.EqualValues(t, 1, inspections.Load())
stats, err = dm.getDockerStats(defaultCacheTimeMs)
require.NoError(t, err)
for _, stat := range stats {
require.True(t, stat.UpdateAvailable)
if stat.Id == "aaaaaaaaaaaa" {
require.Equal(t, "nginx", stat.Image)
} else {
require.Equal(t, "docker.io/library/nginx:latest", stat.Image)
}
}
require.EqualValues(t, 1, lookups.Load())
expire := func() {
dm.imageUpdatesMutex.Lock()
dm.imageUpdates["docker.io/library/nginx:latest"].checkedAt = time.Now().Add(-imageUpdateInterval)
dm.imageUpdatesMutex.Unlock()
}
upToDate.Store(true)
expire()
_, err = dm.getDockerStats(defaultCacheTimeMs)
require.NoError(t, err)
waitForImageUpdates(t, dm)
require.EqualValues(t, 2, lookups.Load())
require.False(t, dm.cachedImageUpdate("nginx:latest"))
// An expired positive result is cleared on failure, and the failure itself
// is cached so realtime stats do not retry a broken registry every second.
dm.imageUpdatesMutex.Lock()
dm.imageUpdates["docker.io/library/nginx:latest"].available = true
dm.imageUpdatesMutex.Unlock()
fail.Store(true)
expire()
_, err = dm.getDockerStats(defaultCacheTimeMs)
require.NoError(t, err)
waitForImageUpdates(t, dm)
failedInspections := inspections.Load()
stats, err = dm.getDockerStats(defaultCacheTimeMs)
require.NoError(t, err)
require.Len(t, stats, 2)
require.Equal(t, failedInspections, inspections.Load())
for _, stat := range stats {
require.False(t, stat.UpdateAvailable)
require.Equal(t, 1.0, stat.Mem)
}
}
func TestImageDiscoveryDoesNotBlockStats(t *testing.T) {
started := make(chan struct{}, 1)
release := make(chan struct{})
server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
if strings.HasPrefix(r.URL.Path, "/images/") {
fmt.Fprintf(w, `{"RepoDigests":["example.com/app@sha256:%s"]}`, strings.Repeat("a", 64))
} else {
fmt.Fprint(w, `{"memory_stats":{"usage":1048576}}`)
}
}))
defer server.Close()
dm := newDockerManagerForVersionTest(server)
defer func() { close(release); waitForImageUpdates(t, dm) }()
dm.registryClient = &http.Client{Transport: roundTripFunc(func(r *http.Request) (*http.Response, error) {
started <- struct{}{}
<-release
return nil, fmt.Errorf("timeout")
})}
ctr := &container.ApiInfo{IdShort: "aaaaaaaaaaaa", Image: "example.com/app", Names: []string{"/one"}}
dm.refreshImageUpdates([]*container.ApiInfo{ctr}, time.Now())
select {
case <-started:
case <-time.After(3 * time.Second):
t.Fatal("check did not start")
}
done := make(chan error, 1)
go func() { done <- dm.updateContainerStats(ctr, defaultCacheTimeMs) }()
select {
case err := <-done:
require.NoError(t, err)
case <-time.After(time.Second):
t.Fatal("registry blocked stats")
}
dm.imageUpdatesMutex.RLock()
require.True(t, dm.imageUpdatesRunning)
dm.imageUpdatesMutex.RUnlock()
}
func TestNormalizeImageUpdateReferences(t *testing.T) {
require.Equal(t, normalizedImageReference("nginx"), normalizedImageReference("docker.io/library/nginx:latest"))
require.Empty(t, normalizedImageReference("bad reference"))
require.Empty(t, normalizedImageReference("nginx@sha256:"+strings.Repeat("a", 64)))
}
// A stats request can return headers promptly and then stall while reading its
// body. The stats-map mutex must remain available during that read.
func TestStatsResponseBodyDoesNotHoldStatsLock(t *testing.T) {
started := make(chan struct{})
release := make(chan struct{})
server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
w.WriteHeader(http.StatusOK)
w.(http.Flusher).Flush()
close(started)
<-release
fmt.Fprint(w, `{"memory_stats":{"usage":1048576}}`)
}))
defer server.Close()
dm := newDockerManagerForVersionTest(server)
done := make(chan error, 1)
go func() {
done <- dm.updateContainerStats(&container.ApiInfo{IdShort: "aaaaaaaaaaaa", Names: []string{"/one"}, Image: "nginx"}, defaultCacheTimeMs)
}()
<-started
locked := make(chan struct{})
go func() { dm.containerStatsMutex.Lock(); dm.containerStatsMutex.Unlock(); close(locked) }()
select {
case <-locked:
case <-time.After(time.Second):
close(release)
<-done
t.Fatal("Docker response body held the stats mutex")
}
close(release)
require.NoError(t, <-done)
}
func TestImageUpdateStatsEncoding(t *testing.T) {
original := container.Stats{Image: "nginx:latest", UpdateAvailable: true}
encoded, err := cbor.Marshal(original)
require.NoError(t, err)
var fields map[int]any
require.NoError(t, cbor.Unmarshal(encoded, &fields))
require.Equal(t, true, fields[11])
require.Equal(t, "nginx:latest", fields[8])
var decoded container.Stats
require.NoError(t, cbor.Unmarshal(encoded, &decoded))
require.True(t, decoded.UpdateAvailable)
require.Equal(t, original.Image, decoded.Image)
encoded, err = json.Marshal(original)
require.NoError(t, err)
require.Contains(t, string(encoded), `"u":true`)
}
func TestImageUpdateCacheExpiryBoundaryAndPruning(t *testing.T) {
now := time.Now()
key := normalizedImageReference("nginx")
dm := &dockerManager{imageUpdates: map[string]*imageUpdateStatus{
key: {available: true, checkedAt: now},
"unused.example/image:latest": {checkedAt: now},
}}
dm.refreshImageUpdates([]*container.ApiInfo{{Image: "nginx"}}, now.Add(imageUpdateInterval-time.Nanosecond))
require.False(t, dm.imageUpdatesRunning)
require.Len(t, dm.imageUpdates, 1)
require.True(t, dm.cachedImageUpdate("nginx:latest"))
dm.refreshImageUpdates(nil, now)
require.Empty(t, dm.imageUpdates)
}
-222
View File
@@ -1,222 +0,0 @@
package agent
import (
_ "crypto/sha256"
"encoding/json"
"fmt"
"net/http"
"net/url"
"strings"
"time"
"github.com/distribution/reference"
"github.com/opencontainers/go-digest"
)
const imageRegistryTimeout = 10 * time.Second
const imageManifestAccept = "application/vnd.docker.distribution.manifest.list.v2+json, " +
"application/vnd.docker.distribution.manifest.v2+json, " +
"application/vnd.oci.image.manifest.v1+json, " +
"application/vnd.oci.image.index.v1+json"
// checkImageUpdate compares the digest recorded by Docker for image with the
// digest currently advertised by its registry. A digest-pinned reference is
// immutable and therefore never has an update available.
func (dm *dockerManager) checkImageUpdate(image string) (bool, error) {
named, err := reference.ParseNormalizedNamed(image)
if err != nil {
return false, fmt.Errorf("parse image reference %q: %w", image, err)
}
if _, pinned := named.(reference.Digested); pinned {
return false, nil
}
named = reference.TagNameOnly(named)
registry := reference.Domain(named)
repository := reference.Path(named)
tag := named.(reference.Tagged).Tag()
localDigest, err := dm.inspectImageDigest(image, registry, repository)
if err != nil {
return false, err
}
remoteDigest, err := dm.registryImageDigest(registry, repository, tag)
if err != nil {
return false, err
}
return remoteDigest != localDigest, nil
}
// inspectImageDigest reads Docker's image metadata without using dm.decode.
// The checker runs in the image-discovery goroutine, so it must not hold any
// of the container statistics locks while waiting on the Docker API.
func (dm *dockerManager) inspectImageDigest(image, registry, repository string) (string, error) {
if dm.client == nil {
return "", fmt.Errorf("inspect image %q: Docker client is unavailable", image)
}
endpoint := "http://localhost/images/" + url.PathEscape(image) + "/json"
resp, err := dm.client.Get(endpoint)
if err != nil {
return "", fmt.Errorf("inspect image %q: %w", image, err)
}
defer resp.Body.Close()
if resp.StatusCode != http.StatusOK {
return "", fmt.Errorf("inspect image %q failed: %s", image, responseStatus(resp))
}
var inspect struct {
RepoDigests []string `json:"RepoDigests"`
}
if err := json.NewDecoder(resp.Body).Decode(&inspect); err != nil {
return "", fmt.Errorf("decode image inspect %q: %w", image, err)
}
if len(inspect.RepoDigests) == 0 {
return "", fmt.Errorf("inspect image %q returned no repository digests", image)
}
localDigest, ok := matchingRepositoryDigest(inspect.RepoDigests, registry, repository)
if !ok {
return "", fmt.Errorf("inspect image %q returned no valid digest for %s/%s", image, registry, repository)
}
return localDigest, nil
}
// matchingRepositoryDigest returns a valid digest belonging to the requested
// repository. Docker can return multiple RepoDigests for one local image; an
// unrelated first entry must never be used for the comparison.
func matchingRepositoryDigest(repoDigests []string, registry, repository string) (string, bool) {
for _, repoDigest := range repoDigests {
repoDigest = strings.TrimSpace(repoDigest)
at := strings.LastIndexByte(repoDigest, '@')
if at <= 0 || at == len(repoDigest)-1 || strings.Contains(repoDigest[:at], "@") {
continue
}
repoRef, err := reference.ParseNormalizedNamed(repoDigest[:at])
if err != nil || reference.Path(repoRef) != repository || !sameRegistry(reference.Domain(repoRef), registry) {
continue
}
if _, hasTag := repoRef.(reference.Tagged); hasTag {
continue
}
d, err := digest.Parse(repoDigest[at+1:])
if err != nil {
continue
}
return d.String(), true
}
return "", false
}
func sameRegistry(left, right string) bool {
left = canonicalRegistry(left)
right = canonicalRegistry(right)
return left == right ||
(left == "ghcr.io" && right == "lscr.io") ||
(left == "lscr.io" && right == "ghcr.io")
}
func canonicalRegistry(registry string) string {
if registry == "index.docker.io" {
return "docker.io"
}
return registry
}
func (dm *dockerManager) registryImageDigest(registry, repository, tag string) (string, error) {
client := dm.registryClient
if client == nil {
client = &http.Client{Timeout: imageRegistryTimeout}
}
token, err := dm.registryToken(client, registry, repository)
if err != nil {
return "", err
}
host := registry
if registry == "docker.io" {
host = "registry-1.docker.io"
}
manifestURL := "https://" + host + "/v2/" + repository + "/manifests/" + url.PathEscape(tag)
req, err := http.NewRequest(http.MethodHead, manifestURL, nil)
if err != nil {
return "", fmt.Errorf("create manifest request: %w", err)
}
req.Header.Set("Accept", imageManifestAccept)
if token != "" {
req.Header.Set("Authorization", "Bearer "+token)
}
resp, err := client.Do(req)
if err != nil {
return "", fmt.Errorf("fetch manifest %s:%s: %w", registry, repository, err)
}
defer resp.Body.Close()
if resp.StatusCode != http.StatusOK {
return "", fmt.Errorf("manifest request for %s:%s failed: %s", repository, tag, responseStatus(resp))
}
remote := strings.TrimSpace(resp.Header.Get("Docker-Content-Digest"))
d, err := digest.Parse(remote)
if err != nil {
return "", fmt.Errorf("manifest request for %s:%s returned invalid digest: %w", repository, tag, err)
}
return d.String(), nil
}
func (dm *dockerManager) registryToken(client *http.Client, registry, repository string) (string, error) {
var authURL string
switch registry {
case "docker.io":
authURL = "https://auth.docker.io/token?service=registry.docker.io&scope=" + url.QueryEscape("repository:"+repository+":pull")
case "ghcr.io", "lscr.io":
// lscr.io is the LinuxServer alias for its GHCR-backed images.
authURL = "https://ghcr.io/token?service=ghcr.io&scope=" + url.QueryEscape("repository:"+repository+":pull")
default:
// Anonymous registries remain supported, as they were before the
// authenticated Docker Hub and GHCR paths were added.
return "", nil
}
req, err := http.NewRequest(http.MethodGet, authURL, nil)
if err != nil {
return "", fmt.Errorf("create registry auth request: %w", err)
}
resp, err := client.Do(req)
if err != nil {
return "", fmt.Errorf("fetch registry auth token for %s: %w", repository, err)
}
defer resp.Body.Close()
if resp.StatusCode != http.StatusOK {
return "", fmt.Errorf("registry auth request for %s failed: %s", repository, responseStatus(resp))
}
var tokenResponse struct {
Token string `json:"token"`
AccessToken string `json:"access_token"`
}
if err := json.NewDecoder(resp.Body).Decode(&tokenResponse); err != nil {
return "", fmt.Errorf("decode registry auth response for %s: %w", repository, err)
}
token := strings.TrimSpace(tokenResponse.Token)
if token == "" {
token = strings.TrimSpace(tokenResponse.AccessToken)
}
if token == "" {
return "", fmt.Errorf("registry auth response for %s contained no token", repository)
}
return token, nil
}
func responseStatus(resp *http.Response) string {
if resp.Status != "" {
return resp.Status
}
return http.StatusText(resp.StatusCode)
}
-204
View File
@@ -1,204 +0,0 @@
//go:build testing
package agent
import (
"fmt"
"io"
"net/http"
"net/http/httptest"
"strings"
"sync/atomic"
"testing"
"github.com/stretchr/testify/require"
)
type registryTransportFunc func(*http.Request) (*http.Response, error)
func (fn registryTransportFunc) RoundTrip(req *http.Request) (*http.Response, error) {
return fn(req)
}
func registryResponse(status int, body string) *http.Response {
return &http.Response{
StatusCode: status,
Status: fmt.Sprintf("%d %s", status, http.StatusText(status)),
Header: make(http.Header),
Body: io.NopCloser(strings.NewReader(body)),
}
}
func registryDigest(fill byte) string {
return "sha256:" + strings.Repeat(string(fill), 64)
}
func newRegistryChecker(t *testing.T, inspectBody string, transport http.RoundTripper) *dockerManager {
t.Helper()
server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
if strings.HasPrefix(r.URL.Path, "/images/") {
w.Header().Set("Content-Type", "application/json")
_, _ = io.WriteString(w, inspectBody)
return
}
http.NotFound(w, r)
}))
t.Cleanup(server.Close)
return &dockerManager{
client: newDockerManagerForVersionTest(server).client,
registryClient: &http.Client{Transport: transport},
}
}
func TestCheckImageUpdateUsesInspectAndManifestDigests(t *testing.T) {
local := registryDigest('a')
remote := registryDigest('b')
var authCalls, manifestCalls atomic.Int32
dm := newRegistryChecker(t, fmt.Sprintf(`{"RepoDigests":["docker.io/library/alpine@%s"]}`, local), registryTransportFunc(func(req *http.Request) (*http.Response, error) {
switch {
case req.Method == http.MethodGet && req.URL.Host == "auth.docker.io":
authCalls.Add(1)
require.Equal(t, "/token", req.URL.Path)
return registryResponse(http.StatusOK, `{"token":"test-token"}`), nil
case req.Method == http.MethodHead && req.URL.Host == "registry-1.docker.io":
manifestCalls.Add(1)
require.Equal(t, "/v2/library/alpine/manifests/latest", req.URL.Path)
require.Equal(t, "Bearer test-token", req.Header.Get("Authorization"))
resp := registryResponse(http.StatusOK, "")
resp.Header.Set("Docker-Content-Digest", remote)
return resp, nil
default:
return registryResponse(http.StatusNotFound, ""), nil
}
}))
available, err := dm.checkImageUpdate("alpine")
require.NoError(t, err)
require.True(t, available)
require.EqualValues(t, 1, authCalls.Load())
require.EqualValues(t, 1, manifestCalls.Load())
}
func TestCheckImageUpdateReportsUnknownInspectState(t *testing.T) {
for _, test := range []struct {
name string
body string
}{
{name: "missing field", body: `{}`},
{name: "empty field", body: `{"RepoDigests":[]}`},
{name: "malformed reference", body: `{"RepoDigests":["not-a-repo-digest"]}`},
{name: "wrong repository", body: `{"RepoDigests":["docker.io/library/busybox@` + registryDigest('a') + `"]}`},
{name: "malformed digest", body: `{"RepoDigests":["docker.io/library/alpine@sha256:not-a-digest"]}`},
} {
t.Run(test.name, func(t *testing.T) {
var registryCalls atomic.Int32
dm := newRegistryChecker(t, test.body, registryTransportFunc(func(req *http.Request) (*http.Response, error) {
registryCalls.Add(1)
return registryResponse(http.StatusOK, `{"token":"unexpected"}`), nil
}))
available, err := dm.checkImageUpdate("alpine")
require.Error(t, err)
require.False(t, available)
require.EqualValues(t, 0, registryCalls.Load(), "invalid local state must not query a registry")
})
}
}
func TestCheckImageUpdateChecksInspectAuthAndManifestStatuses(t *testing.T) {
local := registryDigest('a')
validInspect := fmt.Sprintf(`{"RepoDigests":["docker.io/library/alpine@%s"]}`, local)
tests := []struct {
name string
inspectCode int
authCode int
manifestCode int
remote string
want string
}{
{name: "inspect status", inspectCode: http.StatusNotFound, want: "inspect image"},
{name: "auth status", inspectCode: http.StatusOK, authCode: http.StatusUnauthorized, want: "registry auth"},
{name: "manifest status", inspectCode: http.StatusOK, authCode: http.StatusOK, manifestCode: http.StatusNotFound, remote: local, want: "manifest request"},
{name: "missing digest", inspectCode: http.StatusOK, authCode: http.StatusOK, manifestCode: http.StatusOK, want: "invalid digest"},
}
for _, test := range tests {
t.Run(test.name, func(t *testing.T) {
server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
if test.inspectCode != http.StatusOK && strings.HasPrefix(r.URL.Path, "/images/") {
w.WriteHeader(test.inspectCode)
return
}
_, _ = io.WriteString(w, validInspect)
}))
t.Cleanup(server.Close)
calls := 0
dm := &dockerManager{client: newDockerManagerForVersionTest(server).client, registryClient: &http.Client{Transport: registryTransportFunc(func(req *http.Request) (*http.Response, error) {
calls++
if req.Method == http.MethodGet {
return registryResponse(test.authCode, `{"token":"test"}`), nil
}
response := registryResponse(test.manifestCode, "")
response.Header.Set("Docker-Content-Digest", test.remote)
return response, nil
})}}
_, err := dm.checkImageUpdate("alpine")
require.Error(t, err)
require.Contains(t, err.Error(), test.want)
if test.inspectCode != http.StatusOK {
require.Zero(t, calls)
}
})
}
}
func TestCheckImageUpdateSupportsAnonymousAndLSCRRegistries(t *testing.T) {
t.Run("anonymous registry", func(t *testing.T) {
local := registryDigest('a')
var calls atomic.Int32
dm := newRegistryChecker(t, fmt.Sprintf(`{"RepoDigests":["example.com/app@%s"]}`, local), registryTransportFunc(func(req *http.Request) (*http.Response, error) {
calls.Add(1)
require.Equal(t, http.MethodHead, req.Method)
require.Equal(t, "example.com", req.URL.Host)
resp := registryResponse(http.StatusOK, "")
resp.Header.Set("Docker-Content-Digest", local)
return resp, nil
}))
available, err := dm.checkImageUpdate("example.com/app")
require.NoError(t, err)
require.False(t, available)
require.EqualValues(t, 1, calls.Load())
})
t.Run("lscr ghcr alias", func(t *testing.T) {
local := registryDigest('a')
var authCalls, manifestCalls atomic.Int32
dm := newRegistryChecker(t, fmt.Sprintf(`{"RepoDigests":["ghcr.io/linuxserver/app@%s"]}`, local), registryTransportFunc(func(req *http.Request) (*http.Response, error) {
if req.Method == http.MethodGet {
authCalls.Add(1)
return registryResponse(http.StatusOK, `{"token":"test"}`), nil
}
manifestCalls.Add(1)
require.Equal(t, "lscr.io", req.URL.Host)
resp := registryResponse(http.StatusOK, "")
resp.Header.Set("Docker-Content-Digest", local)
return resp, nil
}))
available, err := dm.checkImageUpdate("lscr.io/linuxserver/app")
require.NoError(t, err)
require.False(t, available)
require.EqualValues(t, 1, authCalls.Load())
require.EqualValues(t, 1, manifestCalls.Load())
})
}
func TestCheckImageUpdateSkipsPinnedDigest(t *testing.T) {
image := "docker.io/library/alpine@" + registryDigest('a')
dm := &dockerManager{}
available, err := dm.checkImageUpdate(image)
require.NoError(t, err)
require.False(t, available)
}
+80 -983
View File
File diff suppressed because it is too large Load Diff
-95
View File
@@ -1,95 +0,0 @@
package agent
import (
"fmt"
"strconv"
"strings"
)
func isEmmcBlockName(name string) bool {
if !strings.HasPrefix(name, "mmcblk") {
return false
}
suffix := strings.TrimPrefix(name, "mmcblk")
if suffix == "" {
return false
}
for _, c := range suffix {
if c < '0' || c > '9' {
return false
}
}
return true
}
func parseHexOrDecByte(s string) (uint8, bool) {
s = strings.TrimSpace(s)
if s == "" {
return 0, false
}
base := 10
if strings.HasPrefix(s, "0x") || strings.HasPrefix(s, "0X") {
base = 16
s = s[2:]
}
parsed, err := strconv.ParseUint(s, base, 8)
if err != nil {
return 0, false
}
return uint8(parsed), true
}
func parseHexBytePair(s string) (uint8, uint8, bool) {
fields := strings.Fields(s)
if len(fields) < 2 {
return 0, 0, false
}
a, okA := parseHexOrDecByte(fields[0])
b, okB := parseHexOrDecByte(fields[1])
if !okA && !okB {
return 0, 0, false
}
return a, b, true
}
func emmcSmartStatus(preEOL uint8) string {
switch preEOL {
case 0x01:
return "PASSED"
case 0x02:
return "WARNING"
case 0x03:
return "FAILED"
default:
return "UNKNOWN"
}
}
func emmcPreEOLString(preEOL uint8) string {
switch preEOL {
case 0x01:
return "0x01 (normal)"
case 0x02:
return "0x02 (warning)"
case 0x03:
return "0x03 (urgent)"
default:
return fmt.Sprintf("0x%02x", preEOL)
}
}
func emmcLifeTimeString(v uint8) string {
// JEDEC eMMC: 0x01..0x0A => 0-100% used in 10% steps, 0x0B => exceeded.
switch {
case v == 0:
return "0x00 (not reported)"
case v >= 0x01 && v <= 0x0A:
low := int(v-1) * 10
high := int(v) * 10
return fmt.Sprintf("0x%02x (%d-%d%% used)", v, low, high)
case v == 0x0B:
return "0x0b (>100% used)"
default:
return fmt.Sprintf("0x%02x", v)
}
}
-78
View File
@@ -1,78 +0,0 @@
package agent
import "testing"
func TestParseHexOrDecByte(t *testing.T) {
tests := []struct {
in string
want uint8
ok bool
}{
{"0x01", 1, true},
{"0X0b", 11, true},
{"01", 1, true},
{" 3 ", 3, true},
{"", 0, false},
{"0x", 0, false},
{"nope", 0, false},
}
for _, tt := range tests {
got, ok := parseHexOrDecByte(tt.in)
if ok != tt.ok || got != tt.want {
t.Fatalf("parseHexOrDecByte(%q) = (%d,%v), want (%d,%v)", tt.in, got, ok, tt.want, tt.ok)
}
}
}
func TestParseHexBytePair(t *testing.T) {
a, b, ok := parseHexBytePair("0x01 0x02\n")
if !ok || a != 1 || b != 2 {
t.Fatalf("parseHexBytePair hex = (%d,%d,%v), want (1,2,true)", a, b, ok)
}
a, b, ok = parseHexBytePair("01 02")
if !ok || a != 1 || b != 2 {
t.Fatalf("parseHexBytePair dec = (%d,%d,%v), want (1,2,true)", a, b, ok)
}
_, _, ok = parseHexBytePair("0x01")
if ok {
t.Fatalf("parseHexBytePair short input ok=true, want false")
}
}
func TestEmmcSmartStatus(t *testing.T) {
if got := emmcSmartStatus(0x01); got != "PASSED" {
t.Fatalf("emmcSmartStatus(0x01) = %q, want PASSED", got)
}
if got := emmcSmartStatus(0x02); got != "WARNING" {
t.Fatalf("emmcSmartStatus(0x02) = %q, want WARNING", got)
}
if got := emmcSmartStatus(0x03); got != "FAILED" {
t.Fatalf("emmcSmartStatus(0x03) = %q, want FAILED", got)
}
if got := emmcSmartStatus(0x00); got != "UNKNOWN" {
t.Fatalf("emmcSmartStatus(0x00) = %q, want UNKNOWN", got)
}
}
func TestIsEmmcBlockName(t *testing.T) {
cases := []struct {
name string
ok bool
}{
{"mmcblk0", true},
{"mmcblk1", true},
{"mmcblk10", true},
{"mmcblk0p1", false},
{"sda", false},
{"mmcblk", false},
{"mmcblkA", false},
}
for _, c := range cases {
if got := isEmmcBlockName(c.name); got != c.ok {
t.Fatalf("isEmmcBlockName(%q) = %v, want %v", c.name, got, c.ok)
}
}
}
-215
View File
@@ -1,215 +0,0 @@
//go:build linux
package agent
import (
"os"
"path/filepath"
"strconv"
"strings"
"github.com/henrygd/beszel/agent/utils"
"github.com/henrygd/beszel/internal/entities/smart"
)
// emmcSysfsRoot is a test hook; production value is "/sys".
var emmcSysfsRoot = "/sys"
type emmcHealth struct {
model string
serial string
revision string
capacity uint64
preEOL uint8
lifeA uint8
lifeB uint8
}
func scanEmmcDevices() []*DeviceInfo {
blockDir := filepath.Join(emmcSysfsRoot, "class", "block")
entries, err := os.ReadDir(blockDir)
if err != nil {
return nil
}
devices := make([]*DeviceInfo, 0, 2)
for _, ent := range entries {
name := ent.Name()
if !isEmmcBlockName(name) {
continue
}
deviceDir := filepath.Join(blockDir, name, "device")
if !hasEmmcHealthFiles(deviceDir) {
continue
}
devPath := filepath.Join("/dev", name)
devices = append(devices, &DeviceInfo{
Name: devPath,
Type: "emmc",
InfoName: devPath + " [eMMC]",
Protocol: "MMC",
})
}
return devices
}
func (sm *SmartManager) collectEmmcHealth(deviceInfo *DeviceInfo) (bool, error) {
if deviceInfo == nil || deviceInfo.Name == "" {
return false, nil
}
base := filepath.Base(deviceInfo.Name)
if !isEmmcBlockName(base) && !strings.EqualFold(deviceInfo.Type, "emmc") && !strings.EqualFold(deviceInfo.Type, "mmc") {
return false, nil
}
health, ok := readEmmcHealth(base)
if !ok {
return false, nil
}
// Normalize the device type to keep pruning logic stable across refreshes.
deviceInfo.Type = "emmc"
key := health.serial
if key == "" {
key = filepath.Join("/dev", base)
}
status := emmcSmartStatus(health.preEOL)
attrs := []*smart.SmartAttribute{
{
Name: "PreEOLInfo",
RawValue: uint64(health.preEOL),
RawString: emmcPreEOLString(health.preEOL),
},
{
Name: "DeviceLifeTimeEstA",
RawValue: uint64(health.lifeA),
RawString: emmcLifeTimeString(health.lifeA),
},
{
Name: "DeviceLifeTimeEstB",
RawValue: uint64(health.lifeB),
RawString: emmcLifeTimeString(health.lifeB),
},
}
sm.Lock()
defer sm.Unlock()
if _, exists := sm.SmartDataMap[key]; !exists {
sm.SmartDataMap[key] = &smart.SmartData{}
}
data := sm.SmartDataMap[key]
data.ModelName = health.model
data.SerialNumber = health.serial
data.FirmwareVersion = health.revision
data.Capacity = health.capacity
data.Temperature = 0
data.SmartStatus = status
data.DiskName = filepath.Join("/dev", base)
data.DiskType = "emmc"
data.Attributes = attrs
return true, nil
}
func readEmmcHealth(blockName string) (emmcHealth, bool) {
var out emmcHealth
if !isEmmcBlockName(blockName) {
return out, false
}
deviceDir := filepath.Join(emmcSysfsRoot, "class", "block", blockName, "device")
preEOL, okPre := readHexByteFile(filepath.Join(deviceDir, "pre_eol_info"))
// Some kernels expose EXT_CSD lifetime via "life_time" (two bytes), others as
// separate files. Support both.
lifeA, lifeB, okLife := readLifeTime(deviceDir)
if !okPre && !okLife {
return out, false
}
out.preEOL = preEOL
out.lifeA = lifeA
out.lifeB = lifeB
out.model = utils.ReadStringFile(filepath.Join(deviceDir, "name"))
out.serial = utils.ReadStringFile(filepath.Join(deviceDir, "serial"))
out.revision = utils.ReadStringFile(filepath.Join(deviceDir, "prv"))
if capBytes, ok := readBlockCapacityBytes(blockName); ok {
out.capacity = capBytes
}
return out, true
}
func readLifeTime(deviceDir string) (uint8, uint8, bool) {
if content, ok := utils.ReadStringFileOK(filepath.Join(deviceDir, "life_time")); ok {
a, b, ok := parseHexBytePair(content)
return a, b, ok
}
a, okA := readHexByteFile(filepath.Join(deviceDir, "device_life_time_est_typ_a"))
b, okB := readHexByteFile(filepath.Join(deviceDir, "device_life_time_est_typ_b"))
if okA || okB {
return a, b, true
}
return 0, 0, false
}
func readBlockCapacityBytes(blockName string) (uint64, bool) {
sizePath := filepath.Join(emmcSysfsRoot, "class", "block", blockName, "size")
lbsPath := filepath.Join(emmcSysfsRoot, "class", "block", blockName, "queue", "logical_block_size")
sizeStr, ok := utils.ReadStringFileOK(sizePath)
if !ok {
return 0, false
}
sectors, err := strconv.ParseUint(sizeStr, 10, 64)
if err != nil || sectors == 0 {
return 0, false
}
lbsStr, ok := utils.ReadStringFileOK(lbsPath)
logicalBlockSize := uint64(512)
if ok {
if parsed, err := strconv.ParseUint(lbsStr, 10, 64); err == nil && parsed > 0 {
logicalBlockSize = parsed
}
}
return sectors * logicalBlockSize, true
}
func readHexByteFile(path string) (uint8, bool) {
content, ok := utils.ReadStringFileOK(path)
if !ok {
return 0, false
}
b, ok := parseHexOrDecByte(content)
return b, ok
}
func hasEmmcHealthFiles(deviceDir string) bool {
entries, err := os.ReadDir(deviceDir)
if err != nil {
return false
}
for _, ent := range entries {
switch ent.Name() {
case "pre_eol_info", "life_time", "device_life_time_est_typ_a", "device_life_time_est_typ_b":
return true
}
}
return false
}
-80
View File
@@ -1,80 +0,0 @@
//go:build linux
package agent
import (
"os"
"path/filepath"
"testing"
"github.com/henrygd/beszel/internal/entities/smart"
)
func TestEmmcMockSysfsScanAndCollect(t *testing.T) {
tmp := t.TempDir()
prev := emmcSysfsRoot
emmcSysfsRoot = tmp
t.Cleanup(func() { emmcSysfsRoot = prev })
// Fake: /sys/class/block/mmcblk0
mmcDeviceDir := filepath.Join(tmp, "class", "block", "mmcblk0", "device")
mmcQueueDir := filepath.Join(tmp, "class", "block", "mmcblk0", "queue")
if err := os.MkdirAll(mmcDeviceDir, 0o755); err != nil {
t.Fatal(err)
}
if err := os.MkdirAll(mmcQueueDir, 0o755); err != nil {
t.Fatal(err)
}
write := func(path, content string) {
t.Helper()
if err := os.WriteFile(path, []byte(content), 0o644); err != nil {
t.Fatal(err)
}
}
write(filepath.Join(mmcDeviceDir, "pre_eol_info"), "0x02\n")
write(filepath.Join(mmcDeviceDir, "life_time"), "0x04 0x05\n")
write(filepath.Join(mmcDeviceDir, "name"), "H26M52103FMR\n")
write(filepath.Join(mmcDeviceDir, "serial"), "01234567\n")
write(filepath.Join(mmcDeviceDir, "prv"), "0x08\n")
write(filepath.Join(mmcQueueDir, "logical_block_size"), "512\n")
write(filepath.Join(tmp, "class", "block", "mmcblk0", "size"), "1024\n") // sectors
devs := scanEmmcDevices()
if len(devs) != 1 {
t.Fatalf("scanEmmcDevices() = %d devices, want 1", len(devs))
}
if devs[0].Name != "/dev/mmcblk0" || devs[0].Type != "emmc" {
t.Fatalf("scanEmmcDevices()[0] = %+v, want Name=/dev/mmcblk0 Type=emmc", devs[0])
}
sm := &SmartManager{SmartDataMap: map[string]*smart.SmartData{}}
ok, err := sm.collectEmmcHealth(devs[0])
if err != nil || !ok {
t.Fatalf("collectEmmcHealth() = (ok=%v, err=%v), want (true,nil)", ok, err)
}
if len(sm.SmartDataMap) != 1 {
t.Fatalf("SmartDataMap len=%d, want 1", len(sm.SmartDataMap))
}
var got *smart.SmartData
for _, v := range sm.SmartDataMap {
got = v
break
}
if got == nil {
t.Fatalf("SmartDataMap value nil")
}
if got.DiskType != "emmc" || got.DiskName != "/dev/mmcblk0" {
t.Fatalf("disk fields = (type=%q name=%q), want (emmc,/dev/mmcblk0)", got.DiskType, got.DiskName)
}
if got.SmartStatus != "WARNING" {
t.Fatalf("SmartStatus=%q, want WARNING", got.SmartStatus)
}
if got.SerialNumber != "01234567" || got.ModelName == "" || got.Capacity == 0 {
t.Fatalf("identity fields = (model=%q serial=%q cap=%d), want non-empty model, serial 01234567, cap>0", got.ModelName, got.SerialNumber, got.Capacity)
}
if len(got.Attributes) < 3 {
t.Fatalf("attributes len=%d, want >= 3", len(got.Attributes))
}
}
-14
View File
@@ -1,14 +0,0 @@
//go:build !linux
package agent
// Non-Linux builds: eMMC health via sysfs is not available.
func scanEmmcDevices() []*DeviceInfo {
return nil
}
func (sm *SmartManager) collectEmmcHealth(deviceInfo *DeviceInfo) (bool, error) {
return false, nil
}
-117
View File
@@ -1,117 +0,0 @@
package agent
import (
"log/slog"
"os"
"path/filepath"
"strings"
"sync"
"github.com/henrygd/beszel/agent/utils"
"github.com/henrygd/beszel/internal/entities/system"
)
type fanSensor struct {
key, path string
}
var getFanSensors = newFanSensorCache(hwmonRoot)
func newFanSensorCache(root string) func() ([]fanSensor, error) {
return sync.OnceValues(func() ([]fanSensor, error) {
return discoverHwmonFans(root)
})
}
// updateFans populates systemStats.Fans from the host's hwmon sysfs tree.
// No-op on platforms where hwmon isn't available (see fans_other.go).
func (a *Agent) updateFans(systemStats *system.Stats) {
if hwmonRoot == "" {
return
}
sensors, err := getFanSensors()
if err != nil {
slog.Debug("Error reading fans", "err", err)
return
}
fans := readFanSensors(sensors)
if len(fans) == 0 {
return
}
systemStats.Fans = fans
// Note: Commented out because we don't currently use this value in the UI.
// Compute the single "dashboard" value used by the FanSpeed alert.
// Per-sensor RPMs live in Stats.Fans and drive the multi-line FanChart
// in the UI; the alert path only needs one number to compare against
// the user's threshold, so we use the highest RPM across all fans
// a.systemInfo.DashboardFan = 0
// for _, rpm := range fans {
// if rpm > a.systemInfo.DashboardFan {
// a.systemInfo.DashboardFan = rpm
// }
// }
}
// readHwmonFans walks the given hwmon root (typically /sys/class/hwmon) and
// returns a map of "<chip>_<label-or-fan-idx>" → RPM for every fan*_input
// file it finds. Zero RPM is retained because it can represent a real fan that
// has stopped; negative and malformed readings are ignored.
func readHwmonFans(root string) (map[string]uint16, error) {
sensors, err := discoverHwmonFans(root)
if err != nil {
return nil, err
}
return readFanSensors(sensors), nil
}
func discoverHwmonFans(root string) ([]fanSensor, error) {
entries, err := os.ReadDir(root)
if err != nil {
return nil, err
}
var sensors []fanSensor
for _, entry := range entries {
chipDir := filepath.Join(root, entry.Name())
sensorDir := chipDir
inputs, _ := filepath.Glob(filepath.Join(sensorDir, "fan*_input"))
// Some legacy hwmon drivers (notably applesmc) register a hwmon class
// device but create fan attributes on the parent platform device. In
// sysfs that parent is exposed through hwmonN/device.
if len(inputs) == 0 {
deviceDir := filepath.Join(chipDir, "device")
if deviceInputs, _ := filepath.Glob(filepath.Join(deviceDir, "fan*_input")); len(deviceInputs) > 0 {
sensorDir = deviceDir
inputs = deviceInputs
}
}
chipName := utils.ReadStringFile(filepath.Join(sensorDir, "name"))
if chipName == "" {
chipName = utils.ReadStringFile(filepath.Join(chipDir, "name"))
}
if chipName == "" {
chipName = entry.Name()
}
for _, inputPath := range inputs {
base := strings.TrimSuffix(filepath.Base(inputPath), "_input")
label := utils.ReadStringFile(filepath.Join(sensorDir, base+"_label"))
key := chipName + "_" + base
if label != "" {
key = chipName + "_" + label
}
sensors = append(sensors, fanSensor{key, inputPath})
}
}
return sensors, nil
}
func readFanSensors(sensors []fanSensor) map[string]uint16 {
fans := make(map[string]uint16, len(sensors))
for _, sensor := range sensors {
if rpm, ok := utils.ReadUintFile(sensor.path); ok {
fans[sensor.key] = uint16(rpm)
}
}
return fans
}
-8
View File
@@ -1,8 +0,0 @@
//go:build linux
package agent
// hwmonRoot is the sysfs entry point for hardware monitor chips. Each
// subdirectory (hwmon0, hwmon1, …) is one chip; fan*_input files inside it
// expose RPM readings.
const hwmonRoot = "/sys/class/hwmon"
-7
View File
@@ -1,7 +0,0 @@
//go:build !linux
package agent
// hwmonRoot is empty on non-Linux platforms — fan RPM reporting via sysfs
// hwmon is Linux-specific. updateFans() short-circuits when this is empty.
const hwmonRoot = ""
-105
View File
@@ -1,105 +0,0 @@
//go:build testing
package agent
import (
"os"
"path/filepath"
"testing"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
// writeFile creates path with parents and writes contents.
func writeFile(t *testing.T, path, contents string) {
t.Helper()
require.NoError(t, os.MkdirAll(filepath.Dir(path), 0o755))
require.NoError(t, os.WriteFile(path, []byte(contents), 0o644))
}
// TestReadHwmonFans verifies the /sys/class/hwmon walker:
// - picks up fan*_input from every chip,
// - keys entries by chip name + sensor label (or fan idx if no label),
// - retains 0 RPM for stopped fans,
// - tolerates chips with no fan files at all.
func TestReadHwmonFans(t *testing.T) {
root := t.TempDir()
// hwmon0: Raspberry Pi 5 active cooler — one fan, no label.
writeFile(t, filepath.Join(root, "hwmon0", "name"), "pwmfan\n")
writeFile(t, filepath.Join(root, "hwmon0", "fan1_input"), "6500\n")
// hwmon1: a thermal-only chip, no fan files. Must not error.
writeFile(t, filepath.Join(root, "hwmon1", "name"), "cpu_thermal\n")
writeFile(t, filepath.Join(root, "hwmon1", "temp1_input"), "55000\n")
// hwmon2: two fans — one stopped (0 RPM) and one labeled "chassis".
writeFile(t, filepath.Join(root, "hwmon2", "name"), "nct6798\n")
writeFile(t, filepath.Join(root, "hwmon2", "fan1_input"), "0\n")
writeFile(t, filepath.Join(root, "hwmon2", "fan2_input"), "1200\n")
writeFile(t, filepath.Join(root, "hwmon2", "fan2_label"), "chassis\n")
fans, err := readHwmonFans(root)
require.NoError(t, err)
assert.Equal(t, map[string]uint16{
"pwmfan_fan1": 6500,
"nct6798_fan1": 0,
"nct6798_chassis": 1200,
}, fans)
}
// TestReadHwmonFansLegacyParent verifies legacy hwmon layouts such as applesmc,
// where the hwmon class node exists but fan attributes live on hwmonN/device.
func TestReadHwmonFansLegacyParent(t *testing.T) {
root := t.TempDir()
deviceDir := filepath.Join(root, "devices", "applesmc.768")
writeFile(t, filepath.Join(deviceDir, "name"), "applesmc\n")
writeFile(t, filepath.Join(deviceDir, "fan1_input"), "1202\n")
writeFile(t, filepath.Join(deviceDir, "fan1_label"), "Exhaust\n")
chipDir := filepath.Join(root, "hwmon1")
require.NoError(t, os.MkdirAll(chipDir, 0o755))
require.NoError(t, os.Symlink(deviceDir, filepath.Join(chipDir, "device")))
fans, err := readHwmonFans(root)
require.NoError(t, err)
assert.Equal(t, map[string]uint16{"applesmc_Exhaust": 1202}, fans)
}
// TestReadHwmonFansMissingRoot returns an error rather than panicking when the
// hwmon root doesn't exist (e.g. running on a kernel without hwmon support).
func TestReadHwmonFansMissingRoot(t *testing.T) {
_, err := readHwmonFans(filepath.Join(t.TempDir(), "does-not-exist"))
assert.Error(t, err)
}
// TestReadHwmonFansEmpty returns an empty map (not nil error) when the root
// exists but contains no chips at all.
func TestReadHwmonFansEmpty(t *testing.T) {
root := t.TempDir()
fans, err := readHwmonFans(root)
require.NoError(t, err)
assert.Empty(t, fans)
}
func TestFanDiscoveryCache(t *testing.T) {
root := t.TempDir()
input := filepath.Join(root, "hwmon0", "fan1_input")
writeFile(t, filepath.Join(root, "hwmon0", "name"), "chip\n")
writeFile(t, input, "1000\n")
getSensors := newFanSensorCache(root)
sensors, err := getSensors()
require.NoError(t, err)
fans := readFanSensors(sensors)
assert.Equal(t, uint16(1000), fans["chip_fan1"])
writeFile(t, input, "1200\n")
writeFile(t, filepath.Join(root, "hwmon0", "fan1_label"), "case\n")
sensors, err = getSensors()
require.NoError(t, err)
fans = readFanSensors(sensors)
assert.Equal(t, map[string]uint16{"chip_fan1": 1200}, fans)
}
-90
View File
@@ -1,90 +0,0 @@
package agent
import (
"crypto/sha256"
"encoding/hex"
"errors"
"os"
"path/filepath"
"strings"
"github.com/shirou/gopsutil/v4/cpu"
"github.com/shirou/gopsutil/v4/host"
)
const fingerprintFileName = "fingerprint"
// knownBadUUID is a commonly known "product_uuid" that is not unique across systems.
const knownBadUUID = "03000200-0400-0500-0006-000700080009"
// GetFingerprint returns the agent fingerprint. It first tries to read a saved
// fingerprint from the data directory. If not found (or dataDir is empty), it
// generates one from system properties. The hostname and cpuModel parameters are
// used as fallback material if host.HostID() fails. If either is empty, they
// are fetched from the system automatically.
//
// If a new fingerprint is generated and a dataDir is provided, it is saved.
func GetFingerprint(dataDir, hostname, cpuModel string) string {
if dataDir != "" {
if fp, err := readFingerprint(dataDir); err == nil {
return fp
}
}
fp := generateFingerprint(hostname, cpuModel)
if dataDir != "" {
_ = SaveFingerprint(dataDir, fp)
}
return fp
}
// generateFingerprint creates a fingerprint from system properties.
// It tries host.HostID() first, falling back to hostname + cpuModel.
// If hostname or cpuModel are empty, they are fetched from the system.
func generateFingerprint(hostname, cpuModel string) string {
fingerprint, err := host.HostID()
if err != nil || fingerprint == "" || fingerprint == knownBadUUID {
if hostname == "" {
hostname, _ = os.Hostname()
}
if cpuModel == "" {
if info, err := cpu.Info(); err == nil && len(info) > 0 {
cpuModel = info[0].ModelName
}
if cpuModel == "" {
cpuModel = getCpuModelFromCpuinfo()
}
}
fingerprint = hostname + cpuModel
}
sum := sha256.Sum256([]byte(fingerprint))
return hex.EncodeToString(sum[:24])
}
// readFingerprint reads the saved fingerprint from the data directory.
func readFingerprint(dataDir string) (string, error) {
fp, err := os.ReadFile(filepath.Join(dataDir, fingerprintFileName))
if err != nil {
return "", err
}
s := strings.TrimSpace(string(fp))
if s == "" {
return "", errors.New("fingerprint file is empty")
}
return s, nil
}
// SaveFingerprint writes the fingerprint to the data directory.
func SaveFingerprint(dataDir, fingerprint string) error {
return os.WriteFile(filepath.Join(dataDir, fingerprintFileName), []byte(fingerprint), 0o644)
}
// DeleteFingerprint removes the saved fingerprint file from the data directory.
// Returns nil if the file does not exist (idempotent).
func DeleteFingerprint(dataDir string) error {
err := os.Remove(filepath.Join(dataDir, fingerprintFileName))
if errors.Is(err, os.ErrNotExist) {
return nil
}
return err
}
-102
View File
@@ -1,102 +0,0 @@
//go:build testing
package agent
import (
"os"
"path/filepath"
"testing"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
func TestGetFingerprint(t *testing.T) {
t.Run("reads existing fingerprint from file", func(t *testing.T) {
dir := t.TempDir()
expected := "abc123def456"
err := os.WriteFile(filepath.Join(dir, fingerprintFileName), []byte(expected), 0644)
require.NoError(t, err)
fp := GetFingerprint(dir, "", "")
assert.Equal(t, expected, fp)
})
t.Run("trims whitespace from file", func(t *testing.T) {
dir := t.TempDir()
err := os.WriteFile(filepath.Join(dir, fingerprintFileName), []byte(" abc123 \n"), 0644)
require.NoError(t, err)
fp := GetFingerprint(dir, "", "")
assert.Equal(t, "abc123", fp)
})
t.Run("generates fingerprint when file does not exist", func(t *testing.T) {
dir := t.TempDir()
fp := GetFingerprint(dir, "", "")
assert.NotEmpty(t, fp)
})
t.Run("generates fingerprint when dataDir is empty", func(t *testing.T) {
fp := GetFingerprint("", "", "")
assert.NotEmpty(t, fp)
})
t.Run("generates consistent fingerprint for same inputs", func(t *testing.T) {
fp1 := GetFingerprint("", "myhost", "mycpu")
fp2 := GetFingerprint("", "myhost", "mycpu")
assert.Equal(t, fp1, fp2)
})
t.Run("prefers saved fingerprint over generated", func(t *testing.T) {
dir := t.TempDir()
require.NoError(t, SaveFingerprint(dir, "saved-fp"))
fp := GetFingerprint(dir, "anyhost", "anycpu")
assert.Equal(t, "saved-fp", fp)
})
}
func TestSaveFingerprint(t *testing.T) {
t.Run("saves fingerprint to file", func(t *testing.T) {
dir := t.TempDir()
err := SaveFingerprint(dir, "abc123")
require.NoError(t, err)
content, err := os.ReadFile(filepath.Join(dir, fingerprintFileName))
require.NoError(t, err)
assert.Equal(t, "abc123", string(content))
})
t.Run("overwrites existing fingerprint", func(t *testing.T) {
dir := t.TempDir()
require.NoError(t, SaveFingerprint(dir, "old"))
require.NoError(t, SaveFingerprint(dir, "new"))
content, err := os.ReadFile(filepath.Join(dir, fingerprintFileName))
require.NoError(t, err)
assert.Equal(t, "new", string(content))
})
}
func TestDeleteFingerprint(t *testing.T) {
t.Run("deletes existing fingerprint", func(t *testing.T) {
dir := t.TempDir()
fp := filepath.Join(dir, fingerprintFileName)
err := os.WriteFile(fp, []byte("abc123"), 0644)
require.NoError(t, err)
err = DeleteFingerprint(dir)
require.NoError(t, err)
// Verify file is gone
_, err = os.Stat(fp)
assert.True(t, os.IsNotExist(err))
})
t.Run("no error when file does not exist", func(t *testing.T) {
dir := t.TempDir()
err := DeleteFingerprint(dir)
assert.NoError(t, err)
})
}
+117 -402
View File
@@ -5,30 +5,29 @@ import (
"bytes" "bytes"
"encoding/json" "encoding/json"
"fmt" "fmt"
"log/slog"
"maps" "maps"
"os/exec" "os/exec"
"regexp" "regexp"
"runtime"
"strconv" "strconv"
"strings" "strings"
"sync" "sync"
"time" "time"
"github.com/henrygd/beszel/agent/utils"
"github.com/henrygd/beszel/internal/entities/system" "github.com/henrygd/beszel/internal/entities/system"
"golang.org/x/exp/slog"
) )
const ( const (
// Commands // Commands
nvidiaSmiCmd string = "nvidia-smi" nvidiaSmiCmd string = "nvidia-smi"
rocmSmiCmd string = "rocm-smi" rocmSmiCmd string = "rocm-smi"
tegraStatsCmd string = "tegrastats" tegraStatsCmd string = "tegrastats"
nvtopCmd string = "nvtop"
powermetricsCmd string = "powermetrics"
macmonCmd string = "macmon"
noGPUFoundMsg string = "no GPU found - see https://beszel.dev/guide/gpu"
// Polling intervals
nvidiaSmiInterval string = "4" // in seconds
tegraStatsInterval string = "3700" // in milliseconds
rocmSmiInterval time.Duration = 4300 * time.Millisecond
// Command retry and timeout constants // Command retry and timeout constants
retryWaitTime time.Duration = 5 * time.Second retryWaitTime time.Duration = 5 * time.Second
maxFailureRetries int = 5 maxFailureRetries int = 5
@@ -41,15 +40,18 @@ const (
// GPUManager manages data collection for GPUs (either Nvidia or AMD) // GPUManager manages data collection for GPUs (either Nvidia or AMD)
type GPUManager struct { type GPUManager struct {
sync.Mutex sync.Mutex
GpuDataMap map[string]*system.GPUData nvidiaSmi bool
rocmSmi bool
tegrastats bool
intelGpuStats bool
nvml bool
GpuDataMap map[string]*system.GPUData
// lastAvgData stores the last calculated averages for each GPU // lastAvgData stores the last calculated averages for each GPU
// Used when a collection happens before new data arrives (Count == 0) // Used when a collection happens before new data arrives (Count == 0)
lastAvgData map[string]system.GPUData lastAvgData map[string]system.GPUData
// Per-cache-key tracking for delta calculations // Per-cache-key tracking for delta calculations
// cacheKey -> gpuId -> snapshot of last count/usage/power values // cacheKey -> gpuId -> snapshot of last count/usage/power values
lastSnapshots map[uint16]map[string]*gpuSnapshot lastSnapshots map[uint16]map[string]*gpuSnapshot
// Per-card energy snapshots for Intel sysfs power calculation.
intelSysfsEnergySnapshots map[string]intelSysfsEnergySnapshot
} }
// gpuSnapshot stores the last observed incremental values for delta tracking // gpuSnapshot stores the last observed incremental values for delta tracking
@@ -84,62 +86,6 @@ type gpuCollector struct {
var errNoValidData = fmt.Errorf("no valid GPU data found") // Error for missing data var errNoValidData = fmt.Errorf("no valid GPU data found") // Error for missing data
// collectorSource identifies a selectable GPU collector in GPU_COLLECTOR.
type collectorSource string
const (
collectorSourceNVTop collectorSource = collectorSource(nvtopCmd)
collectorSourceNVML collectorSource = "nvml"
collectorSourceNvidiaSMI collectorSource = collectorSource(nvidiaSmiCmd)
collectorSourceIntelGpuTop collectorSource = collectorSource(intelGpuStatsCmd)
collectorSourceIntelSysfs collectorSource = "intel_sysfs"
collectorSourceAmdSysfs collectorSource = "amd_sysfs"
collectorSourceRocmSMI collectorSource = collectorSource(rocmSmiCmd)
collectorSourceMacmon collectorSource = collectorSource(macmonCmd)
collectorSourcePowermetrics collectorSource = collectorSource(powermetricsCmd)
collectorGroupNvidia string = "nvidia"
collectorGroupIntel string = "intel"
collectorGroupAmd string = "amd"
collectorGroupApple string = "apple"
)
func isValidCollectorSource(source collectorSource) bool {
switch source {
case collectorSourceNVTop,
collectorSourceNVML,
collectorSourceNvidiaSMI,
collectorSourceIntelGpuTop,
collectorSourceIntelSysfs,
collectorSourceAmdSysfs,
collectorSourceRocmSMI,
collectorSourceMacmon,
collectorSourcePowermetrics:
return true
}
return false
}
// gpuCapabilities describes detected GPU tooling and sysfs support on the host.
type gpuCapabilities struct {
hasNvidiaSmi bool
hasRocmSmi bool
hasAmdSysfs bool
hasTegrastats bool
hasIntelGpuTop bool
hasXe bool
hasIntelSysfs bool
hasNvtop bool
hasMacmon bool
hasPowermetrics bool
}
type collectorDefinition struct {
group string
available bool
start func(onFailure func()) bool
deprecationWarning string
}
// starts and manages the ongoing collection of GPU data for the specified GPU management utility // starts and manages the ongoing collection of GPU data for the specified GPU management utility
func (c *gpuCollector) start() { func (c *gpuCollector) start() {
for { for {
@@ -191,10 +137,10 @@ func (gm *GPUManager) getJetsonParser() func(output []byte) bool {
// use closure to avoid recompiling the regex // use closure to avoid recompiling the regex
ramPattern := regexp.MustCompile(`RAM (\d+)/(\d+)MB`) ramPattern := regexp.MustCompile(`RAM (\d+)/(\d+)MB`)
gr3dPattern := regexp.MustCompile(`GR3D_FREQ (\d+)%`) gr3dPattern := regexp.MustCompile(`GR3D_FREQ (\d+)%`)
tempPattern := regexp.MustCompile(`(?:tj|GPU)@(\d+\.?\d*)C`) tempPattern := regexp.MustCompile(`tj@(\d+\.?\d*)C`)
// Orin Nano / NX do not have GPU specific power monitor // Orin Nano / NX do not have GPU specific power monitor
// TODO: Maybe use VDD_IN for Nano / NX and add a total system power chart // TODO: Maybe use VDD_IN for Nano / NX and add a total system power chart
powerPattern := regexp.MustCompile(`(GPU_SOC|CPU_GPU_CV)\s+(\d+)mW|VDD_SYS_GPU\s+(\d+)/\d+`) powerPattern := regexp.MustCompile(`(GPU_SOC|CPU_GPU_CV) (\d+)mW`)
// jetson devices have only one gpu so we'll just initialize here // jetson devices have only one gpu so we'll just initialize here
gpuData := &system.GPUData{Name: "GPU"} gpuData := &system.GPUData{Name: "GPU"}
@@ -223,13 +169,7 @@ func (gm *GPUManager) getJetsonParser() func(output []byte) bool {
// Parse power usage // Parse power usage
powerMatches := powerPattern.FindSubmatch(output) powerMatches := powerPattern.FindSubmatch(output)
if powerMatches != nil { if powerMatches != nil {
// powerMatches[2] is the "(GPU_SOC|CPU_GPU_CV) <N>mW" capture power, _ := strconv.ParseFloat(string(powerMatches[2]), 64)
// powerMatches[3] is the "VDD_SYS_GPU <N>/<N>" capture
powerStr := string(powerMatches[2])
if powerStr == "" {
powerStr = string(powerMatches[3])
}
power, _ := strconv.ParseFloat(powerStr, 64)
gpuData.Power += power / milliwattsInAWatt gpuData.Power += power / milliwattsInAWatt
} }
gpuData.Count++ gpuData.Count++
@@ -292,14 +232,13 @@ func (gm *GPUManager) parseAmdData(output []byte) bool {
totalMemory, _ := strconv.ParseFloat(v.MemoryTotal, 64) totalMemory, _ := strconv.ParseFloat(v.MemoryTotal, 64)
usage, _ := strconv.ParseFloat(v.Usage, 64) usage, _ := strconv.ParseFloat(v.Usage, 64)
id := v.ID if _, ok := gm.GpuDataMap[v.ID]; !ok {
if _, ok := gm.GpuDataMap[id]; !ok { gm.GpuDataMap[v.ID] = &system.GPUData{Name: v.Name}
gm.GpuDataMap[id] = &system.GPUData{Name: v.Name}
} }
gpu := gm.GpuDataMap[id] gpu := gm.GpuDataMap[v.ID]
gpu.Temperature, _ = strconv.ParseFloat(v.Temperature, 64) gpu.Temperature, _ = strconv.ParseFloat(v.Temperature, 64)
gpu.MemoryUsed = utils.BytesToMegabytes(memoryUsage) gpu.MemoryUsed = bytesToMegabytes(memoryUsage)
gpu.MemoryTotal = utils.BytesToMegabytes(totalMemory) gpu.MemoryTotal = bytesToMegabytes(totalMemory)
gpu.Usage += usage gpu.Usage += usage
gpu.Power += power gpu.Power += power
gpu.Count++ gpu.Count++
@@ -361,33 +300,28 @@ func (gm *GPUManager) calculateGPUAverage(id string, gpu *system.GPUData, cacheK
// If no new data arrived // If no new data arrived
if deltaCount == 0 { if deltaCount == 0 {
// Only discrete GPUs report temp/memory, so treat all-zero as suspended (return zeros). // If GPU appears suspended (instantaneous values are 0), return zero values
// Engine-based (Intel) GPUs don't, so carry the last average forward across sample gaps. // Otherwise return last known average for temporary collection gaps
if gpu.Engines == nil && gpu.Temperature == 0 && gpu.MemoryUsed == 0 { if gpu.Temperature == 0 && gpu.MemoryUsed == 0 {
return system.GPUData{Name: gpu.Name} return system.GPUData{Name: gpu.Name}
} }
lastAvg := gm.lastAvgData[id] // zero value if not found return gm.lastAvgData[id] // zero value if not found
if lastAvg.Name == "" {
lastAvg.Name = gpu.Name
}
return lastAvg
} }
// Calculate new average // Calculate new average
gpuAvg := *gpu gpuAvg := *gpu
deltaUsage, deltaPower, deltaPowerPkg := gm.calculateDeltas(gpu, lastSnapshot) deltaUsage, deltaPower, deltaPowerPkg := gm.calculateDeltas(gpu, lastSnapshot)
gpuAvg.Power = utils.TwoDecimals(deltaPower / float64(deltaCount)) gpuAvg.Power = twoDecimals(deltaPower / float64(deltaCount))
gpuAvg.PowerPkg = utils.TwoDecimals(deltaPowerPkg / float64(deltaCount))
if gpu.Engines != nil { if gpu.Engines != nil {
// make fresh map for averaged engine metrics to avoid mutating // make fresh map for averaged engine metrics to avoid mutating
// the accumulator map stored in gm.GpuDataMap // the accumulator map stored in gm.GpuDataMap
gpuAvg.Engines = make(map[string]float64, len(gpu.Engines)) gpuAvg.Engines = make(map[string]float64, len(gpu.Engines))
gpuAvg.Usage = gm.calculateIntelGPUUsage(&gpuAvg, gpu, lastSnapshot, deltaCount) gpuAvg.Usage = gm.calculateIntelGPUUsage(&gpuAvg, gpu, lastSnapshot, deltaCount)
gpuAvg.PowerPkg = twoDecimals(deltaPowerPkg / float64(deltaCount))
} else { } else {
gpuAvg.Usage = utils.TwoDecimals(deltaUsage / float64(deltaCount)) gpuAvg.Usage = twoDecimals(deltaUsage / float64(deltaCount))
} }
gm.lastAvgData[id] = gpuAvg gm.lastAvgData[id] = gpuAvg
@@ -422,17 +356,17 @@ func (gm *GPUManager) calculateIntelGPUUsage(gpuAvg, gpu *system.GPUData, lastSn
} else { } else {
deltaEngine = engine deltaEngine = engine
} }
gpuAvg.Engines[name] = utils.TwoDecimals(deltaEngine / float64(deltaCount)) gpuAvg.Engines[name] = twoDecimals(deltaEngine / float64(deltaCount))
maxEngineUsage = max(maxEngineUsage, deltaEngine/float64(deltaCount)) maxEngineUsage = max(maxEngineUsage, deltaEngine/float64(deltaCount))
} }
return utils.TwoDecimals(maxEngineUsage) return twoDecimals(maxEngineUsage)
} }
// updateInstantaneousValues updates values that should reflect current state, not averages // updateInstantaneousValues updates values that should reflect current state, not averages
func (gm *GPUManager) updateInstantaneousValues(gpuAvg *system.GPUData, gpu *system.GPUData) { func (gm *GPUManager) updateInstantaneousValues(gpuAvg *system.GPUData, gpu *system.GPUData) {
gpuAvg.Temperature = utils.TwoDecimals(gpu.Temperature) gpuAvg.Temperature = twoDecimals(gpu.Temperature)
gpuAvg.MemoryUsed = utils.TwoDecimals(gpu.MemoryUsed) gpuAvg.MemoryUsed = twoDecimals(gpu.MemoryUsed)
gpuAvg.MemoryTotal = utils.TwoDecimals(gpu.MemoryTotal) gpuAvg.MemoryTotal = twoDecimals(gpu.MemoryTotal)
} }
// storeSnapshot saves the current GPU state for this cache key // storeSnapshot saves the current GPU state for this cache key
@@ -450,337 +384,118 @@ func (gm *GPUManager) storeSnapshot(id string, gpu *system.GPUData, cacheKey uin
gm.lastSnapshots[cacheKey][id] = snapshot gm.lastSnapshots[cacheKey][id] = snapshot
} }
// discoverGpuCapabilities checks for available GPU tooling and sysfs support. // detectGPUs checks for the presence of GPU management tools (nvidia-smi, rocm-smi, tegrastats)
// It only reports capability presence and does not apply policy decisions. // in the system path. It sets the corresponding flags in the GPUManager struct if any of these
func (gm *GPUManager) discoverGpuCapabilities() gpuCapabilities { // tools are found. If none of the tools are found, it returns an error indicating that no GPU
caps := gpuCapabilities{ // management tools are available.
hasAmdSysfs: gm.hasAmdSysfs(), func (gm *GPUManager) detectGPUs() error {
hasXe: gm.hasXe(),
hasIntelSysfs: gm.hasIntelSysfs(),
}
if _, err := exec.LookPath(nvidiaSmiCmd); err == nil { if _, err := exec.LookPath(nvidiaSmiCmd); err == nil {
caps.hasNvidiaSmi = true gm.nvidiaSmi = true
} }
if _, err := exec.LookPath(rocmSmiCmd); err == nil { if _, err := exec.LookPath(rocmSmiCmd); err == nil {
caps.hasRocmSmi = true gm.rocmSmi = true
} }
if _, err := exec.LookPath(tegraStatsCmd); err == nil { if _, err := exec.LookPath(tegraStatsCmd); err == nil {
caps.hasTegrastats = true gm.tegrastats = true
gm.nvidiaSmi = false
} }
if _, err := exec.LookPath(intelGpuStatsCmd); err == nil { if _, err := exec.LookPath(intelGpuStatsCmd); err == nil {
caps.hasIntelGpuTop = true gm.intelGpuStats = true
} }
if _, err := exec.LookPath(nvtopCmd); err == nil { if gm.nvidiaSmi || gm.rocmSmi || gm.tegrastats || gm.intelGpuStats || gm.nvml {
caps.hasNvtop = true return nil
} }
if runtime.GOOS == "darwin" { return fmt.Errorf("no GPU found - install nvidia-smi, rocm-smi, tegrastats, or intel_gpu_top")
if _, err := utils.LookPathHomebrew(macmonCmd); err == nil {
caps.hasMacmon = true
}
if _, err := exec.LookPath(powermetricsCmd); err == nil {
caps.hasPowermetrics = true
}
}
return caps
} }
func hasAnyGpuCollector(caps gpuCapabilities) bool { // startCollector starts the appropriate GPU data collector based on the command
return caps.hasNvidiaSmi || caps.hasRocmSmi || caps.hasAmdSysfs || caps.hasTegrastats || caps.hasIntelGpuTop || caps.hasIntelSysfs || caps.hasNvtop || caps.hasMacmon || caps.hasPowermetrics func (gm *GPUManager) startCollector(command string) {
}
func (gm *GPUManager) startIntelCollector() {
go func() {
failures := 0
for {
if err := gm.collectIntelStats(); err != nil {
failures++
if failures > maxFailureRetries {
break
}
slog.Warn("Error collecting Intel GPU data; see https://beszel.dev/guide/gpu", "err", err)
time.Sleep(retryWaitTime)
continue
}
}
}()
}
func (gm *GPUManager) startNvidiaSmiCollector(intervalSeconds string) {
collector := gpuCollector{ collector := gpuCollector{
name: nvidiaSmiCmd, name: command,
bufSize: 10 * 1024, bufSize: 10 * 1024,
cmdArgs: []string{ }
"-l", intervalSeconds, switch command {
case intelGpuStatsCmd:
go func() {
failures := 0
for {
if err := gm.collectIntelStats(); err != nil {
failures++
if failures > maxFailureRetries {
break
}
slog.Warn("Error collecting Intel GPU data; see https://beszel.dev/guide/gpu", "err", err)
time.Sleep(retryWaitTime)
continue
}
}
}()
case nvidiaSmiCmd:
collector.cmdArgs = []string{
"-l", nvidiaSmiInterval,
"--query-gpu=index,name,temperature.gpu,memory.used,memory.total,utilization.gpu,power.draw", "--query-gpu=index,name,temperature.gpu,memory.used,memory.total,utilization.gpu,power.draw",
"--format=csv,noheader,nounits", "--format=csv,noheader,nounits",
}, }
parse: gm.parseNvidiaData, collector.parse = gm.parseNvidiaData
} go collector.start()
go collector.start() case tegraStatsCmd:
} collector.cmdArgs = []string{"--interval", tegraStatsInterval}
collector.parse = gm.getJetsonParser()
func (gm *GPUManager) startTegraStatsCollector(intervalMilliseconds string) { go collector.start()
collector := gpuCollector{ case rocmSmiCmd:
name: tegraStatsCmd, collector.cmdArgs = []string{"--showid", "--showtemp", "--showuse", "--showpower", "--showproductname", "--showmeminfo", "vram", "--json"}
bufSize: 10 * 1024, collector.parse = gm.parseAmdData
cmdArgs: []string{"--interval", intervalMilliseconds}, go func() {
parse: gm.getJetsonParser(), failures := 0
} for {
go collector.start() if err := collector.collect(); err != nil {
} failures++
if failures > maxFailureRetries {
func (gm *GPUManager) startRocmSmiCollector(pollInterval time.Duration) { break
collector := gpuCollector{ }
name: rocmSmiCmd, slog.Warn("Error collecting AMD GPU data", "err", err)
bufSize: 10 * 1024,
cmdArgs: []string{"--showid", "--showtemp", "--showuse", "--showpower", "--showproductname", "--showmeminfo", "vram", "--json"},
parse: gm.parseAmdData,
}
go func() {
failures := 0
for {
if err := collector.collect(); err != nil {
failures++
if failures > maxFailureRetries {
break
} }
slog.Warn("Error collecting AMD GPU data via rocm-smi", "err", err) time.Sleep(rocmSmiInterval)
} }
time.Sleep(pollInterval) }()
}
}()
}
func (gm *GPUManager) collectorDefinitions(caps gpuCapabilities) map[collectorSource]collectorDefinition {
return map[collectorSource]collectorDefinition{
collectorSourceNVML: {
group: collectorGroupNvidia,
available: true,
start: func(_ func()) bool {
return gm.startNvmlCollector()
},
},
collectorSourceNvidiaSMI: {
group: collectorGroupNvidia,
available: caps.hasNvidiaSmi,
start: func(_ func()) bool {
gm.startNvidiaSmiCollector("4") // seconds
return true
},
},
collectorSourceIntelGpuTop: {
group: collectorGroupIntel,
available: caps.hasIntelGpuTop,
start: func(_ func()) bool {
gm.startIntelCollector()
return true
},
},
collectorSourceIntelSysfs: {
group: collectorGroupIntel,
available: caps.hasIntelSysfs,
start: func(_ func()) bool {
return gm.startIntelSysfsCollector()
},
},
collectorSourceAmdSysfs: {
group: collectorGroupAmd,
available: caps.hasAmdSysfs,
start: func(_ func()) bool {
return gm.startAmdSysfsCollector()
},
},
collectorSourceRocmSMI: {
group: collectorGroupAmd,
available: caps.hasRocmSmi,
deprecationWarning: "rocm-smi is deprecated and may be removed in a future release",
start: func(_ func()) bool {
gm.startRocmSmiCollector(4300 * time.Millisecond)
return true
},
},
collectorSourceNVTop: {
available: caps.hasNvtop,
start: func(onFailure func()) bool {
gm.startNvtopCollector("30", onFailure) // tens of milliseconds
return true
},
},
collectorSourceMacmon: {
group: collectorGroupApple,
available: caps.hasMacmon,
start: func(_ func()) bool {
gm.startMacmonCollector()
return true
},
},
collectorSourcePowermetrics: {
group: collectorGroupApple,
available: caps.hasPowermetrics,
start: func(_ func()) bool {
gm.startPowermetricsCollector()
return true
},
},
} }
} }
// parseCollectorPriority parses GPU_COLLECTOR and returns valid ordered entries.
func parseCollectorPriority(value string) []collectorSource {
parts := strings.Split(value, ",")
priorities := make([]collectorSource, 0, len(parts))
for _, raw := range parts {
name := collectorSource(strings.TrimSpace(strings.ToLower(raw)))
if !isValidCollectorSource(name) {
if name != "" {
slog.Warn("Ignoring unknown GPU collector", "collector", name)
}
continue
}
priorities = append(priorities, name)
}
return priorities
}
// startNvmlCollector initializes NVML and starts its polling loop.
func (gm *GPUManager) startNvmlCollector() bool {
collector := &nvmlCollector{gm: gm}
if err := collector.init(); err != nil {
slog.Warn("Failed to initialize NVML", "err", err)
return false
}
go collector.start()
return true
}
// startAmdSysfsCollector starts AMD GPU collection via sysfs.
func (gm *GPUManager) startAmdSysfsCollector() bool {
go func() {
if err := gm.collectAmdStats(); err != nil {
slog.Warn("Error collecting AMD GPU data via sysfs", "err", err)
}
}()
return true
}
// startCollectorsByPriority starts collectors in order with one source per vendor group.
func (gm *GPUManager) startCollectorsByPriority(priorities []collectorSource, caps gpuCapabilities) int {
definitions := gm.collectorDefinitions(caps)
selectedGroups := make(map[string]bool, 3)
started := 0
for i, source := range priorities {
definition, ok := definitions[source]
if !ok || !definition.available {
continue
}
// nvtop is not a vendor-specific collector, so should only be used if no other collectors are selected or it is first in GPU_COLLECTOR.
if source == collectorSourceNVTop {
if len(selectedGroups) > 0 {
slog.Warn("Skipping nvtop because other collectors are selected")
continue
}
// if nvtop fails, fall back to remaining collectors.
remaining := append([]collectorSource(nil), priorities[i+1:]...)
if definition.start(func() {
gm.startCollectorsByPriority(remaining, caps)
}) {
started++
return started
}
}
group := definition.group
if group == "" || selectedGroups[group] {
continue
}
if definition.deprecationWarning != "" {
slog.Warn(definition.deprecationWarning)
}
if definition.start(nil) {
selectedGroups[group] = true
started++
}
}
return started
}
// resolveLegacyCollectorPriority builds the default collector order when GPU_COLLECTOR is unset.
func (gm *GPUManager) resolveLegacyCollectorPriority(caps gpuCapabilities) []collectorSource {
priorities := make([]collectorSource, 0, 4)
if caps.hasNvidiaSmi && !caps.hasTegrastats {
if nvml, _ := utils.GetEnv("NVML"); nvml == "true" {
priorities = append(priorities, collectorSourceNVML, collectorSourceNvidiaSMI)
} else {
priorities = append(priorities, collectorSourceNvidiaSMI)
}
}
if caps.hasRocmSmi {
if val, _ := utils.GetEnv("AMD_SYSFS"); val == "true" {
priorities = append(priorities, collectorSourceAmdSysfs)
} else {
priorities = append(priorities, collectorSourceRocmSMI)
}
} else if caps.hasAmdSysfs {
priorities = append(priorities, collectorSourceAmdSysfs)
}
if caps.hasIntelGpuTop && !caps.hasXe {
priorities = append(priorities, collectorSourceIntelGpuTop)
}
if caps.hasIntelSysfs {
priorities = append(priorities, collectorSourceIntelSysfs)
}
// Apple collectors are currently opt-in only for testing.
// Enable them with GPU_COLLECTOR=macmon or GPU_COLLECTOR=powermetrics.
// TODO: uncomment below when Apple collectors are confirmed to be working.
//
// Prefer macmon on macOS (no sudo). Fall back to powermetrics if present.
// if caps.hasMacmon {
// priorities = append(priorities, collectorSourceMacmon)
// } else if caps.hasPowermetrics {
// priorities = append(priorities, collectorSourcePowermetrics)
// }
// Keep nvtop as a last resort only when no vendor collector exists.
if len(priorities) == 0 && caps.hasNvtop {
priorities = append(priorities, collectorSourceNVTop)
}
return priorities
}
// NewGPUManager creates and initializes a new GPUManager // NewGPUManager creates and initializes a new GPUManager
func NewGPUManager() (*GPUManager, error) { func NewGPUManager() (*GPUManager, error) {
if skipGPU, _ := utils.GetEnv("SKIP_GPU"); skipGPU == "true" { if skipGPU, _ := GetEnv("SKIP_GPU"); skipGPU == "true" {
return nil, nil return nil, nil
} }
var gm GPUManager var gm GPUManager
caps := gm.discoverGpuCapabilities() if err := gm.detectGPUs(); err != nil {
return nil, err
}
gm.GpuDataMap = make(map[string]*system.GPUData) gm.GpuDataMap = make(map[string]*system.GPUData)
// Jetson devices should always use tegrastats (ignore GPU_COLLECTOR). if gm.nvidiaSmi {
if caps.hasTegrastats { if nvml, _ := GetEnv("NVML"); nvml == "true" {
gm.startTegraStatsCollector("3700") gm.nvml = true
return &gm, nil gm.nvidiaSmi = false
} collector := &nvmlCollector{gm: &gm}
if err := collector.init(); err == nil {
// Respect explicit collector selection before capability auto-detection. go collector.start()
if collectorConfig, ok := utils.GetEnv("GPU_COLLECTOR"); ok && strings.TrimSpace(collectorConfig) != "" { } else {
priorities := parseCollectorPriority(collectorConfig) slog.Warn("Failed to initialize NVML, falling back to nvidia-smi", "err", err)
if gm.startCollectorsByPriority(priorities, caps) == 0 { gm.nvidiaSmi = true
return nil, fmt.Errorf("no configured GPU collectors are available") gm.startCollector(nvidiaSmiCmd)
}
} else {
gm.startCollector(nvidiaSmiCmd)
} }
return &gm, nil
} }
if gm.rocmSmi {
if !hasAnyGpuCollector(caps) { gm.startCollector(rocmSmiCmd)
return nil, fmt.Errorf(noGPUFoundMsg)
} }
if gm.tegrastats {
// auto-detect and start collectors when GPU_COLLECTOR is unset. gm.startCollector(tegraStatsCmd)
if gm.startCollectorsByPriority(gm.resolveLegacyCollectorPriority(caps), caps) == 0 { }
return nil, fmt.Errorf(noGPUFoundMsg) if gm.intelGpuStats {
gm.startCollector(intelGpuStatsCmd)
} }
return &gm, nil return &gm, nil
-303
View File
@@ -1,303 +0,0 @@
//go:build linux
package agent
import (
"bufio"
"fmt"
"log/slog"
"os"
"path/filepath"
"strconv"
"strings"
"sync"
"time"
"github.com/henrygd/beszel/agent/utils"
"github.com/henrygd/beszel/internal/entities/system"
)
var amdgpuNameCache = struct {
sync.RWMutex
hits map[string]string
misses map[string]struct{}
}{
hits: make(map[string]string),
misses: make(map[string]struct{}),
}
// hasAmdSysfs returns true if any AMD GPU sysfs nodes are found
func (gm *GPUManager) hasAmdSysfs() bool {
cards, err := filepath.Glob("/sys/class/drm/card*/device/vendor")
if err != nil {
return false
}
for _, vendorPath := range cards {
vendor, err := utils.ReadStringFileLimited(vendorPath, 64)
if err == nil && vendor == "0x1002" {
return true
}
}
return false
}
// collectAmdStats collects AMD GPU metrics directly from sysfs to avoid the overhead of rocm-smi
func (gm *GPUManager) collectAmdStats() error {
sysfsPollInterval := 3000 * time.Millisecond
cards, err := filepath.Glob("/sys/class/drm/card*")
if err != nil {
return err
}
var amdGpuPaths []string
for _, card := range cards {
// Ignore symbolic links and non-main card directories
if strings.Contains(filepath.Base(card), "-") || !isAmdGpu(card) {
continue
}
amdGpuPaths = append(amdGpuPaths, card)
}
if len(amdGpuPaths) == 0 {
return errNoValidData
}
slog.Debug("Using sysfs for AMD GPU data collection")
failures := 0
for {
hasData := false
for _, cardPath := range amdGpuPaths {
if gm.updateAmdGpuData(cardPath) {
hasData = true
}
}
if !hasData {
failures++
if failures > maxFailureRetries {
return errNoValidData
}
slog.Warn("No AMD GPU data from sysfs", "failures", failures)
time.Sleep(retryWaitTime)
continue
}
failures = 0
time.Sleep(sysfsPollInterval)
}
}
// isAmdGpu checks whether a DRM card path belongs to AMD vendor ID 0x1002.
func isAmdGpu(cardPath string) bool {
vendor, err := utils.ReadStringFileLimited(filepath.Join(cardPath, "device/vendor"), 64)
if err != nil {
return false
}
return vendor == "0x1002"
}
// updateAmdGpuData reads GPU metrics from sysfs and updates the GPU data map.
// Returns true if at least some data was successfully read.
func (gm *GPUManager) updateAmdGpuData(cardPath string) bool {
devicePath := filepath.Join(cardPath, "device")
id := filepath.Base(cardPath)
// Read all sysfs values first (no lock needed - these can be slow)
usage, usageErr := readSysfsFloat(filepath.Join(devicePath, "gpu_busy_percent"))
memUsed, memUsedErr := readSysfsFloat(filepath.Join(devicePath, "mem_info_vram_used"))
memTotal, _ := readSysfsFloat(filepath.Join(devicePath, "mem_info_vram_total"))
// if gtt is present, add it to the memory used and total (https://github.com/henrygd/beszel/issues/1569#issuecomment-3837640484)
if gttUsed, err := readSysfsFloat(filepath.Join(devicePath, "mem_info_gtt_used")); err == nil && gttUsed > 0 {
if gttTotal, err := readSysfsFloat(filepath.Join(devicePath, "mem_info_gtt_total")); err == nil {
memUsed += gttUsed
memTotal += gttTotal
}
}
var temp, power float64
hwmons, _ := filepath.Glob(filepath.Join(devicePath, "hwmon/hwmon*"))
for _, hwmonDir := range hwmons {
if t, err := readSysfsFloat(filepath.Join(hwmonDir, "temp1_input")); err == nil {
temp = t / 1000.0
}
if p, err := readSysfsFloat(filepath.Join(hwmonDir, "power1_average")); err == nil {
power += p / 1000000.0
} else if p, err := readSysfsFloat(filepath.Join(hwmonDir, "power1_input")); err == nil {
power += p / 1000000.0
}
}
// Check if we got any meaningful data
if usageErr != nil && memUsedErr != nil && temp == 0 {
return false
}
// Single lock to update all values atomically
gm.Lock()
defer gm.Unlock()
gpu, ok := gm.GpuDataMap[id]
if !ok {
gpu = &system.GPUData{Name: getAmdGpuName(devicePath)}
gm.GpuDataMap[id] = gpu
}
if usageErr == nil {
gpu.Usage += usage
}
gpu.MemoryUsed = utils.BytesToMegabytes(memUsed)
gpu.MemoryTotal = utils.BytesToMegabytes(memTotal)
gpu.Temperature = temp
gpu.Power += power
gpu.Count++
return true
}
// readSysfsFloat reads and parses a numeric value from a sysfs file.
func readSysfsFloat(path string) (float64, error) {
val, err := utils.ReadStringFileLimited(path, 64)
if err != nil {
slog.Debug("Failed to read sysfs value", "path", path, "error", err)
return 0, err
}
return strconv.ParseFloat(val, 64)
}
// normalizeHexID normalizes hex IDs by trimming spaces, lowercasing, and dropping 0x.
func normalizeHexID(id string) string {
return strings.TrimPrefix(strings.ToLower(strings.TrimSpace(id)), "0x")
}
// cacheKeyForAmdgpu builds the cache key for a device and optional revision.
func cacheKeyForAmdgpu(deviceID, revisionID string) string {
if revisionID != "" {
return deviceID + ":" + revisionID
}
return deviceID
}
// lookupAmdgpuNameInFile resolves an AMDGPU name from amdgpu.ids by device/revision.
func lookupAmdgpuNameInFile(deviceID, revisionID, filePath string) (name string, exact bool, found bool) {
file, err := os.Open(filePath)
if err != nil {
return "", false, false
}
defer file.Close()
var byDevice string
scanner := bufio.NewScanner(file)
for scanner.Scan() {
line := strings.TrimSpace(scanner.Text())
if line == "" || strings.HasPrefix(line, "#") {
continue
}
parts := strings.SplitN(line, ",", 3)
if len(parts) != 3 {
continue
}
dev := normalizeHexID(parts[0])
rev := normalizeHexID(parts[1])
productName := strings.TrimSpace(parts[2])
if dev == "" || productName == "" || dev != deviceID {
continue
}
if byDevice == "" {
byDevice = productName
}
if revisionID != "" && rev == revisionID {
return productName, true, true
}
}
if byDevice != "" {
return byDevice, false, true
}
return "", false, false
}
// getCachedAmdgpuName returns cached hit/miss status for the given device/revision.
func getCachedAmdgpuName(deviceID, revisionID string) (name string, found bool, done bool) {
// Build the list of cache keys to check. We always look up the exact device+revision key.
// When revisionID is set, we also look up deviceID alone, since the cache may store a
// device-only fallback when we couldn't resolve the exact revision.
keys := []string{cacheKeyForAmdgpu(deviceID, revisionID)}
if revisionID != "" {
keys = append(keys, deviceID)
}
knownMisses := 0
amdgpuNameCache.RLock()
defer amdgpuNameCache.RUnlock()
for _, key := range keys {
if name, ok := amdgpuNameCache.hits[key]; ok {
return name, true, true
}
if _, ok := amdgpuNameCache.misses[key]; ok {
knownMisses++
}
}
// done=true means "don't bother doing slow lookup": we either found a name (above) or
// every key we checked was already a known miss, so we've tried before and failed.
return "", false, knownMisses == len(keys)
}
// normalizeAmdgpuName trims standard suffixes from AMDGPU product names.
func normalizeAmdgpuName(name string) string {
for _, suffix := range []string{" Graphics", " Series"} {
name = strings.TrimSuffix(name, suffix)
}
return name
}
// cacheAmdgpuName stores a resolved AMDGPU name in the lookup cache.
func cacheAmdgpuName(deviceID, revisionID, name string, exact bool) {
name = normalizeAmdgpuName(name)
amdgpuNameCache.Lock()
defer amdgpuNameCache.Unlock()
if exact && revisionID != "" {
amdgpuNameCache.hits[cacheKeyForAmdgpu(deviceID, revisionID)] = name
}
amdgpuNameCache.hits[deviceID] = name
}
// cacheMissingAmdgpuName records unresolved device/revision lookups.
func cacheMissingAmdgpuName(deviceID, revisionID string) {
amdgpuNameCache.Lock()
defer amdgpuNameCache.Unlock()
amdgpuNameCache.misses[deviceID] = struct{}{}
if revisionID != "" {
amdgpuNameCache.misses[cacheKeyForAmdgpu(deviceID, revisionID)] = struct{}{}
}
}
// getAmdGpuName attempts to get a descriptive GPU name.
// First tries product_name (rarely available), then looks up the PCI device ID.
// Falls back to showing the raw device ID if not found in the lookup table.
func getAmdGpuName(devicePath string) string {
// Try product_name first (works for some enterprise GPUs)
if prod, err := utils.ReadStringFileLimited(filepath.Join(devicePath, "product_name"), 128); err == nil {
return prod
}
// Read PCI device ID and look it up
if deviceID, err := utils.ReadStringFileLimited(filepath.Join(devicePath, "device"), 64); err == nil {
id := normalizeHexID(deviceID)
revision := ""
if rev, revErr := utils.ReadStringFileLimited(filepath.Join(devicePath, "revision"), 64); revErr == nil {
revision = normalizeHexID(rev)
}
if name, found, done := getCachedAmdgpuName(id, revision); found {
return name
} else if !done {
if name, exact, ok := lookupAmdgpuNameInFile(id, revision, "/usr/share/libdrm/amdgpu.ids"); ok {
cacheAmdgpuName(id, revision, name, exact)
return normalizeAmdgpuName(name)
}
cacheMissingAmdgpuName(id, revision)
}
return fmt.Sprintf("AMD GPU (%s)", id)
}
return "AMD GPU"
}
-265
View File
@@ -1,265 +0,0 @@
//go:build linux
package agent
import (
"os"
"path/filepath"
"testing"
"github.com/henrygd/beszel/agent/utils"
"github.com/henrygd/beszel/internal/entities/system"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
func TestNormalizeHexID(t *testing.T) {
tests := []struct {
in string
want string
}{
{"0x1002", "1002"},
{"C2", "c2"},
{" 15BF ", "15bf"},
{"0x15bf", "15bf"},
{"", ""},
}
for _, tt := range tests {
subName := tt.in
if subName == "" {
subName = "empty_string"
}
t.Run(subName, func(t *testing.T) {
got := normalizeHexID(tt.in)
assert.Equal(t, tt.want, got)
})
}
}
func TestCacheKeyForAmdgpu(t *testing.T) {
tests := []struct {
deviceID string
revisionID string
want string
}{
{"1114", "c2", "1114:c2"},
{"15bf", "", "15bf"},
{"1506", "c1", "1506:c1"},
}
for _, tt := range tests {
got := cacheKeyForAmdgpu(tt.deviceID, tt.revisionID)
assert.Equal(t, tt.want, got)
}
}
func TestReadSysfsFloat(t *testing.T) {
dir := t.TempDir()
validPath := filepath.Join(dir, "val")
require.NoError(t, os.WriteFile(validPath, []byte(" 42.5 \n"), 0o644))
got, err := readSysfsFloat(validPath)
require.NoError(t, err)
assert.Equal(t, 42.5, got)
// Integer and scientific
sciPath := filepath.Join(dir, "sci")
require.NoError(t, os.WriteFile(sciPath, []byte("1e2"), 0o644))
got, err = readSysfsFloat(sciPath)
require.NoError(t, err)
assert.Equal(t, 100.0, got)
// Missing file
_, err = readSysfsFloat(filepath.Join(dir, "missing"))
require.Error(t, err)
// Invalid content
badPath := filepath.Join(dir, "bad")
require.NoError(t, os.WriteFile(badPath, []byte("not a number"), 0o644))
_, err = readSysfsFloat(badPath)
require.Error(t, err)
}
func TestIsAmdGpu(t *testing.T) {
dir := t.TempDir()
deviceDir := filepath.Join(dir, "device")
require.NoError(t, os.MkdirAll(deviceDir, 0o755))
// AMD vendor 0x1002 -> true
require.NoError(t, os.WriteFile(filepath.Join(deviceDir, "vendor"), []byte("0x1002\n"), 0o644))
assert.True(t, isAmdGpu(dir), "vendor 0x1002 should be AMD")
// Non-AMD vendor -> false
require.NoError(t, os.WriteFile(filepath.Join(deviceDir, "vendor"), []byte("0x10de\n"), 0o644))
assert.False(t, isAmdGpu(dir), "vendor 0x10de should not be AMD")
// Missing vendor file -> false
require.NoError(t, os.Remove(filepath.Join(deviceDir, "vendor")))
assert.False(t, isAmdGpu(dir), "missing vendor file should be false")
}
func TestAmdgpuNameCacheRoundTrip(t *testing.T) {
// Cache a name and retrieve it (unique key to avoid affecting other tests)
deviceID, revisionID := "cachedev99", "00"
cacheAmdgpuName(deviceID, revisionID, "AMD Test GPU 99 Graphics", true)
name, found, done := getCachedAmdgpuName(deviceID, revisionID)
assert.True(t, found)
assert.True(t, done)
assert.Equal(t, "AMD Test GPU 99", name)
// Device-only key also stored
name2, found2, _ := getCachedAmdgpuName(deviceID, "")
assert.True(t, found2)
assert.Equal(t, "AMD Test GPU 99", name2)
// Cache a miss
cacheMissingAmdgpuName("missedev99", "ab")
_, found3, done3 := getCachedAmdgpuName("missedev99", "ab")
assert.False(t, found3)
assert.True(t, done3, "done should be true so caller skips file lookup")
}
func TestUpdateAmdGpuDataWithFakeSysfs(t *testing.T) {
tests := []struct {
name string
writeGTT bool
wantMemoryUsed float64
wantMemoryTotal float64
}{
{
name: "sums vram and gtt when gtt is present",
writeGTT: true,
wantMemoryUsed: utils.BytesToMegabytes(1073741824 + 536870912),
wantMemoryTotal: utils.BytesToMegabytes(2147483648 + 4294967296),
},
{
name: "falls back to vram when gtt is missing",
writeGTT: false,
wantMemoryUsed: utils.BytesToMegabytes(1073741824),
wantMemoryTotal: utils.BytesToMegabytes(2147483648),
},
}
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
dir := t.TempDir()
cardPath := filepath.Join(dir, "card0")
devicePath := filepath.Join(cardPath, "device")
hwmonPath := filepath.Join(devicePath, "hwmon", "hwmon0")
require.NoError(t, os.MkdirAll(hwmonPath, 0o755))
write := func(name, content string) {
require.NoError(t, os.WriteFile(filepath.Join(devicePath, name), []byte(content), 0o644))
}
write("vendor", "0x1002")
write("device", "0x1506")
write("revision", "0xc1")
write("gpu_busy_percent", "25")
write("mem_info_vram_used", "1073741824")
write("mem_info_vram_total", "2147483648")
if tt.writeGTT {
write("mem_info_gtt_used", "536870912")
write("mem_info_gtt_total", "4294967296")
}
require.NoError(t, os.WriteFile(filepath.Join(hwmonPath, "temp1_input"), []byte("45000"), 0o644))
require.NoError(t, os.WriteFile(filepath.Join(hwmonPath, "power1_input"), []byte("20000000"), 0o644))
// Pre-cache name so getAmdGpuName returns a known value (it uses system amdgpu.ids path)
cacheAmdgpuName("1506", "c1", "AMD Radeon 610M Graphics", true)
gm := &GPUManager{GpuDataMap: make(map[string]*system.GPUData)}
ok := gm.updateAmdGpuData(cardPath)
require.True(t, ok)
gpu, ok := gm.GpuDataMap["card0"]
require.True(t, ok)
assert.Equal(t, "AMD Radeon 610M", gpu.Name)
assert.Equal(t, 25.0, gpu.Usage)
assert.Equal(t, tt.wantMemoryUsed, gpu.MemoryUsed)
assert.Equal(t, tt.wantMemoryTotal, gpu.MemoryTotal)
assert.Equal(t, 45.0, gpu.Temperature)
assert.Equal(t, 20.0, gpu.Power)
assert.Equal(t, 1.0, gpu.Count)
})
}
}
func TestLookupAmdgpuNameInFile(t *testing.T) {
idsPath := filepath.Join("test-data", "amdgpu.ids")
tests := []struct {
name string
deviceID string
revisionID string
wantName string
wantExact bool
wantFound bool
}{
{
name: "exact device and revision match",
deviceID: "1114",
revisionID: "c2",
wantName: "AMD Radeon 860M Graphics",
wantExact: true,
wantFound: true,
},
{
name: "exact match 15BF revision 01 returns 760M",
deviceID: "15bf",
revisionID: "01",
wantName: "AMD Radeon 760M Graphics",
wantExact: true,
wantFound: true,
},
{
name: "exact match 15BF revision 00 returns 780M",
deviceID: "15bf",
revisionID: "00",
wantName: "AMD Radeon 780M Graphics",
wantExact: true,
wantFound: true,
},
{
name: "device-only match returns first entry for device",
deviceID: "1506",
revisionID: "",
wantName: "AMD Radeon 610M",
wantExact: false,
wantFound: true,
},
{
name: "unknown device not found",
deviceID: "dead",
revisionID: "00",
wantName: "",
wantExact: false,
wantFound: false,
},
}
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
gotName, gotExact, gotFound := lookupAmdgpuNameInFile(tt.deviceID, tt.revisionID, idsPath)
assert.Equal(t, tt.wantName, gotName, "name")
assert.Equal(t, tt.wantExact, gotExact, "exact")
assert.Equal(t, tt.wantFound, gotFound, "found")
})
}
}
func TestGetAmdGpuNameFromIdsFile(t *testing.T) {
// Test that getAmdGpuName resolves a name when we can't inject the ids path.
// We only verify behavior when product_name is missing and device/revision
// would be read from sysfs; the actual lookup uses /usr/share/libdrm/amdgpu.ids.
// So this test focuses on normalizeAmdgpuName and that lookupAmdgpuNameInFile
// returns the expected name for our test-data file.
idsPath := filepath.Join("test-data", "amdgpu.ids")
name, exact, found := lookupAmdgpuNameInFile("1435", "ae", idsPath)
require.True(t, found)
require.True(t, exact)
assert.Equal(t, "AMD Custom GPU 0932", name)
assert.Equal(t, "AMD Custom GPU 0932", normalizeAmdgpuName(name))
// " Graphics" suffix is trimmed by normalizeAmdgpuName
name2 := "AMD Radeon 860M Graphics"
assert.Equal(t, "AMD Radeon 860M", normalizeAmdgpuName(name2))
}
-15
View File
@@ -1,15 +0,0 @@
//go:build !linux
package agent
import (
"errors"
)
func (gm *GPUManager) hasAmdSysfs() bool {
return false
}
func (gm *GPUManager) collectAmdStats() error {
return errors.ErrUnsupported
}
-257
View File
@@ -1,257 +0,0 @@
//go:build darwin
package agent
import (
"bufio"
"bytes"
"encoding/json"
"io"
"log/slog"
"os/exec"
"strconv"
"strings"
"time"
"github.com/henrygd/beszel/agent/utils"
"github.com/henrygd/beszel/internal/entities/system"
)
const (
// powermetricsSampleIntervalMs is the sampling interval passed to powermetrics (-i).
powermetricsSampleIntervalMs = 500
// powermetricsPollInterval is how often we run powermetrics to collect a new sample.
powermetricsPollInterval = 2 * time.Second
// macmonIntervalMs is the sampling interval passed to macmon pipe (-i), in milliseconds.
macmonIntervalMs = 2500
)
const appleGPUID = "0"
// startPowermetricsCollector runs powermetrics --samplers gpu_power in a loop and updates
// GPU usage and power. Requires root (sudo) on macOS. A single logical GPU is reported as id "0".
func (gm *GPUManager) startPowermetricsCollector() {
// Ensure single GPU entry for Apple GPU
if _, ok := gm.GpuDataMap[appleGPUID]; !ok {
gm.GpuDataMap[appleGPUID] = &system.GPUData{Name: "Apple GPU"}
}
go func() {
failures := 0
for {
if err := gm.collectPowermetrics(); err != nil {
failures++
if failures > maxFailureRetries {
slog.Warn("powermetrics GPU collector failed repeatedly, stopping", "err", err)
break
}
slog.Warn("Error collecting macOS GPU data via powermetrics (may require sudo)", "err", err)
time.Sleep(retryWaitTime)
continue
}
failures = 0
time.Sleep(powermetricsPollInterval)
}
}()
}
// collectPowermetrics runs powermetrics once and parses GPU usage and power from its output.
func (gm *GPUManager) collectPowermetrics() error {
interval := strconv.Itoa(powermetricsSampleIntervalMs)
cmd := exec.Command(powermetricsCmd, "--samplers", "gpu_power", "-i", interval, "-n", "1")
cmd.Stderr = nil
out, err := cmd.Output()
if err != nil {
return err
}
if !gm.parsePowermetricsData(out) {
return errNoValidData
}
return nil
}
// parsePowermetricsData parses powermetrics gpu_power output and updates GpuDataMap["0"].
// Example output:
//
// **** GPU usage ****
// GPU HW active frequency: 444 MHz
// GPU HW active residency: 0.97% (444 MHz: .97% ...
// GPU idle residency: 99.03%
// GPU Power: 4 mW
func (gm *GPUManager) parsePowermetricsData(output []byte) bool {
var idleResidency, powerMW float64
var gotIdle, gotPower bool
scanner := bufio.NewScanner(bytes.NewReader(output))
for scanner.Scan() {
line := strings.TrimSpace(scanner.Text())
if strings.HasPrefix(line, "GPU idle residency:") {
// "GPU idle residency: 99.03%"
fields := strings.Fields(strings.TrimPrefix(line, "GPU idle residency:"))
if len(fields) >= 1 {
pct := strings.TrimSuffix(fields[0], "%")
if v, err := strconv.ParseFloat(pct, 64); err == nil {
idleResidency = v
gotIdle = true
}
}
} else if strings.HasPrefix(line, "GPU Power:") {
// "GPU Power: 4 mW"
fields := strings.Fields(strings.TrimPrefix(line, "GPU Power:"))
if len(fields) >= 1 {
if v, err := strconv.ParseFloat(fields[0], 64); err == nil {
powerMW = v
gotPower = true
}
}
}
}
if err := scanner.Err(); err != nil {
return false
}
if !gotIdle && !gotPower {
return false
}
gm.Lock()
defer gm.Unlock()
if _, ok := gm.GpuDataMap[appleGPUID]; !ok {
gm.GpuDataMap[appleGPUID] = &system.GPUData{Name: "Apple GPU"}
}
gpu := gm.GpuDataMap[appleGPUID]
if gotIdle {
// Usage = 100 - idle residency (e.g. 100 - 99.03 = 0.97%)
gpu.Usage += 100 - idleResidency
}
if gotPower {
// mW -> W
gpu.Power += powerMW / milliwattsInAWatt
}
gpu.Count++
return true
}
// startMacmonCollector runs `macmon pipe` in a loop and parses one JSON object per line.
// This collector does not require sudo. A single logical GPU is reported as id "0".
func (gm *GPUManager) startMacmonCollector() {
if _, ok := gm.GpuDataMap[appleGPUID]; !ok {
gm.GpuDataMap[appleGPUID] = &system.GPUData{Name: "Apple GPU"}
}
go func() {
failures := 0
for {
if err := gm.collectMacmonPipe(); err != nil {
failures++
if failures > maxFailureRetries {
slog.Warn("macmon GPU collector failed repeatedly, stopping", "err", err)
break
}
slog.Warn("Error collecting macOS GPU data via macmon", "err", err)
time.Sleep(retryWaitTime)
continue
}
failures = 0
// `macmon pipe` is long-running; if it returns, wait a bit before restarting.
time.Sleep(retryWaitTime)
}
}()
}
type macmonTemp struct {
GPUTempAvg float64 `json:"gpu_temp_avg"`
}
type macmonSample struct {
GPUPower float64 `json:"gpu_power"` // watts (macmon reports fractional values)
GPURAMPower float64 `json:"gpu_ram_power"` // watts
GPUUsage []float64 `json:"gpu_usage"` // [freq_mhz, usage] where usage is typically 0..1
Temp macmonTemp `json:"temp"`
}
func (gm *GPUManager) collectMacmonPipe() (err error) {
macmonPath, err := utils.LookPathHomebrew(macmonCmd)
if err != nil {
return err
}
cmd := exec.Command(macmonPath, "pipe", "-i", strconv.Itoa(macmonIntervalMs))
// Avoid blocking if macmon writes to stderr.
cmd.Stderr = io.Discard
stdout, err := cmd.StdoutPipe()
if err != nil {
return err
}
if err := cmd.Start(); err != nil {
return err
}
// Ensure we always reap the child to avoid zombies on any return path and
// propagate a non-zero exit code if no other error was set.
defer func() {
_ = stdout.Close()
if cmd.ProcessState == nil || !cmd.ProcessState.Exited() {
_ = cmd.Process.Kill()
}
if waitErr := cmd.Wait(); err == nil && waitErr != nil {
err = waitErr
}
}()
scanner := bufio.NewScanner(stdout)
var hadSample bool
for scanner.Scan() {
line := bytes.TrimSpace(scanner.Bytes())
if len(line) == 0 {
continue
}
if gm.parseMacmonLine(line) {
hadSample = true
}
}
if scanErr := scanner.Err(); scanErr != nil {
return scanErr
}
if !hadSample {
return errNoValidData
}
return nil
}
// parseMacmonLine parses a single macmon JSON line and updates Apple GPU metrics.
func (gm *GPUManager) parseMacmonLine(line []byte) bool {
var sample macmonSample
if err := json.Unmarshal(line, &sample); err != nil {
return false
}
usage := 0.0
if len(sample.GPUUsage) >= 2 {
usage = sample.GPUUsage[1]
// Heuristic: macmon typically reports 0..1; convert to percentage.
if usage <= 1.0 {
usage *= 100
}
}
// Consider the line valid if it contains at least one GPU metric.
if usage == 0 && sample.GPUPower == 0 && sample.Temp.GPUTempAvg == 0 {
return false
}
gm.Lock()
defer gm.Unlock()
gpu, ok := gm.GpuDataMap[appleGPUID]
if !ok {
gpu = &system.GPUData{Name: "Apple GPU"}
gm.GpuDataMap[appleGPUID] = gpu
}
gpu.Temperature = sample.Temp.GPUTempAvg
gpu.Usage += usage
// macmon reports power in watts; include VRAM power if present.
gpu.Power += sample.GPUPower + sample.GPURAMPower
gpu.Count++
return true
}
-81
View File
@@ -1,81 +0,0 @@
//go:build darwin
package agent
import (
"testing"
"github.com/henrygd/beszel/internal/entities/system"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
func TestParsePowermetricsData(t *testing.T) {
input := `
Machine model: Mac14,10
OS version: 25D125
*** Sampled system activity (Sat Feb 14 00:42:06 2026 -0500) (503.05ms elapsed) ***
**** GPU usage ****
GPU HW active frequency: 444 MHz
GPU HW active residency: 0.97% (444 MHz: .97% 612 MHz: 0% 808 MHz: 0% 968 MHz: 0% 1110 MHz: 0% 1236 MHz: 0% 1338 MHz: 0% 1398 MHz: 0%)
GPU SW requested state: (P1 : 100% P2 : 0% P3 : 0% P4 : 0% P5 : 0% P6 : 0% P7 : 0% P8 : 0%)
GPU idle residency: 99.03%
GPU Power: 4 mW
`
gm := &GPUManager{
GpuDataMap: make(map[string]*system.GPUData),
}
valid := gm.parsePowermetricsData([]byte(input))
require.True(t, valid)
g0, ok := gm.GpuDataMap["0"]
require.True(t, ok)
assert.Equal(t, "Apple GPU", g0.Name)
// Usage = 100 - 99.03 = 0.97
assert.InDelta(t, 0.97, g0.Usage, 0.01)
// 4 mW -> 0.004 W
assert.InDelta(t, 0.004, g0.Power, 0.0001)
assert.Equal(t, 1.0, g0.Count)
}
func TestParsePowermetricsDataPartial(t *testing.T) {
// Only power line (e.g. older macOS or different sampler output)
input := `
**** GPU usage ****
GPU Power: 120 mW
`
gm := &GPUManager{
GpuDataMap: make(map[string]*system.GPUData),
}
valid := gm.parsePowermetricsData([]byte(input))
require.True(t, valid)
g0, ok := gm.GpuDataMap["0"]
require.True(t, ok)
assert.Equal(t, "Apple GPU", g0.Name)
assert.InDelta(t, 0.12, g0.Power, 0.001)
assert.Equal(t, 1.0, g0.Count)
}
func TestParseMacmonLine(t *testing.T) {
input := `{"all_power":0.6468324661254883,"ane_power":0.0,"cpu_power":0.6359732151031494,"ecpu_usage":[2061,0.1726151406764984],"gpu_power":0.010859241709113121,"gpu_ram_power":0.000965250947047025,"gpu_usage":[503,0.013633215799927711],"memory":{"ram_total":17179869184,"ram_usage":12322914304,"swap_total":0,"swap_usage":0},"pcpu_usage":[1248,0.11792058497667313],"ram_power":0.14885640144348145,"sys_power":10.4955415725708,"temp":{"cpu_temp_avg":23.041261672973633,"gpu_temp_avg":29.44516944885254},"timestamp":"2026-02-17T19:34:27.942556+00:00"}`
gm := &GPUManager{
GpuDataMap: make(map[string]*system.GPUData),
}
valid := gm.parseMacmonLine([]byte(input))
require.True(t, valid)
g0, ok := gm.GpuDataMap["0"]
require.True(t, ok)
assert.Equal(t, "Apple GPU", g0.Name)
// macmon reports usage fraction 0..1; expect percent conversion.
assert.InDelta(t, 1.3633, g0.Usage, 0.05)
// power includes gpu_power + gpu_ram_power
assert.InDelta(t, 0.011824, g0.Power, 0.0005)
assert.InDelta(t, 29.445, g0.Temperature, 0.01)
assert.Equal(t, 1.0, g0.Count)
}
-9
View File
@@ -1,9 +0,0 @@
//go:build !darwin
package agent
// startPowermetricsCollector is a no-op on non-darwin platforms; the real implementation is in gpu_darwin.go.
func (gm *GPUManager) startPowermetricsCollector() {}
// startMacmonCollector is a no-op on non-darwin platforms; the real implementation is in gpu_darwin.go.
func (gm *GPUManager) startMacmonCollector() {}
+3 -5
View File
@@ -7,7 +7,6 @@ import (
"strconv" "strconv"
"strings" "strings"
"github.com/henrygd/beszel/agent/utils"
"github.com/henrygd/beszel/internal/entities/system" "github.com/henrygd/beszel/internal/entities/system"
) )
@@ -28,11 +27,10 @@ func (gm *GPUManager) updateIntelFromStats(sample *intelGpuStats) bool {
defer gm.Unlock() defer gm.Unlock()
// only one gpu for now - cmd doesn't provide all by default // only one gpu for now - cmd doesn't provide all by default
id := "i0" // prefix with i to avoid conflicts with nvidia card ids gpuData, ok := gm.GpuDataMap["0"]
gpuData, ok := gm.GpuDataMap[id]
if !ok { if !ok {
gpuData = &system.GPUData{Name: "GPU", Engines: make(map[string]float64)} gpuData = &system.GPUData{Name: "GPU", Engines: make(map[string]float64)}
gm.GpuDataMap[id] = gpuData gm.GpuDataMap["0"] = gpuData
} }
gpuData.Power += sample.PowerGPU gpuData.Power += sample.PowerGPU
@@ -53,7 +51,7 @@ func (gm *GPUManager) updateIntelFromStats(sample *intelGpuStats) bool {
func (gm *GPUManager) collectIntelStats() (err error) { func (gm *GPUManager) collectIntelStats() (err error) {
// Build command arguments, optionally selecting a device via -d // Build command arguments, optionally selecting a device via -d
args := []string{"-s", intelGpuStatsInterval, "-l"} args := []string{"-s", intelGpuStatsInterval, "-l"}
if dev, ok := utils.GetEnv("INTEL_GPU_DEVICE"); ok && dev != "" { if dev, ok := GetEnv("INTEL_GPU_DEVICE"); ok && dev != "" {
args = append(args, "-d", dev) args = append(args, "-d", dev)
} }
cmd := exec.Command(intelGpuStatsCmd, args...) cmd := exec.Command(intelGpuStatsCmd, args...)
-280
View File
@@ -1,280 +0,0 @@
//go:build linux
package agent
import (
"fmt"
"log/slog"
"os"
"path/filepath"
"strconv"
"strings"
"time"
"github.com/henrygd/beszel/agent/utils"
"github.com/henrygd/beszel/internal/entities/system"
)
var (
drmSysfsRoot = "/sys/class/drm"
intelSysfsNow = time.Now
)
type intelSysfsEnergySnapshot struct {
microjoules uint64
timestamp time.Time
}
type intelSysfsCard struct {
cardPath string
hwmonDir string
}
// hasIntelSysfs returns true if any Intel DRM card exposes an hwmon energy counter.
func (gm *GPUManager) hasIntelSysfs() bool {
cards, err := discoverIntelSysfsCards()
return err == nil && len(cards) > 0
}
// startIntelSysfsCollector starts Intel GPU collection via sysfs.
func (gm *GPUManager) startIntelSysfsCollector() bool {
go func() {
if err := gm.collectIntelSysfsStats(); err != nil {
slog.Warn("Error collecting Intel GPU data via sysfs", "err", err)
}
}()
return true
}
// collectIntelSysfsStats collects Intel GPU metrics directly from DRM sysfs / hwmon.
func (gm *GPUManager) collectIntelSysfsStats() error {
sysfsPollInterval := 3000 * time.Millisecond
cards, err := discoverIntelSysfsCards()
if err != nil {
return err
}
if len(cards) == 0 {
return errNoValidData
}
slog.Debug("Using sysfs for Intel GPU data collection", "cards", len(cards))
for _, card := range cards {
slog.Debug("Intel sysfs card detected", "card", filepath.Base(card.cardPath), "hwmon", card.hwmonDir)
}
failures := 0
for {
hasData := false
for _, card := range cards {
if gm.updateIntelSysfsGpuData(card.cardPath, card.hwmonDir) {
hasData = true
}
}
if !hasData {
failures++
if failures > maxFailureRetries {
return errNoValidData
}
slog.Warn("No Intel GPU data from sysfs", "failures", failures)
time.Sleep(retryWaitTime)
continue
}
failures = 0
time.Sleep(sysfsPollInterval)
}
}
func discoverIntelSysfsCards() ([]intelSysfsCard, error) {
paths, err := filepath.Glob(filepath.Join(drmSysfsRoot, "card*"))
if err != nil {
return nil, err
}
var cards []intelSysfsCard
for _, cardPath := range paths {
if strings.Contains(filepath.Base(cardPath), "-") || !isIntelGpu(cardPath) {
continue
}
hwmonDir := findIntelEnergyHwmon(filepath.Join(cardPath, "device"))
if hwmonDir == "" {
continue
}
cards = append(cards, intelSysfsCard{cardPath: cardPath, hwmonDir: hwmonDir})
}
return cards, nil
}
func isIntelGpu(cardPath string) bool {
vendor, err := utils.ReadStringFileLimited(filepath.Join(cardPath, "device/vendor"), 64)
if err != nil {
return false
}
return strings.EqualFold(strings.TrimSpace(vendor), "0x8086")
}
func findIntelEnergyHwmon(devicePath string) string {
hwmons, _ := filepath.Glob(filepath.Join(devicePath, "hwmon/hwmon*"))
var fallback string
for _, hwmonDir := range hwmons {
if !sysfsFileExists(filepath.Join(hwmonDir, "energy1_input")) {
continue
}
if name, err := utils.ReadStringFileLimited(filepath.Join(hwmonDir, "name"), 64); err == nil && strings.EqualFold(strings.TrimSpace(name), "xe") {
return hwmonDir
}
if fallback == "" {
fallback = hwmonDir
}
}
return fallback
}
func sysfsFileExists(path string) bool {
_, err := utils.ReadStringFileLimited(path, 1)
return err == nil
}
// updateIntelSysfsGpuData reads GPU metrics from sysfs and updates the GPU data map.
// Returns true if the required energy counter was read successfully.
func (gm *GPUManager) updateIntelSysfsGpuData(cardPath, hwmonDir string) bool {
devicePath := filepath.Join(cardPath, "device")
id := filepath.Base(cardPath)
energy, err := readSysfsUint(filepath.Join(hwmonDir, "energy1_input"))
if err != nil {
return false
}
now := intelSysfsNow()
power, hasPower := gm.calculateIntelSysfsPower(id, energy, now)
powerPkg, hasPowerPkg := gm.readIntelSysfsPowerPkg(id, hwmonDir, now)
temp := readIntelSysfsTemperature(hwmonDir)
usage, usageErr := readOptionalSysfsFloat(filepath.Join(devicePath, "gpu_busy_percent"))
memUsed, memUsedErr := readFirstOptionalSysfsFloat(
filepath.Join(devicePath, "mem_info_vram_used"),
filepath.Join(devicePath, "mem_info_lmem_used"),
filepath.Join(devicePath, "mem_info_local_mem_used"),
)
memTotal, memTotalErr := readFirstOptionalSysfsFloat(
filepath.Join(devicePath, "mem_info_vram_total"),
filepath.Join(devicePath, "mem_info_lmem_total"),
filepath.Join(devicePath, "mem_info_local_mem_total"),
)
gm.Lock()
defer gm.Unlock()
gpu, ok := gm.GpuDataMap[id]
if !ok {
gpu = &system.GPUData{Name: getIntelSysfsGpuName(cardPath)}
gm.GpuDataMap[id] = gpu
}
if usageErr == nil {
gpu.Usage += usage
}
if memUsedErr == nil {
gpu.MemoryUsed = utils.BytesToMegabytes(memUsed)
}
if memTotalErr == nil {
gpu.MemoryTotal = utils.BytesToMegabytes(memTotal)
}
if temp > 0 {
gpu.Temperature = temp
}
if hasPower {
gpu.Power += power
slog.Debug("Computed Intel sysfs GPU power", "card", id, "watts", power)
}
if hasPowerPkg {
gpu.PowerPkg += powerPkg
}
gpu.Count++
return true
}
func (gm *GPUManager) calculateIntelSysfsPower(cardID string, microjoules uint64, timestamp time.Time) (float64, bool) {
if gm.intelSysfsEnergySnapshots == nil {
gm.intelSysfsEnergySnapshots = make(map[string]intelSysfsEnergySnapshot)
}
last, ok := gm.intelSysfsEnergySnapshots[cardID]
gm.intelSysfsEnergySnapshots[cardID] = intelSysfsEnergySnapshot{microjoules: microjoules, timestamp: timestamp}
if !ok {
return 0, false
}
if microjoules < last.microjoules {
slog.Debug("Intel sysfs energy counter reset", "card", cardID)
return 0, false
}
elapsed := timestamp.Sub(last.timestamp).Seconds()
if elapsed <= 0 {
return 0, false
}
delta := microjoules - last.microjoules
return float64(delta) / 1_000_000.0 / elapsed, true
}
func (gm *GPUManager) readIntelSysfsPowerPkg(cardID, hwmonDir string, timestamp time.Time) (float64, bool) {
energyPaths, _ := filepath.Glob(filepath.Join(hwmonDir, "energy*_input"))
for _, path := range energyPaths {
if filepath.Base(path) == "energy1_input" {
continue
}
energy, err := readSysfsUint(path)
if err != nil {
continue
}
return gm.calculateIntelSysfsPower(cardID+":"+filepath.Base(path), energy, timestamp)
}
return 0, false
}
func readIntelSysfsTemperature(hwmonDir string) float64 {
tempPaths, _ := filepath.Glob(filepath.Join(hwmonDir, "temp*_input"))
for _, path := range tempPaths {
temp, err := readSysfsFloat(path)
if err == nil && temp > 0 {
return temp / 1000.0
}
}
return 0
}
func readSysfsUint(path string) (uint64, error) {
val, err := utils.ReadStringFileLimited(path, 64)
if err != nil {
slog.Debug("Failed to read sysfs value", "path", path, "error", err)
return 0, err
}
return strconv.ParseUint(strings.TrimSpace(val), 10, 64)
}
func readOptionalSysfsFloat(path string) (float64, error) {
val, err := os.ReadFile(path)
if err != nil {
return 0, err
}
return strconv.ParseFloat(strings.TrimSpace(string(val)), 64)
}
func readFirstOptionalSysfsFloat(paths ...string) (float64, error) {
for _, path := range paths {
val, err := readOptionalSysfsFloat(path)
if err == nil {
return val, nil
}
}
return 0, fmt.Errorf("no sysfs values found")
}
func getIntelSysfsGpuName(cardPath string) string {
devicePath := filepath.Join(cardPath, "device")
if product, err := utils.ReadStringFileLimited(filepath.Join(devicePath, "product_name"), 128); err == nil && strings.TrimSpace(product) != "" {
return strings.TrimSpace(product)
}
if name, err := utils.ReadStringFileLimited(filepath.Join(devicePath, "name"), 128); err == nil && strings.TrimSpace(name) != "" {
return strings.TrimSpace(name)
}
return fmt.Sprintf("Intel GPU %s", filepath.Base(cardPath))
}
-217
View File
@@ -1,217 +0,0 @@
//go:build linux
package agent
import (
"os"
"path/filepath"
"testing"
"time"
"github.com/henrygd/beszel/agent/utils"
"github.com/henrygd/beszel/internal/entities/system"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
func setupIntelSysfsTest(t *testing.T) (root, cardPath, hwmonPath string) {
t.Helper()
root = t.TempDir()
oldRoot := drmSysfsRoot
drmSysfsRoot = root
t.Cleanup(func() {
drmSysfsRoot = oldRoot
})
cardPath = filepath.Join(root, "card0")
devicePath := filepath.Join(cardPath, "device")
hwmonPath = filepath.Join(devicePath, "hwmon", "hwmon0")
require.NoError(t, os.MkdirAll(hwmonPath, 0o755))
return root, cardPath, hwmonPath
}
func writeIntelSysfsFile(t *testing.T, basePath, name, content string) {
t.Helper()
require.NoError(t, os.WriteFile(filepath.Join(basePath, name), []byte(content), 0o644))
}
func setIntelSysfsTime(t *testing.T, now time.Time) {
t.Helper()
oldNow := intelSysfsNow
intelSysfsNow = func() time.Time { return now }
t.Cleanup(func() {
intelSysfsNow = oldNow
})
}
func TestIntelSysfsDetectsIntelCardWithEnergy(t *testing.T) {
_, cardPath, hwmonPath := setupIntelSysfsTest(t)
devicePath := filepath.Join(cardPath, "device")
writeIntelSysfsFile(t, devicePath, "vendor", "0x8086\n")
writeIntelSysfsFile(t, hwmonPath, "name", "xe\n")
writeIntelSysfsFile(t, hwmonPath, "energy1_input", "1000000\n")
gm := &GPUManager{}
assert.True(t, gm.hasIntelSysfs())
cards, err := discoverIntelSysfsCards()
require.NoError(t, err)
require.Len(t, cards, 1)
assert.Equal(t, cardPath, cards[0].cardPath)
assert.Equal(t, hwmonPath, cards[0].hwmonDir)
}
func TestIntelSysfsRejectsNonIntelCard(t *testing.T) {
_, cardPath, hwmonPath := setupIntelSysfsTest(t)
devicePath := filepath.Join(cardPath, "device")
writeIntelSysfsFile(t, devicePath, "vendor", "0x1002\n")
writeIntelSysfsFile(t, hwmonPath, "name", "xe\n")
writeIntelSysfsFile(t, hwmonPath, "energy1_input", "1000000\n")
gm := &GPUManager{}
assert.False(t, gm.hasIntelSysfs())
}
func TestIntelSysfsRequiresEnergyInput(t *testing.T) {
_, cardPath, hwmonPath := setupIntelSysfsTest(t)
devicePath := filepath.Join(cardPath, "device")
writeIntelSysfsFile(t, devicePath, "vendor", "0x8086\n")
writeIntelSysfsFile(t, hwmonPath, "name", "xe\n")
gm := &GPUManager{}
assert.False(t, gm.hasIntelSysfs())
}
func TestIntelSysfsFirstSampleInitializesWithoutBogusPower(t *testing.T) {
_, cardPath, hwmonPath := setupIntelSysfsTest(t)
devicePath := filepath.Join(cardPath, "device")
writeIntelSysfsFile(t, devicePath, "vendor", "0x8086\n")
writeIntelSysfsFile(t, hwmonPath, "name", "xe\n")
writeIntelSysfsFile(t, hwmonPath, "energy1_input", "1000000\n")
setIntelSysfsTime(t, time.Unix(100, 0))
gm := &GPUManager{GpuDataMap: make(map[string]*system.GPUData)}
ok := gm.updateIntelSysfsGpuData(cardPath, hwmonPath)
require.True(t, ok)
gpu := gm.GpuDataMap["card0"]
require.NotNil(t, gpu)
assert.Equal(t, "Intel GPU card0", gpu.Name)
assert.Equal(t, 0.0, gpu.Power)
assert.Equal(t, 1.0, gpu.Count)
}
func TestIntelSysfsSecondSampleComputesWatts(t *testing.T) {
_, cardPath, hwmonPath := setupIntelSysfsTest(t)
devicePath := filepath.Join(cardPath, "device")
writeIntelSysfsFile(t, devicePath, "vendor", "0x8086\n")
writeIntelSysfsFile(t, hwmonPath, "energy1_input", "1000000\n")
gm := &GPUManager{GpuDataMap: make(map[string]*system.GPUData)}
oldNow := intelSysfsNow
intelSysfsNow = func() time.Time { return time.Unix(100, 0) }
t.Cleanup(func() { intelSysfsNow = oldNow })
require.True(t, gm.updateIntelSysfsGpuData(cardPath, hwmonPath))
writeIntelSysfsFile(t, hwmonPath, "energy1_input", "6000000\n")
intelSysfsNow = func() time.Time { return time.Unix(102, 0) }
require.True(t, gm.updateIntelSysfsGpuData(cardPath, hwmonPath))
gpu := gm.GpuDataMap["card0"]
require.NotNil(t, gpu)
assert.Equal(t, 2.5, gpu.Power)
assert.Equal(t, 2.0, gpu.Count)
}
func TestIntelSysfsSecondEnergyCounterMapsToPowerPkg(t *testing.T) {
_, cardPath, hwmonPath := setupIntelSysfsTest(t)
devicePath := filepath.Join(cardPath, "device")
writeIntelSysfsFile(t, devicePath, "vendor", "0x8086\n")
writeIntelSysfsFile(t, hwmonPath, "energy1_input", "1000000\n")
writeIntelSysfsFile(t, hwmonPath, "energy2_input", "2000000\n")
oldNow := intelSysfsNow
t.Cleanup(func() { intelSysfsNow = oldNow })
gm := &GPUManager{GpuDataMap: make(map[string]*system.GPUData)}
intelSysfsNow = func() time.Time { return time.Unix(100, 0) }
require.True(t, gm.updateIntelSysfsGpuData(cardPath, hwmonPath))
writeIntelSysfsFile(t, hwmonPath, "energy1_input", "2000000\n")
writeIntelSysfsFile(t, hwmonPath, "energy2_input", "8000000\n")
intelSysfsNow = func() time.Time { return time.Unix(102, 0) }
require.True(t, gm.updateIntelSysfsGpuData(cardPath, hwmonPath))
gpu := gm.GpuDataMap["card0"]
require.NotNil(t, gpu)
assert.Equal(t, 0.5, gpu.Power)
assert.Equal(t, 3.0, gpu.PowerPkg)
}
func TestIntelSysfsCounterResetSkipsOneSample(t *testing.T) {
gm := &GPUManager{}
power, ok := gm.calculateIntelSysfsPower("card0", 5000000, time.Unix(100, 0))
assert.False(t, ok)
assert.Equal(t, 0.0, power)
power, ok = gm.calculateIntelSysfsPower("card0", 1000000, time.Unix(101, 0))
assert.False(t, ok)
assert.Equal(t, 0.0, power)
power, ok = gm.calculateIntelSysfsPower("card0", 3000000, time.Unix(103, 0))
assert.True(t, ok)
assert.Equal(t, 1.0, power)
}
func TestIntelSysfsTempInputMapsToCelsius(t *testing.T) {
_, cardPath, hwmonPath := setupIntelSysfsTest(t)
devicePath := filepath.Join(cardPath, "device")
writeIntelSysfsFile(t, devicePath, "vendor", "0x8086\n")
writeIntelSysfsFile(t, hwmonPath, "energy1_input", "1000000\n")
writeIntelSysfsFile(t, hwmonPath, "temp1_input", "43500\n")
setIntelSysfsTime(t, time.Unix(100, 0))
gm := &GPUManager{GpuDataMap: make(map[string]*system.GPUData)}
require.True(t, gm.updateIntelSysfsGpuData(cardPath, hwmonPath))
gpu := gm.GpuDataMap["card0"]
require.NotNil(t, gpu)
assert.Equal(t, 43.5, gpu.Temperature)
}
func TestIntelSysfsMissingOptionalFilesDoNotFail(t *testing.T) {
_, cardPath, hwmonPath := setupIntelSysfsTest(t)
devicePath := filepath.Join(cardPath, "device")
writeIntelSysfsFile(t, devicePath, "vendor", "0x8086\n")
writeIntelSysfsFile(t, hwmonPath, "energy1_input", "1000000\n")
setIntelSysfsTime(t, time.Unix(100, 0))
gm := &GPUManager{GpuDataMap: make(map[string]*system.GPUData)}
require.True(t, gm.updateIntelSysfsGpuData(cardPath, hwmonPath))
gpu := gm.GpuDataMap["card0"]
require.NotNil(t, gpu)
assert.Equal(t, 0.0, gpu.Usage)
assert.Equal(t, 0.0, gpu.MemoryUsed)
assert.Equal(t, 0.0, gpu.MemoryTotal)
assert.Equal(t, 0.0, gpu.Temperature)
}
func TestIntelSysfsMapsOpportunisticMemoryAndUsage(t *testing.T) {
_, cardPath, hwmonPath := setupIntelSysfsTest(t)
devicePath := filepath.Join(cardPath, "device")
writeIntelSysfsFile(t, devicePath, "vendor", "0x8086\n")
writeIntelSysfsFile(t, devicePath, "gpu_busy_percent", "37\n")
writeIntelSysfsFile(t, devicePath, "mem_info_lmem_used", "1073741824\n")
writeIntelSysfsFile(t, devicePath, "mem_info_lmem_total", "2147483648\n")
writeIntelSysfsFile(t, hwmonPath, "energy1_input", "1000000\n")
setIntelSysfsTime(t, time.Unix(100, 0))
gm := &GPUManager{GpuDataMap: make(map[string]*system.GPUData)}
require.True(t, gm.updateIntelSysfsGpuData(cardPath, hwmonPath))
gpu := gm.GpuDataMap["card0"]
require.NotNil(t, gpu)
assert.Equal(t, 37.0, gpu.Usage)
assert.Equal(t, utils.BytesToMegabytes(1073741824), gpu.MemoryUsed)
assert.Equal(t, utils.BytesToMegabytes(2147483648), gpu.MemoryTotal)
}
-13
View File
@@ -1,13 +0,0 @@
//go:build !linux
package agent
type intelSysfsEnergySnapshot struct{}
func (gm *GPUManager) hasIntelSysfs() bool {
return false
}
func (gm *GPUManager) startIntelSysfsCollector() bool {
return false
}
+20 -34
View File
@@ -1,16 +1,14 @@
//go:build amd64 && (windows || (linux && glibc))
package agent package agent
import ( import (
"fmt" "fmt"
"log/slog"
"strings" "strings"
"time" "time"
"unsafe" "unsafe"
"github.com/ebitengine/purego" "github.com/ebitengine/purego"
"github.com/henrygd/beszel/internal/entities/system" "github.com/henrygd/beszel/internal/entities/system"
"golang.org/x/exp/slog"
) )
// NVML constants and types // NVML constants and types
@@ -74,7 +72,7 @@ type nvmlCollector struct {
} }
func (c *nvmlCollector) init() error { func (c *nvmlCollector) init() error {
slog.Debug("NVML: Initializing") slog.Info("NVML: Initializing")
libPath := getNVMLPath() libPath := getNVMLPath()
lib, err := openLibrary(libPath) lib, err := openLibrary(libPath)
@@ -161,7 +159,7 @@ func (c *nvmlCollector) collect() {
gpu := c.gm.GpuDataMap[id] gpu := c.gm.GpuDataMap[id]
if bdf != "" && !c.isGPUActive(bdf) { if bdf != "" && !c.isGPUActive(bdf) {
slog.Debug("NVML: GPU is suspended, skipping", "bdf", bdf) slog.Info("NVML: GPU is suspended, skipping", "bdf", bdf)
gpu.Temperature = 0 gpu.Temperature = 0
gpu.MemoryUsed = 0 gpu.MemoryUsed = 0
continue continue
@@ -170,45 +168,31 @@ func (c *nvmlCollector) collect() {
// Utilization // Utilization
var utilization nvmlUtilization var utilization nvmlUtilization
if ret := nvmlDeviceGetUtilizationRates(device, &utilization); ret != nvmlReturn(nvmlSuccess) { if ret := nvmlDeviceGetUtilizationRates(device, &utilization); ret != nvmlReturn(nvmlSuccess) {
slog.Debug("NVML: Utilization failed (GPU likely suspended)", "bdf", bdf, "ret", ret) slog.Info("NVML: Utilization failed (GPU likely suspended)", "bdf", bdf, "ret", ret)
gpu.Temperature = 0 gpu.Temperature = 0
gpu.MemoryUsed = 0 gpu.MemoryUsed = 0
continue continue
} }
slog.Debug("NVML: Collecting data for GPU", "bdf", bdf) slog.Info("NVML: Collecting data for GPU", "bdf", bdf)
// Temperature // Temperature
var temp uint32 var temp uint32
nvmlDeviceGetTemperature(device, 0, &temp) // 0 is NVML_TEMPERATURE_GPU nvmlDeviceGetTemperature(device, 0, &temp) // 0 is NVML_TEMPERATURE_GPU
// Memory: only poll if GPU is active to avoid leaving D3cold state (#1522) // Memory
if utilization.Gpu > 0 { var usedMem, totalMem uint64
var usedMem, totalMem uint64 if c.isV2 {
if c.isV2 { var memory nvmlMemoryV2
var memory nvmlMemoryV2 memory.Version = 0x02000028 // (2 << 24) | 40 bytes
memory.Version = 0x02000028 // (2 << 24) | 40 bytes nvmlDeviceGetMemoryInfo(device, uintptr(unsafe.Pointer(&memory)))
if ret := nvmlDeviceGetMemoryInfo(device, uintptr(unsafe.Pointer(&memory))); ret != nvmlReturn(nvmlSuccess) { usedMem = memory.Used
slog.Debug("NVML: MemoryInfo_v2 failed", "bdf", bdf, "ret", ret) totalMem = memory.Total
} else {
usedMem = memory.Used
totalMem = memory.Total
}
} else {
var memory nvmlMemoryV1
if ret := nvmlDeviceGetMemoryInfo(device, uintptr(unsafe.Pointer(&memory))); ret != nvmlReturn(nvmlSuccess) {
slog.Debug("NVML: MemoryInfo failed", "bdf", bdf, "ret", ret)
} else {
usedMem = memory.Used
totalMem = memory.Total
}
}
if totalMem > 0 {
gpu.MemoryUsed = float64(usedMem) / 1024 / 1024 / mebibytesInAMegabyte
gpu.MemoryTotal = float64(totalMem) / 1024 / 1024 / mebibytesInAMegabyte
}
} else { } else {
slog.Debug("NVML: Skipping memory info (utilization=0)", "bdf", bdf) var memory nvmlMemoryV1
nvmlDeviceGetMemoryInfo(device, uintptr(unsafe.Pointer(&memory)))
usedMem = memory.Used
totalMem = memory.Total
} }
// Power // Power
@@ -216,9 +200,11 @@ func (c *nvmlCollector) collect() {
nvmlDeviceGetPowerUsage(device, &power) nvmlDeviceGetPowerUsage(device, &power)
gpu.Temperature = float64(temp) gpu.Temperature = float64(temp)
gpu.MemoryUsed = float64(usedMem) / 1024 / 1024 / mebibytesInAMegabyte
gpu.MemoryTotal = float64(totalMem) / 1024 / 1024 / mebibytesInAMegabyte
gpu.Usage += float64(utilization.Gpu) gpu.Usage += float64(utilization.Gpu)
gpu.Power += float64(power) / 1000.0 gpu.Power += float64(power) / 1000.0
gpu.Count++ gpu.Count++
slog.Debug("NVML: Collected data", "gpu", gpu) slog.Info("NVML: Collected data", "gpu", gpu)
} }
} }
+4 -5
View File
@@ -1,14 +1,14 @@
//go:build glibc && linux && amd64 //go:build linux
package agent package agent
import ( import (
"log/slog"
"os" "os"
"path/filepath" "path/filepath"
"strings" "strings"
"github.com/ebitengine/purego" "github.com/ebitengine/purego"
"golang.org/x/exp/slog"
) )
func openLibrary(name string) (uintptr, error) { func openLibrary(name string) (uintptr, error) {
@@ -29,12 +29,12 @@ func (c *nvmlCollector) isGPUActive(bdf string) bool {
statusPath := filepath.Join("/sys/bus/pci/devices", bdf, "power/runtime_status") statusPath := filepath.Join("/sys/bus/pci/devices", bdf, "power/runtime_status")
status, err := os.ReadFile(statusPath) status, err := os.ReadFile(statusPath)
if err != nil { if err != nil {
slog.Debug("NVML: Can't read runtime_status", "bdf", bdf, "err", err) slog.Info("NVML: Can't read runtime_status", "bdf", bdf, "err", err)
return true // Assume active if we can't read status return true // Assume active if we can't read status
} }
statusStr := strings.TrimSpace(string(status)) statusStr := strings.TrimSpace(string(status))
if statusStr != "active" && statusStr != "resuming" { if statusStr != "active" && statusStr != "resuming" {
slog.Debug("NVML: GPU not active", "bdf", bdf, "status", statusStr) slog.Info("NVML: GPU is not active", "bdf", bdf, "status", statusStr)
return false return false
} }
@@ -47,7 +47,6 @@ func (c *nvmlCollector) isGPUActive(bdf string) bool {
if err == nil { if err == nil {
pstateStr := strings.TrimSpace(string(pstate)) pstateStr := strings.TrimSpace(string(pstate))
if pstateStr != "D0" { if pstateStr != "D0" {
slog.Debug("NVML: GPU not in D0 state", "bdf", bdf, "pstate", pstateStr)
return false return false
} }
} }
+12 -6
View File
@@ -1,15 +1,21 @@
//go:build (!linux && !windows) || !amd64 || (linux && !glibc) //go:build !linux && !windows
package agent package agent
import "fmt" import "fmt"
type nvmlCollector struct { func openLibrary(name string) (uintptr, error) {
gm *GPUManager return 0, fmt.Errorf("nvml not supported on this platform")
} }
func (c *nvmlCollector) init() error { func getNVMLPath() string {
return fmt.Errorf("nvml not supported on this platform") return ""
} }
func (c *nvmlCollector) start() {} func hasSymbol(lib uintptr, symbol string) bool {
return false
}
func (c *nvmlCollector) isGPUActive(bdf string) bool {
return true
}
+1 -1
View File
@@ -1,4 +1,4 @@
//go:build windows && amd64 //go:build windows
package agent package agent
-201
View File
@@ -1,201 +0,0 @@
package agent
import (
"encoding/json"
"io"
"log/slog"
"os/exec"
"path/filepath"
"strconv"
"strings"
"time"
"github.com/henrygd/beszel/agent/utils"
"github.com/henrygd/beszel/internal/entities/system"
)
type nvtopSnapshot struct {
DeviceName string `json:"device_name"`
Temp *string `json:"temp"`
PowerDraw *string `json:"power_draw"`
GpuUtil *string `json:"gpu_util"`
MemTotal *string `json:"mem_total"`
MemUsed *string `json:"mem_used"`
}
// parseNvtopNumber parses nvtop numeric strings with units (C/W/%).
func parseNvtopNumber(raw string) float64 {
cleaned := strings.TrimSpace(raw)
cleaned = strings.TrimSuffix(cleaned, "C")
cleaned = strings.TrimSuffix(cleaned, "W")
cleaned = strings.TrimSuffix(cleaned, "%")
val, _ := strconv.ParseFloat(cleaned, 64)
return val
}
// parseNvtopData parses a single nvtop JSON snapshot payload.
func (gm *GPUManager) parseNvtopData(output []byte) bool {
var snapshots []nvtopSnapshot
if err := json.Unmarshal(output, &snapshots); err != nil || len(snapshots) == 0 {
return false
}
return gm.updateNvtopSnapshots(snapshots)
}
// updateNvtopSnapshots applies one decoded nvtop snapshot batch to GPU accumulators.
func (gm *GPUManager) updateNvtopSnapshots(snapshots []nvtopSnapshot) bool {
gm.Lock()
defer gm.Unlock()
valid := false
usedIDs := make(map[string]struct{}, len(snapshots))
var xeName string
for i, sample := range snapshots {
// nvtop leaves device_name unset on xe devices.
if sample.DeviceName == "" {
if xeName == "" {
xeName = xeGpuName()
}
sample.DeviceName = xeName
}
indexID := "n" + strconv.Itoa(i)
id := indexID
// nvtop ordering can change, so prefer reusing an existing slot with matching device name.
if existingByIndex, ok := gm.GpuDataMap[indexID]; ok && existingByIndex.Name != "" && existingByIndex.Name != sample.DeviceName {
for existingID, gpu := range gm.GpuDataMap {
if !strings.HasPrefix(existingID, "n") {
continue
}
if _, taken := usedIDs[existingID]; taken {
continue
}
if gpu.Name == sample.DeviceName {
id = existingID
break
}
}
}
if _, ok := gm.GpuDataMap[id]; !ok {
gm.GpuDataMap[id] = &system.GPUData{Name: sample.DeviceName}
}
gpu := gm.GpuDataMap[id]
gpu.Name = sample.DeviceName
if sample.Temp != nil {
gpu.Temperature = parseNvtopNumber(*sample.Temp)
}
if sample.MemUsed != nil {
gpu.MemoryUsed = utils.BytesToMegabytes(parseNvtopNumber(*sample.MemUsed))
}
if sample.MemTotal != nil {
gpu.MemoryTotal = utils.BytesToMegabytes(parseNvtopNumber(*sample.MemTotal))
}
if sample.GpuUtil != nil {
gpu.Usage += parseNvtopNumber(*sample.GpuUtil)
}
if sample.PowerDraw != nil {
gpu.Power += parseNvtopNumber(*sample.PowerDraw)
}
gpu.Count++
usedIDs[id] = struct{}{}
valid = true
}
return valid
}
// collectNvtopStats runs nvtop loop mode and continuously decodes JSON snapshots.
func (gm *GPUManager) collectNvtopStats(interval string) error {
cmd := exec.Command(nvtopCmd, "-lP", "-d", interval)
stdout, err := cmd.StdoutPipe()
if err != nil {
return err
}
if err := cmd.Start(); err != nil {
return err
}
defer func() {
_ = stdout.Close()
if cmd.ProcessState == nil || !cmd.ProcessState.Exited() {
_ = cmd.Process.Kill()
}
_ = cmd.Wait()
}()
decoder := json.NewDecoder(stdout)
foundValid := false
for {
var snapshots []nvtopSnapshot
if err := decoder.Decode(&snapshots); err != nil {
if err == io.EOF {
if foundValid {
return nil
}
return errNoValidData
}
return err
}
if gm.updateNvtopSnapshots(snapshots) {
foundValid = true
}
}
}
// startNvtopCollector starts nvtop collection with retry or fallback callback handling.
func (gm *GPUManager) startNvtopCollector(interval string, onFailure func()) {
go func() {
failures := 0
for {
if err := gm.collectNvtopStats(interval); err != nil {
if onFailure != nil {
slog.Warn("Error collecting GPU data via nvtop", "err", err)
onFailure()
return
}
failures++
if failures > maxFailureRetries {
break
}
slog.Warn("Error collecting GPU data via nvtop", "err", err)
time.Sleep(retryWaitTime)
continue
}
}
}()
}
// xeDevicePath returns the sysfs device path of the first xe GPU, or "".
func xeDevicePath() string {
cards, err := filepath.Glob("/sys/class/drm/card*")
if err != nil {
return ""
}
for _, card := range cards {
if strings.Contains(filepath.Base(card), "-") {
continue
}
if uevent, err := utils.ReadStringFileLimited(filepath.Join(card, "device", "uevent"), 4096); err == nil && strings.Contains(uevent, "DRIVER=xe") {
return filepath.Join(card, "device")
}
}
return ""
}
func (gm *GPUManager) hasXe() bool {
return xeDevicePath() != ""
}
// xeGpuName names an xe GPU from its PCI device id; nvtop leaves device_name unset on xe.
func xeGpuName() string {
devicePath := xeDevicePath()
if devicePath == "" {
return "GPU"
}
id, err := utils.ReadStringFileLimited(filepath.Join(devicePath, "device"), 64)
if err != nil {
return "GPU"
}
id = strings.ToLower(strings.TrimSpace(strings.TrimPrefix(id, "0x")))
return "Intel GPU (" + id + ")"
}
+44 -415
View File
@@ -1,4 +1,5 @@
//go:build testing //go:build testing
// +build testing
package agent package agent
@@ -10,7 +11,6 @@ import (
"testing" "testing"
"time" "time"
"github.com/henrygd/beszel/agent/utils"
"github.com/henrygd/beszel/internal/entities/system" "github.com/henrygd/beszel/internal/entities/system"
"github.com/stretchr/testify/assert" "github.com/stretchr/testify/assert"
@@ -250,101 +250,6 @@ func TestParseAmdData(t *testing.T) {
} }
} }
func TestParseNvtopData(t *testing.T) {
input, err := os.ReadFile("test-data/nvtop.json")
require.NoError(t, err)
gm := &GPUManager{
GpuDataMap: make(map[string]*system.GPUData),
}
valid := gm.parseNvtopData(input)
require.True(t, valid)
g0, ok := gm.GpuDataMap["n0"]
require.True(t, ok)
assert.Equal(t, "NVIDIA GeForce RTX 3050 Ti Laptop GPU", g0.Name)
assert.Equal(t, 48.0, g0.Temperature)
assert.Equal(t, 5.0, g0.Usage)
assert.Equal(t, 13.0, g0.Power)
assert.Equal(t, utils.BytesToMegabytes(349372416), g0.MemoryUsed)
assert.Equal(t, utils.BytesToMegabytes(4294967296), g0.MemoryTotal)
assert.Equal(t, 1.0, g0.Count)
g1, ok := gm.GpuDataMap["n1"]
require.True(t, ok)
assert.Equal(t, "AMD Radeon 680M", g1.Name)
assert.Equal(t, 48.0, g1.Temperature)
assert.Equal(t, 12.0, g1.Usage)
assert.Equal(t, 9.0, g1.Power)
assert.Equal(t, utils.BytesToMegabytes(1213784064), g1.MemoryUsed)
assert.Equal(t, utils.BytesToMegabytes(16929173504), g1.MemoryTotal)
assert.Equal(t, 1.0, g1.Count)
}
func TestUpdateNvtopSnapshotsKeepsDeviceAssociationWhenOrderChanges(t *testing.T) {
strPtr := func(s string) *string { return &s }
gm := &GPUManager{
GpuDataMap: make(map[string]*system.GPUData),
}
firstBatch := []nvtopSnapshot{
{
DeviceName: "NVIDIA GeForce RTX 3050 Ti Laptop GPU",
GpuUtil: strPtr("20%"),
PowerDraw: strPtr("10W"),
},
{
DeviceName: "AMD Radeon 680M",
GpuUtil: strPtr("30%"),
PowerDraw: strPtr("20W"),
},
}
secondBatchSwapped := []nvtopSnapshot{
{
DeviceName: "AMD Radeon 680M",
GpuUtil: strPtr("40%"),
PowerDraw: strPtr("25W"),
},
{
DeviceName: "NVIDIA GeForce RTX 3050 Ti Laptop GPU",
GpuUtil: strPtr("50%"),
PowerDraw: strPtr("15W"),
},
}
require.True(t, gm.updateNvtopSnapshots(firstBatch))
require.True(t, gm.updateNvtopSnapshots(secondBatchSwapped))
nvidia := gm.GpuDataMap["n0"]
require.NotNil(t, nvidia)
assert.Equal(t, "NVIDIA GeForce RTX 3050 Ti Laptop GPU", nvidia.Name)
assert.Equal(t, 70.0, nvidia.Usage)
assert.Equal(t, 25.0, nvidia.Power)
assert.Equal(t, 2.0, nvidia.Count)
amd := gm.GpuDataMap["n1"]
require.NotNil(t, amd)
assert.Equal(t, "AMD Radeon 680M", amd.Name)
assert.Equal(t, 70.0, amd.Usage)
assert.Equal(t, 45.0, amd.Power)
assert.Equal(t, 2.0, amd.Count)
}
func TestParseCollectorPriority(t *testing.T) {
got := parseCollectorPriority(" nvml, nvidia-smi, intel_gpu_top, intel_sysfs, amd_sysfs, nvtop, rocm-smi, bad ")
want := []collectorSource{
collectorSourceNVML,
collectorSourceNvidiaSMI,
collectorSourceIntelGpuTop,
collectorSourceIntelSysfs,
collectorSourceAmdSysfs,
collectorSourceNVTop,
collectorSourceRocmSMI,
}
assert.Equal(t, want, got)
}
func TestParseJetsonData(t *testing.T) { func TestParseJetsonData(t *testing.T) {
tests := []struct { tests := []struct {
name string name string
@@ -402,19 +307,6 @@ func TestParseJetsonData(t *testing.T) {
Count: 1, Count: 1,
}, },
}, },
{
name: "orin-style output with GPU@ temp and VDD_SYS_GPU power",
input: "RAM 3276/7859MB (lfb 5x4MB) SWAP 1626/12122MB (cached 181MB) CPU [44%@1421,49%@2031,67%@2034,17%@1420,25%@1419,8%@1420] EMC_FREQ 1%@1866 GR3D_FREQ 0%@114 APE 150 MTS fg 1% bg 1% PLL@42.5C MCPU@42.5C PMIC@50C Tboard@38C GPU@39.5C BCPU@42.5C thermal@41.3C Tdiode@39.25C VDD_SYS_GPU 182/182 VDD_SYS_SOC 730/730 VDD_4V0_WIFI 0/0 VDD_IN 5297/5297 VDD_SYS_CPU 1917/1917 VDD_SYS_DDR 1241/1241",
wantMetrics: &system.GPUData{
Name: "GPU",
MemoryUsed: 3276.0,
MemoryTotal: 7859.0,
Usage: 0.0,
Power: 0.182, // 182mW -> 0.182W
Temperature: 39.5,
Count: 1,
},
},
} }
for _, tt := range tests { for _, tt := range tests {
@@ -566,42 +458,6 @@ func TestGetCurrentData(t *testing.T) {
assert.EqualValues(t, 2, gm.GpuDataMap["0"].Count, "Count should still be 2") assert.EqualValues(t, 2, gm.GpuDataMap["0"].Count, "Count should still be 2")
}) })
t.Run("carries Intel GPU average forward between samples", func(t *testing.T) {
// Intel GPUs report no temp/memory, so between-sample gaps (delta 0) must
// reuse the last average instead of returning zeros and blanking the chart.
gm := &GPUManager{
GpuDataMap: map[string]*system.GPUData{
"0": {
Name: "GPU",
Usage: 0, // derived from engines for Intel
Power: 200, // averages to 100 over 2 counts
PowerPkg: 60, // averages to 30 over 2 counts
Count: 2,
Engines: map[string]float64{
"Render/3D": 80, // averages to 40
"Video": 20, // averages to 10
},
},
},
}
cacheKey := uint16(1000) // realtime cache key
// First collection - computes and stores averages
result1 := gm.GetCurrentData(cacheKey)
assert.InDelta(t, 100.0, result1["0"].Power, 0.01)
assert.InDelta(t, 30.0, result1["0"].PowerPkg, 0.01)
assert.InDelta(t, 40.0, result1["0"].Engines["Render/3D"], 0.01)
// Second collection with no new sample (count unchanged, temp/mem still 0).
// Must carry the last average forward rather than blanking to zero.
result2 := gm.GetCurrentData(cacheKey)
assert.Equal(t, "GPU", result2["0"].Name, "Name should be preserved")
assert.InDelta(t, 100.0, result2["0"].Power, 0.01, "Should reuse last average power, not 0")
assert.InDelta(t, 30.0, result2["0"].PowerPkg, 0.01, "Should reuse last average package power, not 0")
assert.InDelta(t, 40.0, result2["0"].Engines["Render/3D"], 0.01, "Should reuse last average engine usage")
})
t.Run("tracks separate averages per cache key", func(t *testing.T) { t.Run("tracks separate averages per cache key", func(t *testing.T) {
gm := &GPUManager{ gm := &GPUManager{
GpuDataMap: map[string]*system.GPUData{ GpuDataMap: map[string]*system.GPUData{
@@ -969,7 +825,7 @@ func TestInitializeSnapshots(t *testing.T) {
} }
func TestCalculateGPUAverage(t *testing.T) { func TestCalculateGPUAverage(t *testing.T) {
t.Run("returns cached average when deltaCount is zero", func(t *testing.T) { t.Run("returns zero value when deltaCount is zero", func(t *testing.T) {
gm := &GPUManager{ gm := &GPUManager{
lastSnapshots: map[uint16]map[string]*gpuSnapshot{ lastSnapshots: map[uint16]map[string]*gpuSnapshot{
5000: { 5000: {
@@ -982,10 +838,9 @@ func TestCalculateGPUAverage(t *testing.T) {
} }
gpu := &system.GPUData{ gpu := &system.GPUData{
Count: 10.0, // Same as snapshot, so delta = 0 Count: 10.0, // Same as snapshot, so delta = 0
Usage: 100.0, Usage: 100.0,
Power: 200.0, Power: 200.0,
Temperature: 50.0, // Non-zero to avoid "suspended" check
} }
result := gm.calculateGPUAverage("0", gpu, 5000) result := gm.calculateGPUAverage("0", gpu, 5000)
@@ -994,31 +849,6 @@ func TestCalculateGPUAverage(t *testing.T) {
assert.Equal(t, 100.0, result.Power, "Should return cached average") assert.Equal(t, 100.0, result.Power, "Should return cached average")
}) })
t.Run("returns zero value when GPU is suspended", func(t *testing.T) {
gm := &GPUManager{
lastSnapshots: map[uint16]map[string]*gpuSnapshot{
5000: {
"0": {count: 10, usage: 100, power: 200},
},
},
lastAvgData: map[string]system.GPUData{
"0": {Usage: 50.0, Power: 100.0},
},
}
gpu := &system.GPUData{
Name: "Test GPU",
Count: 10.0,
Temperature: 0,
MemoryUsed: 0,
}
result := gm.calculateGPUAverage("0", gpu, 5000)
assert.Equal(t, 0.0, result.Usage, "Should return zero usage")
assert.Equal(t, 0.0, result.Power, "Should return zero power")
})
t.Run("calculates average for standard GPU", func(t *testing.T) { t.Run("calculates average for standard GPU", func(t *testing.T) {
gm := &GPUManager{ gm := &GPUManager{
lastSnapshots: map[uint16]map[string]*gpuSnapshot{ lastSnapshots: map[uint16]map[string]*gpuSnapshot{
@@ -1118,32 +948,36 @@ func TestCalculateGPUAverage(t *testing.T) {
}) })
} }
func TestGPUCapabilitiesAndLegacyPriority(t *testing.T) { func TestDetectGPUs(t *testing.T) {
hasAmdSysfs := (&GPUManager{}).hasAmdSysfs() // Save original PATH
origPath := os.Getenv("PATH")
defer os.Setenv("PATH", origPath)
// Set up temp dir with the commands
tempDir := t.TempDir()
os.Setenv("PATH", tempDir)
tests := []struct { tests := []struct {
name string name string
setupCommands func(string) error setupCommands func() error
wantNvidiaSmi bool wantNvidiaSmi bool
wantRocmSmi bool wantRocmSmi bool
wantTegrastats bool wantTegrastats bool
wantNvtop bool
wantErr bool wantErr bool
}{ }{
{ {
name: "nvidia-smi not available", name: "nvidia-smi not available",
setupCommands: func(_ string) error { setupCommands: func() error {
return nil return nil
}, },
wantNvidiaSmi: false, wantNvidiaSmi: false,
wantRocmSmi: false, wantRocmSmi: false,
wantTegrastats: false, wantTegrastats: false,
wantNvtop: false,
wantErr: true, wantErr: true,
}, },
{ {
name: "nvidia-smi available", name: "nvidia-smi available",
setupCommands: func(tempDir string) error { setupCommands: func() error {
path := filepath.Join(tempDir, "nvidia-smi") path := filepath.Join(tempDir, "nvidia-smi")
script := `#!/bin/sh script := `#!/bin/sh
echo "test"` echo "test"`
@@ -1155,12 +989,11 @@ echo "test"`
wantNvidiaSmi: true, wantNvidiaSmi: true,
wantTegrastats: false, wantTegrastats: false,
wantRocmSmi: false, wantRocmSmi: false,
wantNvtop: false,
wantErr: false, wantErr: false,
}, },
{ {
name: "rocm-smi available", name: "rocm-smi available",
setupCommands: func(tempDir string) error { setupCommands: func() error {
path := filepath.Join(tempDir, "rocm-smi") path := filepath.Join(tempDir, "rocm-smi")
script := `#!/bin/sh script := `#!/bin/sh
echo "test"` echo "test"`
@@ -1169,15 +1002,14 @@ echo "test"`
} }
return nil return nil
}, },
wantNvidiaSmi: false, wantNvidiaSmi: true,
wantRocmSmi: true, wantRocmSmi: true,
wantTegrastats: false, wantTegrastats: false,
wantNvtop: false,
wantErr: false, wantErr: false,
}, },
{ {
name: "tegrastats available", name: "tegrastats available",
setupCommands: func(tempDir string) error { setupCommands: func() error {
path := filepath.Join(tempDir, "tegrastats") path := filepath.Join(tempDir, "tegrastats")
script := `#!/bin/sh script := `#!/bin/sh
echo "test"` echo "test"`
@@ -1187,32 +1019,14 @@ echo "test"`
return nil return nil
}, },
wantNvidiaSmi: false, wantNvidiaSmi: false,
wantRocmSmi: false, wantRocmSmi: true,
wantTegrastats: true, wantTegrastats: true,
wantNvtop: false,
wantErr: false,
},
{
name: "nvtop available",
setupCommands: func(tempDir string) error {
path := filepath.Join(tempDir, "nvtop")
script := `#!/bin/sh
echo "[]"`
if err := os.WriteFile(path, []byte(script), 0755); err != nil {
return err
}
return nil
},
wantNvidiaSmi: false,
wantRocmSmi: false,
wantTegrastats: false,
wantNvtop: true,
wantErr: false, wantErr: false,
}, },
{ {
name: "no gpu tools available", name: "no gpu tools available",
setupCommands: func(_ string) error { setupCommands: func() error {
// The subtest already restricts PATH to its empty temporary directory. os.Setenv("PATH", "")
return nil return nil
}, },
wantErr: true, wantErr: true,
@@ -1221,56 +1035,36 @@ echo "[]"`
for _, tt := range tests { for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) { t.Run(tt.name, func(t *testing.T) {
tempDir := t.TempDir() if err := tt.setupCommands(); err != nil {
t.Setenv("PATH", tempDir)
if err := tt.setupCommands(tempDir); err != nil {
t.Fatal(err) t.Fatal(err)
} }
gm := &GPUManager{} gm := &GPUManager{}
caps := gm.discoverGpuCapabilities() err := gm.detectGPUs()
var err error
if !hasAnyGpuCollector(caps) {
err = fmt.Errorf(noGPUFoundMsg)
}
priorities := gm.resolveLegacyCollectorPriority(caps)
hasPriority := func(source collectorSource) bool {
for _, s := range priorities {
if s == source {
return true
}
}
return false
}
gotNvidiaSmi := hasPriority(collectorSourceNvidiaSMI)
gotRocmSmi := hasPriority(collectorSourceRocmSMI)
gotTegrastats := caps.hasTegrastats
gotNvtop := caps.hasNvtop
t.Logf("nvidiaSmi: %v, rocmSmi: %v, tegrastats: %v", gotNvidiaSmi, gotRocmSmi, gotTegrastats) t.Logf("nvidiaSmi: %v, rocmSmi: %v, tegrastats: %v", gm.nvidiaSmi, gm.rocmSmi, gm.tegrastats)
wantErr := tt.wantErr if tt.wantErr {
if hasAmdSysfs && (tt.name == "nvidia-smi not available" || tt.name == "no gpu tools available") {
wantErr = false
}
if wantErr {
assert.Error(t, err) assert.Error(t, err)
return return
} }
assert.NoError(t, err) assert.NoError(t, err)
assert.Equal(t, tt.wantNvidiaSmi, gotNvidiaSmi) assert.Equal(t, tt.wantNvidiaSmi, gm.nvidiaSmi)
assert.Equal(t, tt.wantRocmSmi, gotRocmSmi) assert.Equal(t, tt.wantRocmSmi, gm.rocmSmi)
assert.Equal(t, tt.wantTegrastats, gotTegrastats) assert.Equal(t, tt.wantTegrastats, gm.tegrastats)
assert.Equal(t, tt.wantNvtop, gotNvtop)
}) })
} }
} }
func TestCollectorStartHelpers(t *testing.T) { func TestStartCollector(t *testing.T) {
// Save original PATH
origPath := os.Getenv("PATH")
defer os.Setenv("PATH", origPath)
// Set up temp dir with the commands // Set up temp dir with the commands
dir := t.TempDir() dir := t.TempDir()
t.Setenv("PATH", dir) os.Setenv("PATH", dir)
tests := []struct { tests := []struct {
name string name string
@@ -1348,27 +1142,6 @@ echo "11-14-2024 22:54:33 RAM 1024/4096MB GR3D_FREQ 80% tj@70C VDD_GPU_SOC 1000m
}, },
}, },
}, },
{
name: "nvtop collector",
command: "nvtop",
setup: func(t *testing.T) error {
path := filepath.Join(dir, "nvtop")
script := `#!/bin/sh
echo '[{"device_name":"NVIDIA Test GPU","temp":"52C","power_draw":"31W","gpu_util":"37%","mem_total":"4294967296","mem_used":"536870912","processes":[]}]'`
if err := os.WriteFile(path, []byte(script), 0755); err != nil {
return err
}
return nil
},
validate: func(t *testing.T, gm *GPUManager) {
gpu, exists := gm.GpuDataMap["n0"]
assert.True(t, exists)
if exists {
assert.Equal(t, "NVIDIA Test GPU", gpu.Name)
assert.Equal(t, 52.0, gpu.Temperature)
}
},
},
} }
for _, tt := range tests { for _, tt := range tests {
@@ -1381,161 +1154,13 @@ echo '[{"device_name":"NVIDIA Test GPU","temp":"52C","power_draw":"31W","gpu_uti
GpuDataMap: make(map[string]*system.GPUData), GpuDataMap: make(map[string]*system.GPUData),
} }
} }
switch tt.command { tt.gm.startCollector(tt.command)
case nvidiaSmiCmd:
tt.gm.startNvidiaSmiCollector("4")
case rocmSmiCmd:
tt.gm.startRocmSmiCollector(4300 * time.Millisecond)
case tegraStatsCmd:
tt.gm.startTegraStatsCollector("3700")
case nvtopCmd:
tt.gm.startNvtopCollector("30", nil)
default:
t.Fatalf("unknown test command %q", tt.command)
}
time.Sleep(50 * time.Millisecond) // Give collector time to run time.Sleep(50 * time.Millisecond) // Give collector time to run
tt.validate(t, tt.gm) tt.validate(t, tt.gm)
}) })
} }
} }
func TestNewGPUManagerPriorityNvtopFallback(t *testing.T) {
dir := t.TempDir()
t.Setenv("PATH", dir)
t.Setenv("BESZEL_AGENT_GPU_COLLECTOR", "nvtop,nvidia-smi")
nvtopPath := filepath.Join(dir, "nvtop")
nvtopScript := `#!/bin/sh
echo 'not-json'`
require.NoError(t, os.WriteFile(nvtopPath, []byte(nvtopScript), 0755))
nvidiaPath := filepath.Join(dir, "nvidia-smi")
nvidiaScript := `#!/bin/sh
echo "0, NVIDIA Priority GPU, 45, 512, 2048, 12, 25"`
require.NoError(t, os.WriteFile(nvidiaPath, []byte(nvidiaScript), 0755))
gm, err := NewGPUManager()
require.NoError(t, err)
require.NotNil(t, gm)
time.Sleep(150 * time.Millisecond)
gpu, ok := gm.GpuDataMap["0"]
require.True(t, ok)
assert.Equal(t, "Priority GPU", gpu.Name)
assert.Equal(t, 45.0, gpu.Temperature)
}
func TestNewGPUManagerPriorityMixedCollectors(t *testing.T) {
dir := t.TempDir()
t.Setenv("PATH", dir)
t.Setenv("BESZEL_AGENT_GPU_COLLECTOR", "intel_gpu_top,rocm-smi")
intelPath := filepath.Join(dir, "intel_gpu_top")
intelScript := `#!/bin/sh
echo "Freq MHz IRQ RC6 Power W IMC MiB/s RCS VCS"
echo " req act /s % gpu pkg rd wr % se wa % se wa"
echo "226 223 338 58 2.00 2.69 1820 965 0.00 0 0 0.00 0 0"
echo "189 187 412 67 1.80 2.45 1950 823 8.50 2 1 15.00 1 0"
`
require.NoError(t, os.WriteFile(intelPath, []byte(intelScript), 0755))
rocmPath := filepath.Join(dir, "rocm-smi")
rocmScript := `#!/bin/sh
echo '{"card0": {"Temperature (Sensor edge) (C)": "49.0", "Current Socket Graphics Package Power (W)": "28.159", "GPU use (%)": "0", "VRAM Total Memory (B)": "536870912", "VRAM Total Used Memory (B)": "445550592", "Card Series": "Rembrandt [Radeon 680M]", "GUID": "34756"}}'
`
require.NoError(t, os.WriteFile(rocmPath, []byte(rocmScript), 0755))
gm, err := NewGPUManager()
require.NoError(t, err)
require.NotNil(t, gm)
time.Sleep(150 * time.Millisecond)
_, intelOk := gm.GpuDataMap["i0"]
_, amdOk := gm.GpuDataMap["34756"]
assert.True(t, intelOk)
assert.True(t, amdOk)
}
func TestNewGPUManagerPriorityNvmlFallbackToNvidiaSmi(t *testing.T) {
dir := t.TempDir()
t.Setenv("PATH", dir)
t.Setenv("BESZEL_AGENT_GPU_COLLECTOR", "nvml,nvidia-smi")
nvidiaPath := filepath.Join(dir, "nvidia-smi")
nvidiaScript := `#!/bin/sh
echo "0, NVIDIA Fallback GPU, 41, 256, 1024, 8, 14"`
require.NoError(t, os.WriteFile(nvidiaPath, []byte(nvidiaScript), 0755))
gm, err := NewGPUManager()
require.NoError(t, err)
require.NotNil(t, gm)
time.Sleep(150 * time.Millisecond)
gpu, ok := gm.GpuDataMap["0"]
require.True(t, ok)
assert.Equal(t, "Fallback GPU", gpu.Name)
}
func TestNewGPUManagerConfiguredCollectorsMustStart(t *testing.T) {
dir := t.TempDir()
t.Setenv("PATH", dir)
t.Run("configured valid collector unavailable", func(t *testing.T) {
t.Setenv("BESZEL_AGENT_GPU_COLLECTOR", "nvidia-smi")
gm, err := NewGPUManager()
require.Nil(t, gm)
require.Error(t, err)
assert.Contains(t, err.Error(), "no configured GPU collectors are available")
})
t.Run("configured collector list has only unknown entries", func(t *testing.T) {
t.Setenv("BESZEL_AGENT_GPU_COLLECTOR", "bad,unknown")
gm, err := NewGPUManager()
require.Nil(t, gm)
require.Error(t, err)
assert.Contains(t, err.Error(), "no configured GPU collectors are available")
})
}
func TestCollectorDefinitionsNvmlDoesNotRequireNvidiaSmi(t *testing.T) {
gm := &GPUManager{}
definitions := gm.collectorDefinitions(gpuCapabilities{})
require.Contains(t, definitions, collectorSourceNVML)
assert.True(t, definitions[collectorSourceNVML].available)
}
func TestNewGPUManagerConfiguredNvmlBypassesCapabilityGate(t *testing.T) {
dir := t.TempDir()
t.Setenv("PATH", dir)
t.Setenv("BESZEL_AGENT_GPU_COLLECTOR", "nvml")
gm, err := NewGPUManager()
require.Nil(t, gm)
require.Error(t, err)
assert.Contains(t, err.Error(), "no configured GPU collectors are available")
assert.NotContains(t, err.Error(), noGPUFoundMsg)
}
func TestNewGPUManagerJetsonIgnoresCollectorConfig(t *testing.T) {
dir := t.TempDir()
t.Setenv("PATH", dir)
t.Setenv("BESZEL_AGENT_GPU_COLLECTOR", "nvidia-smi")
tegraPath := filepath.Join(dir, "tegrastats")
tegraScript := `#!/bin/sh
echo "11-14-2024 22:54:33 RAM 1024/4096MB GR3D_FREQ 80% tj@70C VDD_GPU_SOC 1000mW"`
require.NoError(t, os.WriteFile(tegraPath, []byte(tegraScript), 0755))
gm, err := NewGPUManager()
require.NoError(t, err)
require.NotNil(t, gm)
time.Sleep(100 * time.Millisecond)
gpu, ok := gm.GpuDataMap["0"]
require.True(t, ok)
assert.Equal(t, "GPU", gpu.Name)
}
// TestAccumulationTableDriven tests the accumulation behavior for all three GPU types // TestAccumulationTableDriven tests the accumulation behavior for all three GPU types
func TestAccumulation(t *testing.T) { func TestAccumulation(t *testing.T) {
type expectedGPUValues struct { type expectedGPUValues struct {
@@ -1721,7 +1346,7 @@ func TestIntelUpdateFromStats(t *testing.T) {
ok := gm.updateIntelFromStats(&sample1) ok := gm.updateIntelFromStats(&sample1)
assert.True(t, ok) assert.True(t, ok)
gpu := gm.GpuDataMap["i0"] gpu := gm.GpuDataMap["0"]
require.NotNil(t, gpu) require.NotNil(t, gpu)
assert.Equal(t, "GPU", gpu.Name) assert.Equal(t, "GPU", gpu.Name)
assert.EqualValues(t, 10.5, gpu.Power) assert.EqualValues(t, 10.5, gpu.Power)
@@ -1743,7 +1368,7 @@ func TestIntelUpdateFromStats(t *testing.T) {
ok = gm.updateIntelFromStats(&sample2) ok = gm.updateIntelFromStats(&sample2)
assert.True(t, ok) assert.True(t, ok)
gpu = gm.GpuDataMap["i0"] gpu = gm.GpuDataMap["0"]
require.NotNil(t, gpu) require.NotNil(t, gpu)
assert.EqualValues(t, 10.5, gpu.Power) assert.EqualValues(t, 10.5, gpu.Power)
assert.EqualValues(t, 30.0, gpu.Engines["Render/3D"]) // 20 + 10 assert.EqualValues(t, 30.0, gpu.Engines["Render/3D"]) // 20 + 10
@@ -1753,8 +1378,12 @@ func TestIntelUpdateFromStats(t *testing.T) {
} }
func TestIntelCollectorStreaming(t *testing.T) { func TestIntelCollectorStreaming(t *testing.T) {
// Save and override PATH
origPath := os.Getenv("PATH")
defer os.Setenv("PATH", origPath)
dir := t.TempDir() dir := t.TempDir()
t.Setenv("PATH", dir) os.Setenv("PATH", dir)
// Create a fake intel_gpu_top that prints -l format with four samples (first will be skipped) and exits // Create a fake intel_gpu_top that prints -l format with four samples (first will be skipped) and exits
scriptPath := filepath.Join(dir, "intel_gpu_top") scriptPath := filepath.Join(dir, "intel_gpu_top")
@@ -1778,7 +1407,7 @@ echo "298 295 278 51 2.20 3.12 1675 942 5.75 1 2 9.50
t.Fatalf("collectIntelStats error: %v", err) t.Fatalf("collectIntelStats error: %v", err)
} }
gpu := gm.GpuDataMap["i0"] gpu := gm.GpuDataMap["0"]
require.NotNil(t, gpu) require.NotNil(t, gpu)
// Power should be sum of samples 2-4 (first is skipped): 2.0 + 1.8 + 2.2 = 6.0 // Power should be sum of samples 2-4 (first is skipped): 2.0 + 1.8 + 2.2 = 6.0
assert.EqualValues(t, 6.0, gpu.Power) assert.EqualValues(t, 6.0, gpu.Power)
+6 -46
View File
@@ -7,10 +7,9 @@ import (
"github.com/fxamacker/cbor/v2" "github.com/fxamacker/cbor/v2"
"github.com/henrygd/beszel/internal/common" "github.com/henrygd/beszel/internal/common"
"github.com/henrygd/beszel/internal/entities/monitor"
"github.com/henrygd/beszel/internal/entities/smart" "github.com/henrygd/beszel/internal/entities/smart"
"log/slog" "golang.org/x/exp/slog"
) )
// HandlerContext provides context for request handlers // HandlerContext provides context for request handlers
@@ -52,8 +51,6 @@ func NewHandlerRegistry() *HandlerRegistry {
registry.Register(common.GetContainerInfo, &GetContainerInfoHandler{}) registry.Register(common.GetContainerInfo, &GetContainerInfoHandler{})
registry.Register(common.GetSmartData, &GetSmartDataHandler{}) registry.Register(common.GetSmartData, &GetSmartDataHandler{})
registry.Register(common.GetSystemdInfo, &GetSystemdInfoHandler{}) registry.Register(common.GetSystemdInfo, &GetSystemdInfoHandler{})
registry.Register(common.SyncNetworkMonitors, &SyncNetworkMonitorsHandler{})
registry.Register(common.GetZfsData, &GetZfsDataHandler{})
return registry return registry
} }
@@ -169,33 +166,14 @@ type GetSmartDataHandler struct{}
func (h *GetSmartDataHandler) Handle(hctx *HandlerContext) error { func (h *GetSmartDataHandler) Handle(hctx *HandlerContext) error {
if hctx.Agent.smartManager == nil { if hctx.Agent.smartManager == nil {
return hctx.SendResponse(smart.SmartDataResponse{Data: map[string]smart.SmartData{}}, hctx.RequestID) // return empty map to indicate no data
return hctx.SendResponse(map[string]smart.SmartData{}, hctx.RequestID)
} }
complete, err := hctx.Agent.smartManager.Refresh(false) if err := hctx.Agent.smartManager.Refresh(false); err != nil {
if err != nil {
slog.Debug("smart refresh failed", "err", err) slog.Debug("smart refresh failed", "err", err)
} }
return hctx.SendResponse(smart.SmartDataResponse{ data := hctx.Agent.smartManager.GetCurrentData()
Data: hctx.Agent.smartManager.GetCurrentData(), return hctx.SendResponse(data, hctx.RequestID)
Complete: complete,
}, hctx.RequestID)
}
////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////
// GetZfsDataHandler handles ZFS detail data requests
type GetZfsDataHandler struct{}
func (h *GetZfsDataHandler) Handle(hctx *HandlerContext) error {
if hctx.Agent.storagePoolManager == nil {
return hctx.SendResponse(nil, hctx.RequestID)
}
var req common.ZfsDataRequest
if err := cbor.Unmarshal(hctx.Request.Data, &req); err != nil {
return err
}
return hctx.SendResponse(hctx.Agent.storagePoolManager.GetDetail(req.Force), hctx.RequestID)
} }
//////////////////////////////////////////////////////////////////////////// ////////////////////////////////////////////////////////////////////////////
@@ -225,21 +203,3 @@ func (h *GetSystemdInfoHandler) Handle(hctx *HandlerContext) error {
return hctx.SendResponse(details, hctx.RequestID) return hctx.SendResponse(details, hctx.RequestID)
} }
////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////
// SyncNetworkMonitorsHandler handles monitor configuration sync from hub
type SyncNetworkMonitorsHandler struct{}
func (h *SyncNetworkMonitorsHandler) Handle(hctx *HandlerContext) error {
var req monitor.SyncRequest
if err := cbor.Unmarshal(hctx.Request.Data, &req); err != nil {
return err
}
resp, err := hctx.Agent.monitorManager.HandleSyncRequest(req)
if err != nil {
return err
}
return hctx.SendResponse(resp, hctx.RequestID)
}
+1 -41
View File
@@ -1,15 +1,13 @@
//go:build testing //go:build testing
// +build testing
package agent package agent
import ( import (
"testing" "testing"
"time"
"github.com/fxamacker/cbor/v2" "github.com/fxamacker/cbor/v2"
"github.com/henrygd/beszel/agent/zfs"
"github.com/henrygd/beszel/internal/common" "github.com/henrygd/beszel/internal/common"
"github.com/henrygd/beszel/internal/entities/smart"
"github.com/stretchr/testify/assert" "github.com/stretchr/testify/assert"
) )
@@ -20,44 +18,6 @@ type MockHandler struct {
handleFunc func(ctx *HandlerContext) error handleFunc func(ctx *HandlerContext) error
} }
func TestNewAgentResponseSmartData(t *testing.T) {
response := newAgentResponse(smart.SmartDataResponse{
Data: map[string]smart.SmartData{
"AAA": {SerialNumber: "AAA"},
},
Complete: true,
}, nil)
assert.Equal(t, "AAA", response.SmartData["AAA"].SerialNumber)
assert.True(t, response.SmartComplete)
}
func TestGetZfsDataHandlerForceRefresh(t *testing.T) {
poolCalls := 0
zm := &StoragePoolManager{detailInterval: time.Hour, backends: []*poolBackend{{name: "zfs"}}}
zm.backends[0].poolStatsFn = func() ([]zfs.PoolStat, error) {
poolCalls++
return []zfs.PoolStat{{Name: "tank", Alloc: uint64(poolCalls)}}, nil
}
zm.backends[0].poolStatusesFn = func() ([]zfs.PoolStatus, error) { return nil, nil }
zm.backends[0].datasetsFn = func() ([]zfs.Dataset, error) { return nil, nil }
zm.GetDetail(false)
requestData, err := cbor.Marshal(common.ZfsDataRequest{Force: true})
assert.NoError(t, err)
ctx := &HandlerContext{
Agent: &Agent{storagePoolManager: zm},
Request: &common.HubRequest[cbor.RawMessage]{
Action: common.GetZfsData,
Data: requestData,
},
SendResponse: func(any, *uint32) error { return nil },
}
assert.NoError(t, (&GetZfsDataHandler{}).Handle(ctx))
assert.Equal(t, 2, poolCalls)
}
func (m *MockHandler) Handle(ctx *HandlerContext) error { func (m *MockHandler) Handle(ctx *HandlerContext) error {
if m.handleFunc != nil { if m.handleFunc != nil {
return m.handleFunc(ctx) return m.handleFunc(ctx)
+6 -22
View File
@@ -9,31 +9,11 @@ import (
"log" "log"
"os" "os"
"path/filepath" "path/filepath"
"runtime"
"time" "time"
) )
// healthFile is the path to the health file // healthFile is the path to the health file
var healthFile = getHealthFilePath() var healthFile = filepath.Join(os.TempDir(), "beszel_health")
func getHealthFilePath() string {
filename := "beszel_health"
if runtime.GOOS == "linux" {
fullPath := filepath.Join("/dev/shm", filename)
if err := updateHealthFile(fullPath); err == nil {
return fullPath
}
}
return filepath.Join(os.TempDir(), filename)
}
func updateHealthFile(path string) error {
file, err := os.Create(path)
if err != nil {
return err
}
return file.Close()
}
// Check checks if the agent is connected by checking the modification time of the health file // Check checks if the agent is connected by checking the modification time of the health file
func Check() error { func Check() error {
@@ -50,7 +30,11 @@ func Check() error {
// Update updates the modification time of the health file // Update updates the modification time of the health file
func Update() error { func Update() error {
return updateHealthFile(healthFile) file, err := os.Create(healthFile)
if err != nil {
return err
}
return file.Close()
} }
// CleanUp removes the health file // CleanUp removes the health file
+2
View File
@@ -1,4 +1,5 @@
//go:build testing //go:build testing
// +build testing
package health package health
@@ -36,6 +37,7 @@ func TestHealth(t *testing.T) {
}) })
// This test uses synctest to simulate time passing. // This test uses synctest to simulate time passing.
// NOTE: This test requires GOEXPERIMENT=synctest to run.
t.Run("check with simulated time", func(t *testing.T) { t.Run("check with simulated time", func(t *testing.T) {
synctest.Test(t, func(t *testing.T) { synctest.Test(t, func(t *testing.T) {
// Update the file to set the initial timestamp. // Update the file to set the initial timestamp.
+1 -6
View File
@@ -52,12 +52,7 @@ class Program
foreach (var sensor in hardware.Sensors) foreach (var sensor in hardware.Sensors)
{ {
var validTemp = sensor.SensorType == SensorType.Temperature && sensor.Value.HasValue; var validTemp = sensor.SensorType == SensorType.Temperature && sensor.Value.HasValue;
if (!validTemp || if (!validTemp || sensor.Name.Contains("Distance"))
sensor.Name.IndexOf("Distance", StringComparison.OrdinalIgnoreCase) >= 0 ||
sensor.Name.IndexOf("Limit", StringComparison.OrdinalIgnoreCase) >= 0 ||
sensor.Name.IndexOf("Critical", StringComparison.OrdinalIgnoreCase) >= 0 ||
sensor.Name.IndexOf("Warning", StringComparison.OrdinalIgnoreCase) >= 0 ||
sensor.Name.IndexOf("Resolution", StringComparison.OrdinalIgnoreCase) >= 0)
{ {
continue; continue;
} }
+1 -3
View File
@@ -3,11 +3,9 @@
<OutputType>Exe</OutputType> <OutputType>Exe</OutputType>
<TargetFramework>net48</TargetFramework> <TargetFramework>net48</TargetFramework>
<Platforms>x64</Platforms> <Platforms>x64</Platforms>
<RuntimeIdentifier>win-x64</RuntimeIdentifier>
<AppendRuntimeIdentifierToOutputPath>false</AppendRuntimeIdentifierToOutputPath>
</PropertyGroup> </PropertyGroup>
<ItemGroup> <ItemGroup>
<PackageReference Include="LibreHardwareMonitorLib" Version="0.9.6" /> <PackageReference Include="LibreHardwareMonitorLib" Version="0.9.4" />
</ItemGroup> </ItemGroup>
</Project> </Project>
-293
View File
@@ -1,293 +0,0 @@
//go:build linux
package agent
import (
"fmt"
"os"
"path/filepath"
"strconv"
"strings"
"github.com/henrygd/beszel/agent/utils"
"github.com/henrygd/beszel/internal/entities/smart"
)
// mdraidSysfsRoot is a test hook; production value is "/sys".
var mdraidSysfsRoot = "/sys"
type mdraidHealth struct {
level string
arrayState string
degraded uint64
faultyDisks uint64
populatedDisks uint64
raidDisks uint64
syncAction string
syncCompleted string
syncSpeed string
mismatchCnt uint64
capacity uint64
}
// scanMdraidDevices discovers Linux md arrays exposed in sysfs.
func scanMdraidDevices() []*DeviceInfo {
blockDir := filepath.Join(mdraidSysfsRoot, "block")
entries, err := os.ReadDir(blockDir)
if err != nil {
return nil
}
devices := make([]*DeviceInfo, 0, 2)
for _, ent := range entries {
name := ent.Name()
if !isMdraidBlockName(name) {
continue
}
mdDir := filepath.Join(blockDir, name, "md")
if !utils.FileExists(filepath.Join(mdDir, "array_state")) {
continue
}
devPath := filepath.Join("/dev", name)
devices = append(devices, &DeviceInfo{
Name: devPath,
Type: "mdraid",
InfoName: devPath + " [mdraid]",
Protocol: "MD",
})
}
return devices
}
// collectMdraidHealth reads mdraid health and stores it in SmartDataMap.
func (sm *SmartManager) collectMdraidHealth(deviceInfo *DeviceInfo) (bool, error) {
if deviceInfo == nil || deviceInfo.Name == "" {
return false, nil
}
base := filepath.Base(deviceInfo.Name)
if !isMdraidBlockName(base) && !strings.EqualFold(deviceInfo.Type, "mdraid") {
return false, nil
}
health, ok := readMdraidHealth(base)
if !ok {
return false, nil
}
deviceInfo.Type = "mdraid"
key := fmt.Sprintf("mdraid:%s", base)
status := mdraidSmartStatus(health)
attrs := make([]*smart.SmartAttribute, 0, 10)
if health.arrayState != "" {
attrs = append(attrs, &smart.SmartAttribute{Name: "ArrayState", RawString: health.arrayState})
}
if health.level != "" {
attrs = append(attrs, &smart.SmartAttribute{Name: "RaidLevel", RawString: health.level})
}
if health.raidDisks > 0 {
attrs = append(attrs, &smart.SmartAttribute{Name: "RaidDisks", RawValue: health.raidDisks})
}
if health.degraded > 0 {
attrs = append(attrs, &smart.SmartAttribute{Name: "Degraded", RawValue: health.degraded})
}
if health.faultyDisks > 0 {
attrs = append(attrs, &smart.SmartAttribute{Name: "FaultyDisks", RawValue: health.faultyDisks})
}
if health.syncAction != "" {
attrs = append(attrs, &smart.SmartAttribute{Name: "SyncAction", RawString: health.syncAction})
}
if health.syncCompleted != "" {
attrs = append(attrs, &smart.SmartAttribute{Name: "SyncCompleted", RawString: health.syncCompleted})
}
if health.syncSpeed != "" {
attrs = append(attrs, &smart.SmartAttribute{Name: "SyncSpeed", RawString: health.syncSpeed})
}
if health.mismatchCnt > 0 {
attrs = append(attrs, &smart.SmartAttribute{Name: "MismatchCount", RawValue: health.mismatchCnt})
}
sm.Lock()
defer sm.Unlock()
if _, exists := sm.SmartDataMap[key]; !exists {
sm.SmartDataMap[key] = &smart.SmartData{}
}
data := sm.SmartDataMap[key]
data.ModelName = "Linux MD RAID"
if health.level != "" {
data.ModelName = "Linux MD RAID (" + health.level + ")"
}
data.Capacity = health.capacity
data.SmartStatus = status
data.DiskName = filepath.Join("/dev", base)
data.DiskType = "mdraid"
data.Attributes = attrs
return true, nil
}
// readMdraidHealth reads md array health fields from sysfs.
func readMdraidHealth(blockName string) (mdraidHealth, bool) {
var out mdraidHealth
if !isMdraidBlockName(blockName) {
return out, false
}
mdDir := filepath.Join(mdraidSysfsRoot, "block", blockName, "md")
arrayState, okState := utils.ReadStringFileOK(filepath.Join(mdDir, "array_state"))
if !okState {
return out, false
}
out.arrayState = arrayState
out.level = utils.ReadStringFile(filepath.Join(mdDir, "level"))
out.syncAction = utils.ReadStringFile(filepath.Join(mdDir, "sync_action"))
out.syncCompleted = utils.ReadStringFile(filepath.Join(mdDir, "sync_completed"))
out.syncSpeed = utils.ReadStringFile(filepath.Join(mdDir, "sync_speed"))
if val, ok := utils.ReadUintFile(filepath.Join(mdDir, "raid_disks")); ok {
out.raidDisks = val
}
if val, ok := utils.ReadUintFile(filepath.Join(mdDir, "degraded")); ok {
out.degraded = val
}
out.faultyDisks, out.populatedDisks = countMdraidMemberStates(blockName, mdraidSysfsRoot)
if val, ok := utils.ReadUintFile(filepath.Join(mdDir, "mismatch_cnt")); ok {
out.mismatchCnt = val
}
if capBytes, ok := readMdraidBlockCapacityBytes(blockName, mdraidSysfsRoot); ok {
out.capacity = capBytes
}
return out, true
}
// mdraidSmartStatus maps md state/sync signals to a SMART-like status.
func mdraidSmartStatus(health mdraidHealth) string {
state := strings.ToLower(strings.TrimSpace(health.arrayState))
switch state {
case "inactive", "faulty", "broken", "stopped":
return "FAILED"
}
// During rebuild/recovery, arrays are often temporarily degraded; report as
// warning instead of hard failure while synchronization is in progress.
syncAction := strings.ToLower(strings.TrimSpace(health.syncAction))
switch syncAction {
case "resync", "recover", "reshape":
return "WARNING"
}
// Use actual faulty member count rather than the degraded counter, which
// equals raid_disks minus active_disks. On QNAP systems raid_disks may be
// set to a large value (e.g. 32) while only a few slots are ever used,
// making degraded misleadingly large despite zero failed disks.
if health.faultyDisks > 0 {
return "FAILED"
}
if health.degraded > 0 {
if isSparseSlotDegraded(health) {
// A sysfs snapshot cannot distinguish reserved slots from a removed
// member on sparse arrays, so report the ambiguity as a warning.
return "WARNING"
}
return "FAILED"
}
if health.mismatchCnt > 0 {
return "WARNING"
}
// "check" scans for consistency problems without repairing mismatches.
// With no mismatches, keep it green while reporting progress attributes.
switch syncAction {
case "repair":
return "WARNING"
}
switch state {
case "clean", "active", "active-idle", "write-pending", "read-auto", "readonly":
return "PASSED"
}
return "UNKNOWN"
}
// countMdraidMemberStates reads member device directories under
// block/<name>/md and returns how many are explicitly marked "faulty", plus
// how many are populated at all (regardless of state). populatedDisks lets
// callers distinguish RAID slots that were never used (QNAP reserves far
// more raid_disks than it ever populates) from members that went missing.
func countMdraidMemberStates(blockName, root string) (faultyDisks, populatedDisks uint64) {
devDir := filepath.Join(root, "block", blockName, "md")
entries, err := os.ReadDir(devDir)
if err != nil {
return 0, 0
}
for _, ent := range entries {
if !strings.HasPrefix(ent.Name(), "dev-") {
continue
}
populatedDisks++
statePath := filepath.Join(devDir, ent.Name(), "state")
state := utils.ReadStringFile(statePath)
if strings.Contains(state, "faulty") {
faultyDisks++
}
}
return faultyDisks, populatedDisks
}
// isSparseSlotDegraded reports whether a non-zero "degraded" count may be
// explained by RAID slots that were never populated. QNAP configures system
// arrays with raid_disks set to a large fixed maximum (e.g. 32) far beyond the
// handful of slots it ever populates, so sparse slots outnumber populated ones.
func isSparseSlotDegraded(health mdraidHealth) bool {
if health.populatedDisks == 0 || health.raidDisks <= health.populatedDisks {
return false
}
sparseSlots := health.raidDisks - health.populatedDisks
return sparseSlots > health.populatedDisks
}
// isMdraidBlockName matches /dev/mdN-style block device names.
func isMdraidBlockName(name string) bool {
if !strings.HasPrefix(name, "md") {
return false
}
suffix := strings.TrimPrefix(name, "md")
if suffix == "" {
return false
}
for _, c := range suffix {
if c < '0' || c > '9' {
return false
}
}
return true
}
// readMdraidBlockCapacityBytes converts block size metadata into bytes.
func readMdraidBlockCapacityBytes(blockName, root string) (uint64, bool) {
sizePath := filepath.Join(root, "block", blockName, "size")
lbsPath := filepath.Join(root, "block", blockName, "queue", "logical_block_size")
sizeStr, ok := utils.ReadStringFileOK(sizePath)
if !ok {
return 0, false
}
sectors, err := strconv.ParseUint(sizeStr, 10, 64)
if err != nil || sectors == 0 {
return 0, false
}
logicalBlockSize := uint64(512)
if lbsStr, ok := utils.ReadStringFileOK(lbsPath); ok {
if parsed, err := strconv.ParseUint(lbsStr, 10, 64); err == nil && parsed > 0 {
logicalBlockSize = parsed
}
}
return sectors * logicalBlockSize, true
}
-186
View File
@@ -1,186 +0,0 @@
//go:build linux
package agent
import (
"os"
"path/filepath"
"testing"
"github.com/henrygd/beszel/internal/entities/smart"
)
func TestMdraidMockSysfsScanAndCollect(t *testing.T) {
tmp := t.TempDir()
prev := mdraidSysfsRoot
mdraidSysfsRoot = tmp
t.Cleanup(func() { mdraidSysfsRoot = prev })
mdDir := filepath.Join(tmp, "block", "md0", "md")
queueDir := filepath.Join(tmp, "block", "md0", "queue")
if err := os.MkdirAll(mdDir, 0o755); err != nil {
t.Fatal(err)
}
if err := os.MkdirAll(queueDir, 0o755); err != nil {
t.Fatal(err)
}
write := func(path, content string) {
t.Helper()
if err := os.WriteFile(path, []byte(content), 0o644); err != nil {
t.Fatal(err)
}
}
write(filepath.Join(mdDir, "array_state"), "active\n")
write(filepath.Join(mdDir, "level"), "raid1\n")
write(filepath.Join(mdDir, "raid_disks"), "2\n")
write(filepath.Join(mdDir, "degraded"), "0\n")
write(filepath.Join(mdDir, "sync_action"), "resync\n")
write(filepath.Join(mdDir, "sync_completed"), "10%\n")
write(filepath.Join(mdDir, "sync_speed"), "100M\n")
write(filepath.Join(mdDir, "mismatch_cnt"), "0\n")
// Simulate two healthy member devices (no faulty state).
for _, dev := range []string{"dev-sda", "dev-sdb"} {
devPath := filepath.Join(mdDir, dev)
if err := os.MkdirAll(devPath, 0o755); err != nil {
t.Fatal(err)
}
write(filepath.Join(devPath, "state"), "in_sync\n")
}
write(filepath.Join(queueDir, "logical_block_size"), "512\n")
write(filepath.Join(tmp, "block", "md0", "size"), "2048\n")
devs := scanMdraidDevices()
if len(devs) != 1 {
t.Fatalf("scanMdraidDevices() = %d devices, want 1", len(devs))
}
if devs[0].Name != "/dev/md0" || devs[0].Type != "mdraid" {
t.Fatalf("scanMdraidDevices()[0] = %+v, want Name=/dev/md0 Type=mdraid", devs[0])
}
sm := &SmartManager{SmartDataMap: map[string]*smart.SmartData{}}
ok, err := sm.collectMdraidHealth(devs[0])
if err != nil || !ok {
t.Fatalf("collectMdraidHealth() = (ok=%v, err=%v), want (true,nil)", ok, err)
}
if len(sm.SmartDataMap) != 1 {
t.Fatalf("SmartDataMap len=%d, want 1", len(sm.SmartDataMap))
}
var got *smart.SmartData
for _, v := range sm.SmartDataMap {
got = v
break
}
if got == nil {
t.Fatalf("SmartDataMap value nil")
}
if got.DiskType != "mdraid" || got.DiskName != "/dev/md0" {
t.Fatalf("disk fields = (type=%q name=%q), want (mdraid,/dev/md0)", got.DiskType, got.DiskName)
}
if got.SmartStatus != "WARNING" {
t.Fatalf("SmartStatus=%q, want WARNING", got.SmartStatus)
}
if got.ModelName == "" || got.Capacity == 0 {
t.Fatalf("identity fields = (model=%q cap=%d), want non-empty model and cap>0", got.ModelName, got.Capacity)
}
if len(got.Attributes) < 5 {
t.Fatalf("attributes len=%d, want >= 5", len(got.Attributes))
}
}
func TestCountMdraidMemberStates(t *testing.T) {
tmp := t.TempDir()
write := func(path, content string) {
t.Helper()
if err := os.MkdirAll(filepath.Dir(path), 0o755); err != nil {
t.Fatal(err)
}
if err := os.WriteFile(path, []byte(content), 0o644); err != nil {
t.Fatal(err)
}
}
mdDir := filepath.Join(tmp, "block", "md0", "md")
// No dev-* entries: zero faulty, zero populated.
if faulty, populated := countMdraidMemberStates("md0", tmp); faulty != 0 || populated != 0 {
t.Fatalf("no members: got (faulty=%d populated=%d), want (0,0)", faulty, populated)
}
// Two healthy members.
write(filepath.Join(mdDir, "dev-sda", "state"), "in_sync\n")
write(filepath.Join(mdDir, "dev-sdb", "state"), "in_sync\n")
if faulty, populated := countMdraidMemberStates("md0", tmp); faulty != 0 || populated != 2 {
t.Fatalf("all in_sync: got (faulty=%d populated=%d), want (0,2)", faulty, populated)
}
// One faulty member.
write(filepath.Join(mdDir, "dev-sdb", "state"), "faulty\n")
if faulty, populated := countMdraidMemberStates("md0", tmp); faulty != 1 || populated != 2 {
t.Fatalf("one faulty: got (faulty=%d populated=%d), want (1,2)", faulty, populated)
}
// QNAP-style: 28 degraded slots but no dev-* entries for them, 4 in_sync.
write(filepath.Join(mdDir, "dev-sdb", "state"), "in_sync\n")
write(filepath.Join(mdDir, "dev-sdc", "state"), "in_sync\n")
write(filepath.Join(mdDir, "dev-sdd", "state"), "in_sync\n")
if faulty, populated := countMdraidMemberStates("md0", tmp); faulty != 0 || populated != 4 {
t.Fatalf("qnap sparse: got (faulty=%d populated=%d), want (0,4)", faulty, populated)
}
}
func TestMdraidSmartStatus(t *testing.T) {
if got := mdraidSmartStatus(mdraidHealth{arrayState: "inactive"}); got != "FAILED" {
t.Fatalf("mdraidSmartStatus(inactive) = %q, want FAILED", got)
}
if got := mdraidSmartStatus(mdraidHealth{arrayState: "active", degraded: 1, faultyDisks: 1, syncAction: "recover"}); got != "WARNING" {
t.Fatalf("mdraidSmartStatus(degraded+recover) = %q, want WARNING", got)
}
if got := mdraidSmartStatus(mdraidHealth{arrayState: "active", degraded: 1, faultyDisks: 1}); got != "FAILED" {
t.Fatalf("mdraidSmartStatus(degraded+faulty) = %q, want FAILED", got)
}
// QNAP-style: raid_disks=32 but only 4 populated; degraded=28 but no faulty devices.
if got := mdraidSmartStatus(mdraidHealth{arrayState: "clean", degraded: 28, faultyDisks: 0, raidDisks: 32, populatedDisks: 4}); got != "WARNING" {
t.Fatalf("mdraidSmartStatus(qnap sparse) = %q, want WARNING", got)
}
// A member disappearing from the same sparse array is indistinguishable
// from another reserved slot, so it must not be reported as healthy.
if got := mdraidSmartStatus(mdraidHealth{arrayState: "clean", degraded: 29, faultyDisks: 0, raidDisks: 32, populatedDisks: 3}); got != "WARNING" {
t.Fatalf("mdraidSmartStatus(qnap sparse missing member) = %q, want WARNING", got)
}
// A genuinely missing member (removed dev-* entry, not just an unpopulated
// QNAP reserve slot) must still fail: raid_disks=4, only 3 populated, all
// of them in_sync, so faultyDisks==0 but degraded==1.
if got := mdraidSmartStatus(mdraidHealth{arrayState: "clean", degraded: 1, faultyDisks: 0, raidDisks: 4, populatedDisks: 3}); got != "FAILED" {
t.Fatalf("mdraidSmartStatus(missing member) = %q, want FAILED", got)
}
// Degraded with no member-state info at all (e.g. sysfs read failed) must
// still fail rather than being silently treated as a sparse QNAP array.
if got := mdraidSmartStatus(mdraidHealth{arrayState: "clean", degraded: 1, faultyDisks: 0, raidDisks: 4, populatedDisks: 0}); got != "FAILED" {
t.Fatalf("mdraidSmartStatus(degraded, no member info) = %q, want FAILED", got)
}
if got := mdraidSmartStatus(mdraidHealth{arrayState: "active", syncAction: "recover"}); got != "WARNING" {
t.Fatalf("mdraidSmartStatus(recover) = %q, want WARNING", got)
}
if got := mdraidSmartStatus(mdraidHealth{arrayState: "clean", syncAction: "check"}); got != "PASSED" {
t.Fatalf("mdraidSmartStatus(clean+check) = %q, want PASSED", got)
}
if got := mdraidSmartStatus(mdraidHealth{arrayState: "clean", syncAction: "check", mismatchCnt: 1}); got != "WARNING" {
t.Fatalf("mdraidSmartStatus(clean+check+mismatch) = %q, want WARNING", got)
}
if got := mdraidSmartStatus(mdraidHealth{arrayState: "clean", mismatchCnt: 1}); got != "WARNING" {
t.Fatalf("mdraidSmartStatus(clean+mismatch) = %q, want WARNING", got)
}
if got := mdraidSmartStatus(mdraidHealth{arrayState: "clean", syncAction: "repair"}); got != "WARNING" {
t.Fatalf("mdraidSmartStatus(repair) = %q, want WARNING", got)
}
if got := mdraidSmartStatus(mdraidHealth{arrayState: "clean"}); got != "PASSED" {
t.Fatalf("mdraidSmartStatus(clean) = %q, want PASSED", got)
}
if got := mdraidSmartStatus(mdraidHealth{arrayState: "unknown"}); got != "UNKNOWN" {
t.Fatalf("mdraidSmartStatus(unknown) = %q, want UNKNOWN", got)
}
}
-11
View File
@@ -1,11 +0,0 @@
//go:build !linux
package agent
func scanMdraidDevices() []*DeviceInfo {
return nil
}
func (sm *SmartManager) collectMdraidHealth(deviceInfo *DeviceInfo) (bool, error) {
return false, nil
}
+14 -18
View File
@@ -8,7 +8,6 @@ import (
"time" "time"
"github.com/henrygd/beszel/agent/deltatracker" "github.com/henrygd/beszel/agent/deltatracker"
"github.com/henrygd/beszel/agent/utils"
"github.com/henrygd/beszel/internal/entities/system" "github.com/henrygd/beszel/internal/entities/system"
psutilNet "github.com/shirou/gopsutil/v4/net" psutilNet "github.com/shirou/gopsutil/v4/net"
) )
@@ -95,7 +94,7 @@ func (a *Agent) initializeNetIoStats() {
a.netInterfaces = make(map[string]struct{}, 0) a.netInterfaces = make(map[string]struct{}, 0)
// parse NICS env var for whitelist / blacklist // parse NICS env var for whitelist / blacklist
nicsEnvVal, nicsEnvExists := utils.GetEnv("NICS") nicsEnvVal, nicsEnvExists := GetEnv("NICS")
var nicCfg *NicConfig var nicCfg *NicConfig
if nicsEnvExists { if nicsEnvExists {
nicCfg = newNicConfig(nicsEnvVal) nicCfg = newNicConfig(nicsEnvVal)
@@ -104,7 +103,10 @@ func (a *Agent) initializeNetIoStats() {
// get current network I/O stats and record valid interfaces // get current network I/O stats and record valid interfaces
if netIO, err := psutilNet.IOCounters(true); err == nil { if netIO, err := psutilNet.IOCounters(true); err == nil {
for _, v := range netIO { for _, v := range netIO {
if skipNetworkInterface(v, nicCfg) { if nicsEnvExists && !isValidNic(v.Name, nicCfg) {
continue
}
if a.skipNetworkInterface(v) {
continue continue
} }
slog.Info("Detected network interface", "name", v.Name, "sent", v.BytesSent, "recv", v.BytesRecv) slog.Info("Detected network interface", "name", v.Name, "sent", v.BytesSent, "recv", v.BytesRecv)
@@ -213,8 +215,10 @@ func (a *Agent) applyNetworkTotals(
totalBytesSent, totalBytesRecv uint64, totalBytesSent, totalBytesRecv uint64,
bytesSentPerSecond, bytesRecvPerSecond uint64, bytesSentPerSecond, bytesRecvPerSecond uint64,
) { ) {
if bytesSentPerSecond > 10_000_000_000 || bytesRecvPerSecond > 10_000_000_000 { networkSentPs := bytesToMegabytes(float64(bytesSentPerSecond))
slog.Warn("Invalid net stats. Resetting.", "sent", bytesSentPerSecond, "recv", bytesRecvPerSecond) networkRecvPs := bytesToMegabytes(float64(bytesRecvPerSecond))
if networkSentPs > 10_000 || networkRecvPs > 10_000 {
slog.Warn("Invalid net stats. Resetting.", "sent", networkSentPs, "recv", networkRecvPs)
for _, v := range netIO { for _, v := range netIO {
if _, exists := a.netInterfaces[v.Name]; !exists { if _, exists := a.netInterfaces[v.Name]; !exists {
continue continue
@@ -224,29 +228,21 @@ func (a *Agent) applyNetworkTotals(
a.initializeNetIoStats() a.initializeNetIoStats()
delete(a.netIoStats, cacheTimeMs) delete(a.netIoStats, cacheTimeMs)
delete(a.netInterfaceDeltaTrackers, cacheTimeMs) delete(a.netInterfaceDeltaTrackers, cacheTimeMs)
systemStats.NetworkSent = 0
systemStats.NetworkRecv = 0
systemStats.Bandwidth[0], systemStats.Bandwidth[1] = 0, 0 systemStats.Bandwidth[0], systemStats.Bandwidth[1] = 0, 0
return return
} }
systemStats.NetworkSent = networkSentPs
systemStats.NetworkRecv = networkRecvPs
systemStats.Bandwidth[0], systemStats.Bandwidth[1] = bytesSentPerSecond, bytesRecvPerSecond systemStats.Bandwidth[0], systemStats.Bandwidth[1] = bytesSentPerSecond, bytesRecvPerSecond
nis.BytesSent = totalBytesSent nis.BytesSent = totalBytesSent
nis.BytesRecv = totalBytesRecv nis.BytesRecv = totalBytesRecv
a.netIoStats[cacheTimeMs] = nis a.netIoStats[cacheTimeMs] = nis
} }
// skipNetworkInterface returns true if the network interface should be ignored. func (a *Agent) skipNetworkInterface(v psutilNet.IOCountersStat) bool {
func skipNetworkInterface(v psutilNet.IOCountersStat, nicCfg *NicConfig) bool {
if nicCfg != nil {
if !isValidNic(v.Name, nicCfg) {
return true
}
// In whitelist mode, we honor explicit inclusion without auto-filtering.
if !nicCfg.isBlacklist {
return false
}
// In blacklist mode, still apply the auto-filter below.
}
switch { switch {
case strings.HasPrefix(v.Name, "lo"), case strings.HasPrefix(v.Name, "lo"),
strings.HasPrefix(v.Name, "docker"), strings.HasPrefix(v.Name, "docker"),
-176
View File
@@ -1,176 +0,0 @@
package agent
import (
"errors"
"fmt"
"net/http"
"sync"
"time"
"github.com/henrygd/beszel/internal/entities/monitor"
)
// MonitorManager manages network monitor configurations and task lifetimes.
type MonitorManager struct {
mu sync.RWMutex
monitors map[string]*monitorTask // keyed by monitor ID
probe monitorProbe
resumeGuard monitorResumeGuard
}
func newMonitorManager() *MonitorManager {
return newMonitorManagerWithProbe(networkMonitorProbe(&http.Client{Timeout: monitor.MaxProbeTimeout}))
}
func newMonitorManagerWithProbe(probe monitorProbe) *MonitorManager {
return &MonitorManager{monitors: make(map[string]*monitorTask), probe: probe}
}
// SyncMonitors replaces all monitor tasks with the given configs.
func (pm *MonitorManager) SyncMonitors(configs []monitor.Config) {
pm.mu.Lock()
defer pm.mu.Unlock()
// Build set of new keys
newKeys := make(map[string]monitor.Config, len(configs))
for _, cfg := range configs {
if cfg.ID == "" {
continue
}
newKeys[cfg.ID] = cfg
}
// Stop removed monitors
for key, task := range pm.monitors {
if _, exists := newKeys[key]; !exists {
task.cancel()
delete(pm.monitors, key)
}
}
// Start new monitors and restart tasks whose config changed.
for key, cfg := range newKeys {
task, exists := pm.monitors[key]
if exists && task.config == cfg {
continue
}
if exists {
task.cancel()
}
task = newMonitorTaskFromExisting(cfg, task)
task.resumeGuard = &pm.resumeGuard
pm.resumeGuard.start()
pm.monitors[key] = task
pm.startMonitor(task)
}
if len(pm.monitors) == 0 {
pm.resumeGuard.shutdown()
}
}
// HandleSyncRequest applies a full or incremental monitor sync request.
func (pm *MonitorManager) HandleSyncRequest(req monitor.SyncRequest) (monitor.SyncResponse, error) {
switch req.Action {
case monitor.SyncActionReplace:
pm.SyncMonitors(req.Configs)
return monitor.SyncResponse{}, nil
case monitor.SyncActionUpsert:
result, err := pm.UpsertMonitor(req.Config, req.RunNow)
if err != nil {
return monitor.SyncResponse{}, err
}
if result == nil {
return monitor.SyncResponse{}, nil
}
return monitor.SyncResponse{Result: *result}, nil
case monitor.SyncActionDelete:
if req.Config.ID == "" {
return monitor.SyncResponse{}, errors.New("missing monitor ID for delete")
}
pm.DeleteMonitor(req.Config.ID)
return monitor.SyncResponse{}, nil
default:
return monitor.SyncResponse{}, fmt.Errorf("unknown monitor sync action: %d", req.Action)
}
}
// UpsertMonitor creates or replaces a single monitor task.
func (pm *MonitorManager) UpsertMonitor(config monitor.Config, runNow bool) (*monitor.Result, error) {
if config.ID == "" {
return nil, errors.New("missing monitor ID")
}
pm.mu.Lock()
task, exists := pm.monitors[config.ID]
if exists && task.config == config {
pm.mu.Unlock()
if !runNow {
return nil, nil
}
return task.runProbe(pm.probe), nil
}
if exists {
task.cancel()
}
task = newMonitorTaskFromExisting(config, task)
task.resumeGuard = &pm.resumeGuard
pm.resumeGuard.start()
pm.monitors[config.ID] = task
pm.mu.Unlock()
if runNow {
result := task.runProbe(pm.probe)
pm.startMonitor(task)
return result, nil
}
pm.startMonitor(task)
return nil, nil
}
// DeleteMonitor stops and removes a single monitor task.
func (pm *MonitorManager) DeleteMonitor(id string) {
if id == "" {
return
}
pm.mu.Lock()
defer pm.mu.Unlock()
if task, exists := pm.monitors[id]; exists {
task.cancel()
delete(pm.monitors, id)
}
if len(pm.monitors) == 0 {
pm.resumeGuard.shutdown()
}
}
// GetResults returns aggregated results for all monitors over the last supplied duration in ms.
func (pm *MonitorManager) GetResults(durationMs uint16) map[string]monitor.Result {
pm.mu.RLock()
defer pm.mu.RUnlock()
results := make(map[string]monitor.Result, len(pm.monitors))
now := time.Now()
duration := time.Duration(durationMs) * time.Millisecond
for _, task := range pm.monitors {
result, ok := task.history.result(duration, now)
if !ok {
continue
}
results[task.config.ID] = result
}
return results
}
// Stop stops all monitor tasks.
func (pm *MonitorManager) Stop() {
pm.mu.Lock()
defer pm.mu.Unlock()
for key, task := range pm.monitors {
task.cancel()
delete(pm.monitors, key)
}
pm.resumeGuard.shutdown()
}
-274
View File
@@ -1,274 +0,0 @@
package agent
import (
"math"
"sync"
"time"
"github.com/henrygd/beszel/internal/entities/monitor"
)
// Monitors run at user-defined intervals (e.g., every 10s).
// To keep memory usage low and constant, data is stored in two layers:
// 1. Raw samples: The most recent individual results (kept for monitorRawRetention).
// 2. Minute buckets: A ring buffer of 61 buckets, each representing one
// wall-clock minute. Samples collected within the same minute are aggregated
// (sum, min, max, count) into a single bucket.
//
// Short-term requests (<= 61s) use raw samples.
// Long-term requests (up to 1h) use the minute buckets to avoid storing thousands
// of individual data points.
const (
// monitorRawRetention is the duration to keep individual samples
monitorRawRetention = 61 * time.Second
// monitorMinuteBucketLen is the number of 1-minute buckets to keep (1 hour + 1 for partials)
monitorMinuteBucketLen int32 = 61
)
// monitorHistory owns retention and aggregation, independently of probe execution.
type monitorHistory struct {
mu sync.Mutex
sampleCount int64
samples []monitorSample
buckets [monitorMinuteBucketLen]monitorBucket
}
func newMonitorHistory() *monitorHistory {
// Start small for typical intervals; append grows the buffer for faster probes.
return &monitorHistory{samples: make([]monitorSample, 0, 4)}
}
func (h *monitorHistory) clone() *monitorHistory {
h.mu.Lock()
defer h.mu.Unlock()
cloned := newMonitorHistory()
cloned.samples = append(cloned.samples, h.samples...)
cloned.buckets = h.buckets
cloned.sampleCount = h.sampleCount
return cloned
}
func (h *monitorHistory) result(duration time.Duration, now time.Time) (monitor.Result, bool) {
h.mu.Lock()
defer h.mu.Unlock()
return h.resultLocked(duration, now)
}
func (h *monitorHistory) record(sample monitorSample) monitor.Result {
h.mu.Lock()
defer h.mu.Unlock()
h.addSampleLocked(sample)
result, _ := h.resultLocked(time.Minute, sample.timestamp)
return result
}
// monitorSample stores one monitor attempt and its collection time.
type monitorSample struct {
responseUs int64 // -1 means loss
timestamp time.Time
}
// monitorBucket stores one minute of aggregated monitor data.
type monitorBucket struct {
minute int32
filled bool
stats monitorAggregate
}
// monitorAggregate accumulates successful response stats and total sample counts.
type monitorAggregate struct {
sumUs int64
minUs int64
maxUs int64
totalCount int64
successCount int64
}
// newMonitorAggregate initializes an aggregate with an unset minimum value.
func newMonitorAggregate() monitorAggregate {
return monitorAggregate{minUs: math.MaxInt64}
}
// addResponse folds a single monitor sample into the aggregate.
func (agg *monitorAggregate) addResponse(responseUs int64) {
agg.totalCount++
if responseUs < 0 {
return
}
agg.successCount++
agg.sumUs += responseUs
if responseUs < agg.minUs {
agg.minUs = responseUs
}
if responseUs > agg.maxUs {
agg.maxUs = responseUs
}
}
// addAggregate merges another aggregate into this one.
func (agg *monitorAggregate) addAggregate(other monitorAggregate) {
if other.totalCount == 0 {
return
}
agg.totalCount += other.totalCount
agg.successCount += other.successCount
agg.sumUs += other.sumUs
if other.successCount == 0 {
return
}
if agg.minUs == math.MaxInt64 || other.minUs < agg.minUs {
agg.minUs = other.minUs
}
if other.maxUs > agg.maxUs {
agg.maxUs = other.maxUs
}
}
// hasData reports whether the aggregate contains any samples.
func (agg monitorAggregate) hasData() bool {
return agg.totalCount > 0
}
// result converts the aggregate into the monitor result format.
func (agg monitorAggregate) result() monitor.Result {
avg := agg.avgResponse()
result := monitor.Result{
AvgResponse: avg,
MinResponse: agg.minUs,
MaxResponse: agg.maxUs,
PacketLoss: agg.lossPercentage(),
TotalCount: agg.totalCount,
SuccessCount: agg.successCount,
ResponseSum: agg.sumUs,
}
if agg.successCount == 0 {
result.MinResponse, result.MaxResponse = 0, 0
}
return result
}
// avgResponse returns the rounded average of successful samples.
func (agg monitorAggregate) avgResponse() int64 {
if agg.successCount == 0 {
return 0
}
return agg.sumUs / agg.successCount
}
// lossPercentage returns the rounded failure rate for the aggregate.
func (agg monitorAggregate) lossPercentage() float64 {
if agg.totalCount == 0 {
return 0
}
return math.Round(float64(agg.totalCount-agg.successCount)/float64(agg.totalCount)*10000) / 100
}
// resultLocked returns the aggregated monitor result for the requested duration along with a bool indicating whether any data was available.
func (h *monitorHistory) resultLocked(duration time.Duration, now time.Time) (monitor.Result, bool) {
agg := h.aggregateLocked(duration, now)
if !agg.hasData() {
// short realtime windows (e.g. the 1s window used for 1m/realtime charts) often fall
// between monitor samples since monitors run at longer, user-defined intervals; fall back to
// the most recent sample so realtime requests still report current status.
agg = h.latestSampleAggregateLocked()
}
hourAgg := h.aggregateLocked(time.Hour, now)
if !agg.hasData() {
return monitor.Result{}, false
}
result := agg.result()
if len(h.samples) > 0 {
result.LastProbeAt = h.samples[len(h.samples)-1].timestamp.UnixMilli()
}
result.AvgResponse1h = hourAgg.avgResponse()
result.MinResponse1h = hourAgg.minUs
result.MaxResponse1h = hourAgg.maxUs
result.PacketLoss1h = hourAgg.lossPercentage()
result.SampleCount = h.sampleCount
if hourAgg.successCount == 0 {
result.MinResponse1h, result.MaxResponse1h = 0, 0
}
return result, true
}
// latestSampleAggregateLocked returns an aggregate containing only the most recent sample, if any.
func (h *monitorHistory) latestSampleAggregateLocked() monitorAggregate {
agg := newMonitorAggregate()
if len(h.samples) == 0 {
return agg
}
agg.addResponse(h.samples[len(h.samples)-1].responseUs)
return agg
}
// aggregateLocked collects monitor data for the requested time window.
func (h *monitorHistory) aggregateLocked(duration time.Duration, now time.Time) monitorAggregate {
cutoff := now.Add(-duration)
// Keep short windows exact; longer windows read from minute buckets to avoid raw-sample retention.
if duration <= monitorRawRetention {
return aggregateSamplesSince(h.samples, cutoff)
}
return aggregateBucketsSince(h.buckets[:], cutoff, now)
}
// aggregateSamplesSince aggregates raw samples newer than the cutoff.
func aggregateSamplesSince(samples []monitorSample, cutoff time.Time) monitorAggregate {
agg := newMonitorAggregate()
for _, sample := range samples {
if sample.timestamp.Before(cutoff) {
continue
}
agg.addResponse(sample.responseUs)
}
return agg
}
// aggregateBucketsSince aggregates minute buckets overlapping the requested window.
func aggregateBucketsSince(buckets []monitorBucket, cutoff, now time.Time) monitorAggregate {
agg := newMonitorAggregate()
startMinute := int32(cutoff.Unix() / 60)
endMinute := int32(now.Unix() / 60)
for _, bucket := range buckets {
if !bucket.filled || bucket.minute < startMinute || bucket.minute > endMinute {
continue
}
agg.addAggregate(bucket.stats)
}
return agg
}
// addSampleLocked stores a fresh sample in both raw and per-minute retention buffers.
func (h *monitorHistory) addSampleLocked(sample monitorSample) {
h.sampleCount++
cutoff := sample.timestamp.Add(-monitorRawRetention)
start := 0
for i := range h.samples {
if !h.samples[i].timestamp.Before(cutoff) {
start = i
break
}
if i == len(h.samples)-1 {
start = len(h.samples)
}
}
if start > 0 {
size := copy(h.samples, h.samples[start:])
h.samples = h.samples[:size]
}
h.samples = append(h.samples, sample)
minute := int32(sample.timestamp.Unix() / 60)
// Each slot stores one wall-clock minute, so the ring stays fixed-size at ~1h per monitor.
bucket := &h.buckets[minute%monitorMinuteBucketLen]
if !bucket.filled || bucket.minute != minute {
bucket.minute = minute
bucket.filled = true
bucket.stats = newMonitorAggregate()
}
bucket.stats.addResponse(sample.responseUs)
}
-154
View File
@@ -1,154 +0,0 @@
package agent
import (
"testing"
"time"
"github.com/fxamacker/cbor/v2"
"github.com/henrygd/beszel/internal/entities/monitor"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
func TestMonitorHistoryWindowCounts(t *testing.T) {
history := newMonitorHistory()
now := time.Now()
// This older success counts toward lifetime warm-up, but not this window.
history.record(monitorSample{responseUs: 1000, timestamp: now.Add(-2 * time.Minute)})
history.record(monitorSample{responseUs: 10, timestamp: now.Add(-30 * time.Second)})
history.record(monitorSample{responseUs: 21, timestamp: now.Add(-20 * time.Second)})
history.record(monitorSample{responseUs: -1, timestamp: now.Add(-10 * time.Second)})
result, ok := history.result(time.Minute, now)
require.True(t, ok)
assert.EqualValues(t, 4, result.SampleCount)
assert.EqualValues(t, 3, result.TotalCount)
assert.EqualValues(t, 2, result.SuccessCount)
assert.EqualValues(t, 31, result.ResponseSum, "preserve the sum before average rounding")
assert.EqualValues(t, 15, result.AvgResponse)
assert.Equal(t, 33.33, result.PacketLoss)
encoded, err := cbor.Marshal(result)
require.NoError(t, err)
var decoded monitor.Result
require.NoError(t, cbor.Unmarshal(encoded, &decoded))
assert.Equal(t, result, decoded)
stats := monitor.Stats{}.FromResult(decoded)
assert.Equal(t, result.TotalCount, stats.TotalCount)
assert.Equal(t, result.SuccessCount, stats.SuccessCount)
assert.Equal(t, result.ResponseSum, stats.ResponseSum)
// Reads do not consume samples. A short window's latest-sample fallback
// carries the count for that single failure, not the minute or lifetime count.
repeated, _ := history.result(time.Minute, now)
assert.Equal(t, result, repeated)
fallback, ok := history.result(time.Second, now)
require.True(t, ok)
assert.EqualValues(t, 1, fallback.TotalCount)
assert.Zero(t, fallback.SuccessCount)
assert.Zero(t, fallback.ResponseSum)
assert.Equal(t, 100.0, fallback.PacketLoss)
assert.EqualValues(t, 4, fallback.SampleCount)
}
func TestMonitorHistoryAggregateLockedUsesRawSamplesForShortWindows(t *testing.T) {
now := time.Date(2026, time.April, 21, 12, 0, 0, 0, time.UTC)
history := newMonitorHistory()
history.addSampleLocked(monitorSample{responseUs: 10, timestamp: now.Add(-90 * time.Second)})
history.addSampleLocked(monitorSample{responseUs: 20, timestamp: now.Add(-30 * time.Second)})
history.addSampleLocked(monitorSample{responseUs: -1, timestamp: now.Add(-10 * time.Second)})
agg := history.aggregateLocked(time.Minute, now)
require.True(t, agg.hasData())
assert.Equal(t, int64(2), agg.totalCount)
assert.Equal(t, int64(1), agg.successCount)
result := agg.result()
assert.Equal(t, int64(20), result.AvgResponse)
assert.Equal(t, int64(20), result.MinResponse)
assert.Equal(t, int64(20), result.MaxResponse)
assert.Equal(t, 50.0, result.PacketLoss)
}
func TestMonitorHistoryAggregateLockedUsesMinuteBucketsForLongWindows(t *testing.T) {
now := time.Date(2026, time.April, 21, 12, 0, 30, 0, time.UTC)
history := newMonitorHistory()
history.addSampleLocked(monitorSample{responseUs: 10, timestamp: now.Add(-11 * time.Minute)})
history.addSampleLocked(monitorSample{responseUs: 20, timestamp: now.Add(-9 * time.Minute)})
history.addSampleLocked(monitorSample{responseUs: 40, timestamp: now.Add(-5 * time.Minute)})
history.addSampleLocked(monitorSample{responseUs: -1, timestamp: now.Add(-90 * time.Second)})
history.addSampleLocked(monitorSample{responseUs: 30, timestamp: now.Add(-30 * time.Second)})
agg := history.aggregateLocked(10*time.Minute, now)
require.True(t, agg.hasData())
assert.Equal(t, int64(4), agg.totalCount)
assert.Equal(t, int64(3), agg.successCount)
result := agg.result()
assert.Equal(t, int64(30), result.AvgResponse)
assert.Equal(t, int64(20), result.MinResponse)
assert.Equal(t, int64(40), result.MaxResponse)
assert.Equal(t, 25.0, result.PacketLoss)
}
func TestMonitorHistoryAddSampleLockedTrimsRawSamplesButKeepsBucketHistory(t *testing.T) {
now := time.Date(2026, time.April, 21, 12, 0, 0, 0, time.UTC)
history := newMonitorHistory()
history.addSampleLocked(monitorSample{responseUs: 10, timestamp: now.Add(-10 * time.Minute)})
history.addSampleLocked(monitorSample{responseUs: 20, timestamp: now})
require.Len(t, history.samples, 1)
assert.Equal(t, int64(20), history.samples[0].responseUs)
agg := history.aggregateLocked(10*time.Minute, now)
require.True(t, agg.hasData())
assert.Equal(t, int64(2), agg.totalCount)
assert.Equal(t, int64(2), agg.successCount)
result := agg.result()
assert.Equal(t, int64(15), result.AvgResponse)
assert.Equal(t, int64(10), result.MinResponse)
assert.Equal(t, int64(20), result.MaxResponse)
assert.Equal(t, 0.0, result.PacketLoss)
}
func TestMonitorHistoryProbeTimestamp(t *testing.T) {
history := newMonitorHistory()
start := time.Date(2026, time.September, 14, 12, 0, 0, 0, time.UTC)
_, ok := history.result(time.Minute, start)
require.False(t, ok)
first := history.record(monitorSample{responseUs: 20, timestamp: start})
assert.Equal(t, start.UnixMilli(), first.LastProbeAt)
for minute := 0; minute < 5; minute++ {
now := start.Add(time.Duration(minute)*time.Minute + time.Second)
// Realtime reads must not consume freshness for the persistence request.
for _, window := range []time.Duration{time.Second, time.Minute} {
result, ok := history.result(window, now)
require.True(t, ok)
assert.Equal(t, first.LastProbeAt, result.LastProbeAt)
assert.Equal(t, int64(20), result.AvgResponse)
}
}
next := start.Add(5 * time.Minute)
failed := history.record(monitorSample{responseUs: -1, timestamp: next})
assert.Equal(t, next.UnixMilli(), failed.LastProbeAt)
assert.Equal(t, float64(100), failed.PacketLoss)
repeated, ok := history.result(time.Minute, next.Add(2*time.Minute))
require.True(t, ok)
assert.Equal(t, failed.LastProbeAt, repeated.LastProbeAt)
assert.Equal(t, float64(100), repeated.PacketLoss)
}
func TestMonitorHistorySampleCount(t *testing.T) {
history := newMonitorHistory()
now := time.Now()
// Both failed and successful probes count, including older samples so
// monitors with hourly intervals can finish warming up.
history.record(monitorSample{responseUs: -1, timestamp: now.Add(-2 * time.Hour)})
for i, response := range []int64{10, -1, 20} {
result := history.record(monitorSample{responseUs: response, timestamp: now.Add(time.Duration(i) * time.Second)})
assert.EqualValues(t, i+2, result.SampleCount)
}
result, ok := history.clone().result(time.Minute, now.Add(3*time.Second))
require.True(t, ok)
assert.EqualValues(t, 4, result.SampleCount)
}
-312
View File
@@ -1,312 +0,0 @@
package agent
import (
"bytes"
"context"
"crypto/rand"
"errors"
"fmt"
"math"
"net"
"os"
"os/exec"
"regexp"
"runtime"
"strconv"
"strings"
"sync"
"sync/atomic"
"time"
"golang.org/x/net/icmp"
"golang.org/x/net/ipv4"
"golang.org/x/net/ipv6"
"log/slog"
)
// Match the numeric RTT independently of the localized label used by Windows.
var pingTimeRegex = regexp.MustCompile(`(?i)[=<]\s*([0-9]+(?:[.,][0-9]+)?)\s*ms\b`)
var icmpSequence atomic.Uint32
type icmpPacketConn interface {
Close() error
}
// icmpMethod tracks which ICMP approach to use. Once a method succeeds or
// all native methods fail, the choice is cached so subsequent monitors skip
// the trial-and-error overhead.
type icmpMethod uint8
const (
icmpUntried icmpMethod = iota // haven't tried yet
icmpRaw // privileged raw socket
icmpDatagram // unprivileged datagram socket
icmpExecFallback // shell out to system ping command
)
// icmpFamily holds the network parameters and cached detection result for one address family.
type icmpFamily struct {
rawNetwork string // e.g. "ip4:icmp" or "ip6:ipv6-icmp"
dgramNetwork string // e.g. "udp4" or "udp6"
listenAddr string // "0.0.0.0" or "::"
echoType icmp.Type // outgoing echo request type
replyType icmp.Type // expected echo reply type
proto int // IANA protocol number for parsing replies
isIPv6 bool
mode icmpMethod // cached detection result (guarded by icmpModeMu)
}
var (
icmpV4 = icmpFamily{
rawNetwork: "ip4:icmp",
dgramNetwork: "udp4",
listenAddr: "0.0.0.0",
echoType: ipv4.ICMPTypeEcho,
replyType: ipv4.ICMPTypeEchoReply,
proto: 1,
}
icmpV6 = icmpFamily{
rawNetwork: "ip6:ipv6-icmp",
dgramNetwork: "udp6",
listenAddr: "::",
echoType: ipv6.ICMPTypeEchoRequest,
replyType: ipv6.ICMPTypeEchoReply,
proto: 58,
isIPv6: true,
}
icmpModeMu sync.Mutex
icmpListen = func(network, listenAddr string) (icmpPacketConn, error) {
return icmp.ListenPacket(network, listenAddr)
}
)
// monitorICMP sends an ICMP echo request and measures round-trip response.
// Supports both IPv4 and IPv6 targets. The ICMP method (raw socket,
// unprivileged datagram, or exec fallback) is detected once per address
// family and cached for subsequent monitors.
// Returns response in microseconds, or -1 and an error on failure.
func monitorICMP(ctx context.Context, target string) (int64, error) {
ctx, cancel := context.WithTimeout(ctx, 3*time.Second)
defer cancel()
family, ip, err := resolveICMPTarget(ctx, target)
if err != nil {
return -1, err
}
icmpModeMu.Lock()
if family.mode == icmpUntried {
family.mode = detectICMPMode(family, icmpListen)
}
mode := family.mode
icmpModeMu.Unlock()
switch mode {
case icmpRaw:
return monitorICMPNative(ctx, family.rawNetwork, family, &net.IPAddr{IP: ip})
case icmpDatagram:
return monitorICMPNative(ctx, family.dgramNetwork, family, &net.UDPAddr{IP: ip})
case icmpExecFallback:
return monitorICMPExec(ctx, ip.String(), family.isIPv6)
default:
return -1, errors.New("unsupported ICMP mode")
}
}
// resolveICMPTarget resolves a target hostname or IP to determine the address
// family and concrete IP address. Prefers IPv4 for dual-stack hostnames.
func resolveICMPTarget(ctx context.Context, target string) (*icmpFamily, net.IP, error) {
if ip := net.ParseIP(target); ip != nil {
if ip.To4() != nil {
return &icmpV4, ip.To4(), nil
}
return &icmpV6, ip, nil
}
ips, err := net.DefaultResolver.LookupIP(ctx, "ip", target)
if err != nil || len(ips) == 0 {
return nil, nil, err
}
for _, ip := range ips {
if v4 := ip.To4(); v4 != nil {
return &icmpV4, v4, nil
}
}
return &icmpV6, ips[0], nil
}
func detectICMPMode(family *icmpFamily, listen func(network, listenAddr string) (icmpPacketConn, error)) icmpMethod {
label := "IPv4"
if family.isIPv6 {
label = "IPv6"
}
conn, err := listen(family.rawNetwork, family.listenAddr)
slog.Debug("ICMP raw socket test", "family", label, "err", err)
if err == nil {
conn.Close()
return icmpRaw
}
conn, err = listen(family.dgramNetwork, family.listenAddr)
slog.Debug("ICMP datagram socket test", "family", label, "err", err)
if err == nil {
conn.Close()
return icmpDatagram
}
return icmpExecFallback
}
// monitorICMPNative sends an ICMP echo request using Go's x/net/icmp package.
func monitorICMPNative(ctx context.Context, network string, family *icmpFamily, dst net.Addr) (int64, error) {
conn, err := icmp.ListenPacket(network, family.listenAddr)
if err != nil {
return -1, err
}
defer conn.Close()
return monitorICMPPacket(ctx, conn, family, dst)
}
func monitorICMPPacket(ctx context.Context, conn net.PacketConn, family *icmpFamily, dst net.Addr) (int64, error) {
if err := ctx.Err(); err != nil {
return -1, err
}
// Closing the socket interrupts both reads and writes on cancellation.
stop := context.AfterFunc(ctx, func() { _ = conn.Close() })
defer stop()
// Prepare correlation data before starting the round-trip timer. The token
// also distinguishes delayed replies after the 16-bit sequence wraps.
token := make([]byte, 16)
if _, err := rand.Read(token); err != nil {
return -1, err
}
echo := &icmp.Echo{
ID: os.Getpid() & 0xffff,
Seq: int(icmpSequence.Add(1) & 0xffff),
Data: token,
}
// Linux ping sockets replace the Echo ID with their bound port. Darwin
// datagram sockets and raw sockets preserve the supplied ID.
if local, ok := conn.LocalAddr().(*net.UDPAddr); ok && runtime.GOOS == "linux" {
echo.ID = local.Port
}
targetIP := icmpAddrIP(dst)
msg := &icmp.Message{
Type: family.echoType,
Code: 0,
Body: echo,
}
msgBytes, err := msg.Marshal(nil)
if err != nil {
return -1, err
}
// Set deadline before sending
if err := conn.SetDeadline(time.Now().Add(3 * time.Second)); err != nil {
return -1, err
}
buf := make([]byte, 1500)
start := time.Now()
if _, err := conn.WriteTo(msgBytes, dst); err != nil {
return -1, err
}
// Read reply
for {
n, peer, err := conn.ReadFrom(buf)
received := time.Now()
if err != nil {
return -1, err
}
if !targetIP.Equal(icmpAddrIP(peer)) {
continue
}
reply, err := icmp.ParseMessage(family.proto, buf[:n])
if err != nil || reply.Type != family.replyType || reply.Code != 0 {
continue
}
body, ok := reply.Body.(*icmp.Echo)
if ok && body.ID == echo.ID && body.Seq == echo.Seq && bytes.Equal(body.Data, echo.Data) {
return received.Sub(start).Microseconds(), nil
}
// Keep waiting for our reply without extending the original deadline.
}
}
func icmpAddrIP(addr net.Addr) net.IP {
switch addr := addr.(type) {
case *net.IPAddr:
return addr.IP
case *net.UDPAddr:
return addr.IP
default:
return nil
}
}
// pingCommand selects the executable and arguments for the supported agent platforms.
// The context deadline enforces the timeout: -W has incompatible meanings across
// Linux, BSD IPv4 ping, and macOS ping6.
func pingCommand(goos, target string, isIPv6 bool) (string, []string, error) {
family := "-4"
if isIPv6 {
family = "-6"
}
switch goos {
case "windows":
return "ping", []string{family, "-n", "1", "-w", "3000", target}, nil
case "linux":
return "ping", []string{family, "-n", "-c", "1", target}, nil
case "darwin", "freebsd", "openbsd":
command := "ping"
if isIPv6 {
command = "ping6"
}
return command, []string{"-n", "-c", "1", target}, nil
default:
return "", nil, fmt.Errorf("ping fallback is unsupported on %s", goos)
}
}
// monitorICMPExec falls back to the system ping command. Returns -1 and an error on failure.
func monitorICMPExec(ctx context.Context, target string, isIPv6 bool) (int64, error) {
ctx, cancel := context.WithTimeout(ctx, 3*time.Second)
defer cancel()
name, args, err := pingCommand(runtime.GOOS, target, isIPv6)
if err != nil {
return -1, err
}
cmd := exec.CommandContext(ctx, name, args...)
// Keep Unix output and decimal formatting stable. Windows ignores LC_ALL.
cmd.Env = append(os.Environ(), "LC_ALL=C")
output, err := cmd.Output()
if ctx.Err() != nil {
return -1, ctx.Err()
}
if err != nil {
return -1, fmt.Errorf("%s failed: %w", name, err)
}
return parsePingResponse(output)
}
// parsePingResponse returns the reported RTT, never subprocess execution time.
// For a bounded value such as Windows' time<1ms, retain the reported upper bound.
func parsePingResponse(output []byte) (int64, error) {
matches := pingTimeRegex.FindSubmatch(output)
if len(matches) < 2 {
return -1, errors.New("ping output contains no round-trip time")
}
ms, err := strconv.ParseFloat(strings.ReplaceAll(string(matches[1]), ",", "."), 64)
if err != nil || math.IsInf(ms, 0) || ms >= float64(math.MaxInt64)/1000 {
return -1, errors.New("invalid round-trip time in ping output")
}
return int64(math.Round(ms * 1000)), nil
}
-433
View File
@@ -1,433 +0,0 @@
//go:build testing
package agent
import (
"context"
"errors"
"fmt"
"net"
"os"
"path/filepath"
"runtime"
"testing"
"time"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
"golang.org/x/net/icmp"
)
type testICMPPacketConn struct{}
func (testICMPPacketConn) Close() error { return nil }
type blockingICMPConn struct {
net.PacketConn
reading chan struct{}
}
func (c *blockingICMPConn) WriteTo(p []byte, addr net.Addr) (int, error) {
return len(p), nil
}
func (c *blockingICMPConn) ReadFrom(p []byte) (int, net.Addr, error) {
close(c.reading)
return c.PacketConn.ReadFrom(p)
}
func TestMonitorICMPPacketCancellation(t *testing.T) {
conn, err := net.ListenPacket("udp4", "127.0.0.1:0")
require.NoError(t, err)
defer conn.Close()
blocking := &blockingICMPConn{PacketConn: conn, reading: make(chan struct{})}
ctx, cancel := context.WithCancel(context.Background())
defer cancel()
done := make(chan error, 1)
go func() {
_, err := monitorICMPPacket(ctx, blocking, &icmpV4, conn.LocalAddr())
done <- err
}()
select {
case <-blocking.reading:
case <-time.After(time.Second):
t.Fatal("probe did not begin reading")
}
cancel()
select {
case err := <-done:
require.Error(t, err)
case <-time.After(time.Second):
t.Fatal("cancellation did not interrupt the socket read")
}
}
func TestMonitorICMPExecCancellation(t *testing.T) {
if runtime.GOOS == "windows" {
t.Skip("test uses a POSIX shell stub for ping")
}
dir := t.TempDir()
require.NoError(t, os.WriteFile(filepath.Join(dir, "ping"), []byte("#!/bin/sh\nexec sleep 30\n"), 0o755))
t.Setenv("PATH", dir+string(os.PathListSeparator)+os.Getenv("PATH"))
ctx, cancel := context.WithTimeout(context.Background(), 100*time.Millisecond)
defer cancel()
done := make(chan error, 1)
go func() {
_, err := monitorICMPExec(ctx, "127.0.0.1", false)
done <- err
}()
select {
case err := <-done:
require.ErrorIs(t, err, context.DeadlineExceeded)
case <-time.After(time.Second):
t.Fatal("cancellation did not terminate ping")
}
}
func TestPingCommand(t *testing.T) {
for _, goos := range []string{"linux", "windows", "darwin", "freebsd", "openbsd"} {
for _, ipv6 := range []bool{false, true} {
t.Run(fmt.Sprintf("%s/ipv6=%t", goos, ipv6), func(t *testing.T) {
target, family := "192.0.2.1", "-4"
if ipv6 {
target, family = "2001:db8::1", "-6"
}
name, args, err := pingCommand(goos, target, ipv6)
require.NoError(t, err)
wantName := "ping"
wantArgs := []string{"-n", "-c", "1", target}
switch goos {
case "windows":
wantArgs = []string{family, "-n", "1", "-w", "3000", target}
case "linux":
wantArgs = append([]string{family}, wantArgs...)
default:
if ipv6 {
wantName = "ping6"
}
}
assert.Equal(t, wantName, name)
assert.Equal(t, wantArgs, args)
})
}
}
_, _, err := pingCommand("unsupported", "192.0.2.1", false)
require.Error(t, err)
}
func TestParsePingResponse(t *testing.T) {
for _, tc := range []struct {
name string
output string
wantUs int64
}{
{"linux", "64 bytes from 192.0.2.1: icmp_seq=1 ttl=64 time=12.345 ms", 12345},
{"bsd", "64 bytes from 192.0.2.1: icmp_seq=0 ttl=64 time=0.023 ms", 23},
{"ipv6", "64 bytes from 2001:db8::1: icmp_seq=0 hlim=64 time=1.234 ms", 1234},
{"windows", "Reply from 192.0.2.1: bytes=32 time=12ms TTL=128", 12000},
{"windows submillisecond", "Reply from ::1: time<1ms", 1000},
{"localized windows", "Antwort von 192.0.2.1: Bytes=32 Zeit=12ms TTL=128", 12000},
{"decimal comma", "64 bytes from 192.0.2.1: time=1,234 ms", 1234},
{"rounding", "time=0.1236 ms", 124},
{"empty", "", -1},
{"timeout", "Request timed out.", -1},
{"unreachable", "Reply from 192.0.2.1: Destination host unreachable.", -1},
{"malformed", "time=oops ms", -1},
{"negative", "time=-1 ms", -1},
{"overflow", "time=999999999999999999999 ms", -1},
} {
t.Run(tc.name, func(t *testing.T) {
responseUs, err := parsePingResponse([]byte(tc.output))
if tc.wantUs < 0 {
require.Error(t, err)
} else {
require.NoError(t, err)
}
assert.Equal(t, tc.wantUs, responseUs)
})
}
}
func TestMonitorICMPExecOutput(t *testing.T) {
if runtime.GOOS == "windows" {
t.Skip("test uses a POSIX shell stub for ping")
}
for _, tc := range []struct {
name string
output string
exit int
wantUs int64
}{
{"success", "time=1.234 ms", 0, 1234},
{"missing RTT", "unrecognized output", 0, -1},
{"failed command with RTT", "time=1.234 ms", 1, -1},
} {
t.Run(tc.name, func(t *testing.T) {
dir := t.TempDir()
// Also verify an inherited locale cannot override the C locale.
script := fmt.Sprintf("#!/bin/sh\n[ \"$LC_ALL\" = C ] || exit 2\nprintf '%%s\\n' '%s'\nexit %d\n", tc.output, tc.exit)
require.NoError(t, os.WriteFile(filepath.Join(dir, "ping"), []byte(script), 0o755))
t.Setenv("PATH", dir+string(os.PathListSeparator)+os.Getenv("PATH"))
t.Setenv("LC_ALL", "de_DE.UTF-8")
responseUs, err := monitorICMPExec(t.Context(), "127.0.0.1", false)
if tc.wantUs < 0 {
require.Error(t, err)
} else {
require.NoError(t, err)
}
assert.Equal(t, tc.wantUs, responseUs)
})
}
}
type icmpTestReply struct {
data []byte
peer net.Addr
}
type scriptedICMPConn struct {
net.PacketConn
local net.Addr
onWrite func([]byte, net.Addr)
replies []icmpTestReply
reads int
deadlineSets int
}
func (c *scriptedICMPConn) LocalAddr() net.Addr { return c.local }
func (c *scriptedICMPConn) SetDeadline(deadline time.Time) error {
c.deadlineSets++
return nil
}
func (c *scriptedICMPConn) WriteTo(data []byte, dst net.Addr) (int, error) {
c.onWrite(data, dst)
return len(data), nil
}
func (c *scriptedICMPConn) ReadFrom(buf []byte) (int, net.Addr, error) {
c.reads++
if len(c.replies) == 0 {
return 0, nil, os.ErrDeadlineExceeded
}
reply := c.replies[0]
c.replies = c.replies[1:]
return copy(buf, reply.data), reply.peer, nil
}
func TestMonitorICMPReplyCorrelation(t *testing.T) {
for _, family := range []*icmpFamily{&icmpV4, &icmpV6} {
for _, datagram := range []bool{false, true} {
network := family.rawNetwork
ip, other := net.ParseIP("192.0.2.1"), net.ParseIP("192.0.2.2")
if family.isIPv6 {
ip, other = net.ParseIP("2001:db8::1"), net.ParseIP("2001:db8::2")
}
var dst net.Addr = &net.IPAddr{IP: ip}
var wrongPeer net.Addr = &net.IPAddr{IP: other}
if datagram {
network = family.dgramNetwork
dst = &net.UDPAddr{IP: ip}
wrongPeer = &net.UDPAddr{IP: other}
}
for _, mismatch := range []string{"source", "id", "sequence", "payload", "type", "code", "malformed"} {
for _, eventuallyMatches := range []bool{false, true} {
ending := "timeout"
if eventuallyMatches {
ending = "success"
}
t.Run(network+"/"+mismatch+"/"+ending, func(t *testing.T) {
conn := &scriptedICMPConn{local: &net.IPAddr{IP: net.IPv4zero}}
if datagram {
conn.local = &net.UDPAddr{Port: 12345}
if runtime.GOOS == "linux" {
// Deliberately differ from the process ID.
conn.local = &net.UDPAddr{Port: (os.Getpid() % 65534) + 1}
}
}
conn.onWrite = func(data []byte, target net.Addr) {
require.Equal(t, dst, target)
request, err := icmp.ParseMessage(family.proto, data)
require.NoError(t, err)
echo := request.Body.(*icmp.Echo)
expectedID := os.Getpid() & 0xffff
if datagram && runtime.GOOS == "linux" {
expectedID = conn.local.(*net.UDPAddr).Port
}
require.Equal(t, expectedID, echo.ID)
reply := &icmp.Message{Type: family.replyType, Body: echo}
valid, err := reply.Marshal(nil)
require.NoError(t, err)
peer := dst
switch mismatch {
case "source":
peer = wrongPeer
case "id":
echo.ID ^= 1
case "sequence":
echo.Seq ^= 1
case "payload":
echo.Data[0] ^= 1
case "type":
reply.Type = family.echoType
case "code":
reply.Code = 1
}
invalid, err := reply.Marshal(nil)
require.NoError(t, err)
if mismatch == "malformed" {
invalid = invalid[:2]
}
conn.replies = []icmpTestReply{{invalid, peer}}
if eventuallyMatches {
conn.replies = append(conn.replies, icmpTestReply{valid, dst})
}
}
elapsed, err := monitorICMPPacket(context.Background(), conn, family, dst)
if eventuallyMatches {
require.NoError(t, err)
assert.GreaterOrEqual(t, elapsed, int64(0))
} else {
require.ErrorIs(t, err, os.ErrDeadlineExceeded)
assert.Equal(t, int64(-1), elapsed)
}
assert.Equal(t, 2, conn.reads)
assert.Equal(t, 1, conn.deadlineSets)
})
}
}
}
}
}
func TestMonitorICMPLoopback(t *testing.T) {
for _, family := range []*icmpFamily{&icmpV4, &icmpV6} {
for _, network := range []string{family.rawNetwork, family.dgramNetwork} {
t.Run(network, func(t *testing.T) {
conn, err := icmp.ListenPacket(network, family.listenAddr)
if err != nil {
t.Skipf("ICMP socket unavailable: %v", err)
}
defer conn.Close()
ip := net.ParseIP("127.0.0.1")
if family.isIPv6 {
ip = net.ParseIP("::1")
}
var dst net.Addr = &net.IPAddr{IP: ip}
if network == family.dgramNetwork {
dst = &net.UDPAddr{IP: ip}
}
elapsed, err := monitorICMPPacket(context.Background(), conn, family, dst)
require.NoError(t, err)
assert.GreaterOrEqual(t, elapsed, int64(0))
})
}
}
}
func TestDetectICMPMode(t *testing.T) {
tests := []struct {
name string
family *icmpFamily
rawErr error
udpErr error
want icmpMethod
wantNetworks []string
}{
{
name: "IPv4 prefers raw socket when available",
family: &icmpV4,
want: icmpRaw,
wantNetworks: []string{"ip4:icmp"},
},
{
name: "IPv4 uses datagram when raw unavailable",
family: &icmpV4,
rawErr: errors.New("operation not permitted"),
want: icmpDatagram,
wantNetworks: []string{"ip4:icmp", "udp4"},
},
{
name: "IPv4 falls back to exec when both unavailable",
family: &icmpV4,
rawErr: errors.New("operation not permitted"),
udpErr: errors.New("protocol not supported"),
want: icmpExecFallback,
wantNetworks: []string{"ip4:icmp", "udp4"},
},
{
name: "IPv6 prefers raw socket when available",
family: &icmpV6,
want: icmpRaw,
wantNetworks: []string{"ip6:ipv6-icmp"},
},
{
name: "IPv6 uses datagram when raw unavailable",
family: &icmpV6,
rawErr: errors.New("operation not permitted"),
want: icmpDatagram,
wantNetworks: []string{"ip6:ipv6-icmp", "udp6"},
},
{
name: "IPv6 falls back to exec when both unavailable",
family: &icmpV6,
rawErr: errors.New("operation not permitted"),
udpErr: errors.New("protocol not supported"),
want: icmpExecFallback,
wantNetworks: []string{"ip6:ipv6-icmp", "udp6"},
},
}
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
calls := make([]string, 0, 2)
listen := func(network, listenAddr string) (icmpPacketConn, error) {
require.Equal(t, tt.family.listenAddr, listenAddr)
calls = append(calls, network)
switch network {
case tt.family.rawNetwork:
if tt.rawErr != nil {
return nil, tt.rawErr
}
case tt.family.dgramNetwork:
if tt.udpErr != nil {
return nil, tt.udpErr
}
default:
t.Fatalf("unexpected network %q", network)
}
return testICMPPacketConn{}, nil
}
assert.Equal(t, tt.want, detectICMPMode(tt.family, listen))
assert.Equal(t, tt.wantNetworks, calls)
})
}
}
func TestResolveICMPTarget(t *testing.T) {
t.Run("IPv4 literal", func(t *testing.T) {
family, ip, err := resolveICMPTarget(context.Background(), "127.0.0.1")
require.NoError(t, err)
require.NotNil(t, family)
assert.False(t, family.isIPv6)
assert.Equal(t, "127.0.0.1", ip.String())
})
t.Run("IPv6 literal", func(t *testing.T) {
family, ip, err := resolveICMPTarget(context.Background(), "::1")
require.NoError(t, err)
require.NotNil(t, family)
assert.True(t, family.isIPv6)
assert.Equal(t, "::1", ip.String())
})
t.Run("IPv4-mapped IPv6 resolves as IPv4", func(t *testing.T) {
family, ip, err := resolveICMPTarget(context.Background(), "::ffff:127.0.0.1")
require.NoError(t, err)
require.NotNil(t, family)
assert.False(t, family.isIPv6)
assert.Equal(t, "127.0.0.1", ip.String())
})
}
-105
View File
@@ -1,105 +0,0 @@
package agent
import (
"context"
"errors"
"fmt"
"net"
"net/http"
"time"
"github.com/henrygd/beszel/internal/entities/monitor"
)
// monitorProbe performs one check. Errors are recorded as loss by the task runner.
// Implementations must honor cancellation and bound their execution time.
type monitorProbe func(context.Context, monitor.Config) (int64, error)
func networkMonitorProbe(client *http.Client) monitorProbe {
return func(ctx context.Context, config monitor.Config) (int64, error) {
switch config.Protocol {
case "icmp":
return monitorICMP(ctx, config.Target)
case "tcp":
return monitorTCP(ctx, config.Target, config.Port)
case "http":
return monitorHTTP(ctx, client, config.Target)
case "dns":
return monitorDNS(ctx, config.Target)
default:
return -1, fmt.Errorf("unknown monitor protocol: %s", config.Protocol)
}
}
}
// monitorTCP measures connection establishment time, including address fallback
// but excluding DNS resolution.
// Returns -1 and an error on failure.
func monitorTCP(ctx context.Context, target string, port uint16) (int64, error) {
ctx, cancel := context.WithTimeout(ctx, 3*time.Second)
defer cancel()
// Resolve DNS first, outside the timing window but within the probe deadline.
ips, err := net.DefaultResolver.LookupHost(ctx, target)
if err != nil {
return -1, err
}
if len(ips) == 0 {
return -1, errors.New("no addresses resolved for TCP monitor")
}
portString := fmt.Sprintf("%d", port)
deadline, _ := ctx.Deadline()
// Share the remaining probe budget across addresses so an unresponsive
// first address cannot consume all the time available for alternatives.
start := time.Now()
for i, ip := range ips {
if err := ctx.Err(); err != nil {
return -1, err
}
dialer := net.Dialer{Timeout: time.Until(deadline) / time.Duration(len(ips)-i)}
var conn net.Conn
conn, err = dialer.DialContext(ctx, "tcp", net.JoinHostPort(ip, portString))
if err != nil {
continue
}
responseUs := time.Since(start).Microseconds()
conn.Close()
return responseUs, nil
}
return -1, err
}
// monitorDNS measures DNS resolution response time in microseconds. Returns -1 and an error on failure.
func monitorDNS(ctx context.Context, target string) (int64, error) {
ctx, cancel := context.WithTimeout(ctx, 3*time.Second)
defer cancel()
start := time.Now()
ips, err := net.DefaultResolver.LookupHost(ctx, target)
if err != nil || len(ips) == 0 {
return -1, err
}
return time.Since(start).Microseconds(), nil
}
// monitorHTTP measures HTTP GET request response in microseconds. Returns -1 and an error on failure.
func monitorHTTP(ctx context.Context, client *http.Client, url string) (int64, error) {
if client == nil {
client = http.DefaultClient
}
start := time.Now()
req, err := http.NewRequestWithContext(ctx, http.MethodGet, url, nil)
if err != nil {
return -1, err
}
resp, err := client.Do(req)
if err != nil {
return -1, err
}
resp.Body.Close()
if resp.StatusCode >= 400 {
return -1, fmt.Errorf("HTTP error: %s", resp.Status)
}
return time.Since(start).Microseconds(), nil
}
-88
View File
@@ -1,88 +0,0 @@
package agent
import (
"sync"
"time"
)
const (
monitorResumeHeartbeat = 10 * time.Second
// Allow scheduling jitter without mistaking an ordinary tick for resume.
monitorResumeGap = 2 * monitorResumeHeartbeat
monitorResumePause = 10 * time.Second
)
// monitorResumeGuard detects likely suspend/resume using wall time. A long
// process stall or forward clock adjustment can also trigger the bounded pause.
// One heartbeat is shared by all configured monitors.
type monitorResumeGuard struct {
mu sync.Mutex
stop chan struct{}
lastTick time.Time
pauseUntil time.Time
generation uint32
}
func (g *monitorResumeGuard) start() {
g.mu.Lock()
defer g.mu.Unlock()
if g.stop != nil {
return
}
stop := make(chan struct{})
g.stop = stop
g.lastTick = time.Now().Round(0)
g.pauseUntil = time.Time{}
go func() {
ticker := time.NewTicker(monitorResumeHeartbeat)
defer ticker.Stop()
for {
select {
case <-stop:
return
case <-ticker.C:
g.mu.Lock()
if g.stop == stop {
g.observe(time.Now())
}
g.mu.Unlock()
}
}
}()
}
func (g *monitorResumeGuard) shutdown() {
g.mu.Lock()
defer g.mu.Unlock()
if g.stop != nil {
close(g.stop)
g.stop = nil
g.generation++
}
}
// observe requires mu. Strip the monotonic component because it can stop during
// suspend. Read the current time rather than the ticker's queued timestamp.
func (g *monitorResumeGuard) observe(now time.Time) {
now = now.Round(0)
if now.Sub(g.lastTick) > monitorResumeGap {
g.pauseUntil = now.Add(monitorResumePause)
g.generation++
}
g.lastTick = now
}
// snapshot also observes time so a probe waking before the heartbeat detects
// resume itself. A changed generation invalidates probes spanning suspend.
func (g *monitorResumeGuard) snapshot() (generation uint32, allowed bool) {
if g == nil {
return 0, true
}
g.mu.Lock()
defer g.mu.Unlock()
if g.stop == nil {
return g.generation, true
}
g.observe(time.Now())
return g.generation, !g.lastTick.Before(g.pauseUntil)
}
-121
View File
@@ -1,121 +0,0 @@
//go:build testing
package agent
import (
"context"
"errors"
"sync/atomic"
"testing"
"testing/synctest"
"time"
"github.com/henrygd/beszel/internal/entities/monitor"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
func simulateMonitorSleep(g *monitorResumeGuard) {
g.mu.Lock()
g.lastTick = time.Now().Add(-time.Hour).Round(0)
g.mu.Unlock()
}
func TestMonitorResumePause(t *testing.T) {
synctest.Test(t, func(t *testing.T) {
var g monitorResumeGuard
g.start()
defer g.shutdown()
generation, allowed := g.snapshot()
require.True(t, allowed)
// Heartbeats alone must keep the guard current between infrequent probes.
time.Sleep(time.Minute)
synctest.Wait()
steadyGeneration, allowed := g.snapshot()
require.True(t, allowed)
require.Equal(t, generation, steadyGeneration)
// The probe, rather than the heartbeat, must detect this gap.
simulateMonitorSleep(&g)
next, allowed := g.snapshot()
assert.False(t, allowed)
assert.NotEqual(t, generation, next)
time.Sleep(9 * time.Second)
_, allowed = g.snapshot()
assert.False(t, allowed)
time.Sleep(time.Second)
end, allowed := g.snapshot()
assert.True(t, allowed)
assert.Equal(t, next, end)
})
}
func TestMonitorResumeGuardLifecycle(t *testing.T) {
synctest.Test(t, func(t *testing.T) {
pm := newMonitorManagerWithProbe(func(context.Context, monitor.Config) (int64, error) { return 1, nil })
defer pm.Stop()
assert.Nil(t, pm.resumeGuard.stop)
pm.SyncMonitors([]monitor.Config{{ID: "a", Interval: 3600}, {ID: "b", Interval: 3600}})
stop := pm.resumeGuard.stop
require.NotNil(t, stop)
pm.DeleteMonitor("a")
assert.Equal(t, stop, pm.resumeGuard.stop)
pm.DeleteMonitor("b")
assert.Nil(t, pm.resumeGuard.stop)
select {
case <-stop:
default:
t.Fatal("heartbeat was not stopped")
}
time.Sleep(time.Hour)
_, err := pm.UpsertMonitor(monitor.Config{ID: "c", Interval: 3600}, false)
require.NoError(t, err)
_, allowed := pm.resumeGuard.snapshot()
assert.True(t, allowed, "idle time must not trigger a resume pause")
pm.SyncMonitors(nil)
assert.Nil(t, pm.resumeGuard.stop)
})
}
func TestMonitorResumeDiscardsInflightProbe(t *testing.T) {
synctest.Test(t, func(t *testing.T) {
var g monitorResumeGuard
g.start()
defer g.shutdown()
task := newMonitorTask(monitor.Config{ID: "test"})
defer task.cancel()
task.resumeGuard = &g
result := task.runProbe(func(context.Context, monitor.Config) (int64, error) {
simulateMonitorSleep(&g)
return 0, errors.New("network not ready")
})
assert.Nil(t, result)
assert.Empty(t, task.history.samples)
// Explicit requests may still run during the pause and record real failures.
result = task.runProbe(func(context.Context, monitor.Config) (int64, error) {
return 0, errors.New("unreachable")
})
require.NotNil(t, result)
assert.Equal(t, 100.0, result.PacketLoss)
})
}
func TestMonitorResumeSkipsScheduledProbes(t *testing.T) {
synctest.Test(t, func(t *testing.T) {
var calls atomic.Int32
pm := newMonitorManagerWithProbe(func(context.Context, monitor.Config) (int64, error) {
calls.Add(1)
return 1, nil
})
defer pm.Stop()
pm.SyncMonitors([]monitor.Config{{ID: "test", Interval: 1}})
simulateMonitorSleep(&pm.resumeGuard)
pm.resumeGuard.snapshot()
time.Sleep(9 * time.Second)
synctest.Wait()
assert.Zero(t, calls.Load())
assert.Empty(t, pm.GetResults(1000))
time.Sleep(2 * time.Second)
synctest.Wait()
assert.Positive(t, calls.Load())
})
}
-60
View File
@@ -1,60 +0,0 @@
package agent
import (
"context"
"log/slog"
"math/rand"
"time"
)
func (pm *MonitorManager) startMonitor(task *monitorTask) {
interval := time.Duration(task.config.Interval) * time.Second
if interval < time.Second {
interval = 30 * time.Second
}
delay := getStagger(interval.Milliseconds())
slog.Debug("starting monitor task", "target", task.config.Target, "delay", delay, "interval", interval)
go runMonitorSchedule(task.ctx, interval, delay, func() {
if _, allowed := task.resumeGuard.snapshot(); allowed {
task.runProbe(pm.probe)
}
})
}
// runMonitorSchedule owns only timing. Checks run serially, and slow checks
// naturally drop missed ticks rather than building an execution backlog.
func runMonitorSchedule(ctx context.Context, interval, delay time.Duration, run func()) {
timer := time.NewTimer(delay)
defer timer.Stop()
select {
case <-ctx.Done():
return
case <-timer.C:
}
if ctx.Err() != nil {
return
}
run()
ticker := time.NewTicker(interval)
defer ticker.Stop()
for {
select {
case <-ctx.Done():
return
case <-ticker.C:
if ctx.Err() != nil {
return
}
run()
}
}
}
// getStagger returns an initial delay between half an interval and one interval.
func getStagger(intervalMilli int64) time.Duration {
delay := rand.Intn(int(intervalMilli))
if delay < int(intervalMilli)/2 {
delay += int(intervalMilli) / 2
}
return time.Duration(delay) * time.Millisecond
}
-167
View File
@@ -1,167 +0,0 @@
//go:build testing
package agent
import (
"context"
"sync/atomic"
"testing"
"testing/synctest"
"time"
"github.com/henrygd/beszel/internal/entities/monitor"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
func TestMonitorScheduleTiming(t *testing.T) {
synctest.Test(t, func(t *testing.T) {
ctx, cancel := context.WithCancel(t.Context())
defer cancel()
var calls atomic.Int32
go runMonitorSchedule(ctx, 10*time.Second, 5*time.Second, func() { calls.Add(1) })
synctest.Wait()
time.Sleep(4 * time.Second)
synctest.Wait()
assert.Equal(t, 0, int(calls.Load()))
time.Sleep(time.Second)
synctest.Wait()
assert.Equal(t, 1, int(calls.Load()))
time.Sleep(10 * time.Second)
synctest.Wait()
assert.Equal(t, 2, int(calls.Load()))
cancel()
synctest.Wait()
time.Sleep(time.Minute)
synctest.Wait()
assert.Equal(t, 2, int(calls.Load()))
})
}
func TestMonitorScheduleSlowProbe(t *testing.T) {
synctest.Test(t, func(t *testing.T) {
ctx, cancel := context.WithCancel(t.Context())
defer cancel()
var calls atomic.Int32
release := make(chan struct{})
go runMonitorSchedule(ctx, time.Second, 0, func() {
calls.Add(1)
select {
case <-release:
case <-ctx.Done():
}
})
synctest.Wait()
assert.Equal(t, 1, int(calls.Load()))
time.Sleep(time.Minute)
synctest.Wait()
assert.Equal(t, 1, int(calls.Load()), "a slow probe must not spawn overlapping checks")
close(release)
synctest.Wait()
assert.Equal(t, 1, int(calls.Load()), "missed intervals must not accumulate a backlog")
time.Sleep(time.Second)
synctest.Wait()
assert.Equal(t, 2, int(calls.Load()))
cancel()
synctest.Wait()
})
}
func TestMonitorScheduledAndImmediateRequestsShareProbe(t *testing.T) {
synctest.Test(t, func(t *testing.T) {
var calls atomic.Int32
release := make(chan struct{})
cfg := monitor.Config{ID: "test", Interval: 10}
pm := newMonitorManagerWithProbe(func(ctx context.Context, config monitor.Config) (int64, error) {
assert.Equal(t, cfg, config)
calls.Add(1)
<-release
return 42, nil
})
defer pm.Stop()
task := newMonitorTask(cfg)
pm.monitors[cfg.ID] = task
go runMonitorSchedule(task.ctx, 10*time.Second, 0, func() { task.runProbe(pm.probe) })
synctest.Wait()
results := make(chan *monitor.Result, 2)
for range 2 {
go func() {
result, _ := pm.UpsertMonitor(cfg, true)
results <- result
}()
}
synctest.Wait()
assert.Equal(t, 1, int(calls.Load()))
assert.Empty(t, pm.GetResults(1000), "reading history must not wait for network I/O")
close(release)
synctest.Wait()
first, second := <-results, <-results
require.NotNil(t, first)
require.NotNil(t, second)
assert.Equal(t, int64(42), first.AvgResponse)
assert.Equal(t, first, second)
assert.NotSame(t, first, second, "callers must not share mutable result pointers")
assert.Len(t, task.history.samples, 1)
// A later explicit request must still perform a fresh probe.
_, err := pm.UpsertMonitor(cfg, true)
require.NoError(t, err)
assert.Equal(t, 2, int(calls.Load()))
assert.Len(t, task.history.samples, 2)
})
}
func TestMonitorReplacementCancelsSharedProbe(t *testing.T) {
synctest.Test(t, func(t *testing.T) {
cfg := monitor.Config{ID: "test", Interval: 10}
pm := newMonitorManagerWithProbe(func(ctx context.Context, config monitor.Config) (int64, error) {
if config.Interval == 10 {
<-ctx.Done()
return 0, ctx.Err()
}
return 30, nil
})
defer pm.Stop()
task := newMonitorTask(cfg)
task.history.record(monitorSample{responseUs: 10, timestamp: time.Now()})
pm.monitors[cfg.ID] = task
results := make(chan *monitor.Result, 2)
for range 2 {
go func() {
result, _ := pm.UpsertMonitor(cfg, true)
results <- result
}()
}
synctest.Wait()
updated := cfg
updated.Interval = 20
result, err := pm.UpsertMonitor(updated, true)
require.NoError(t, err)
require.NotNil(t, result)
assert.Equal(t, int64(20), result.AvgResponse)
assert.Zero(t, result.PacketLoss)
synctest.Wait()
assert.Nil(t, <-results)
assert.Nil(t, <-results)
assert.Len(t, task.history.samples, 1)
assert.Len(t, pm.monitors[cfg.ID].history.samples, 2)
})
}
func TestMonitorInjectedProbeTimeoutRecordsLoss(t *testing.T) {
synctest.Test(t, func(t *testing.T) {
pm := newMonitorManagerWithProbe(func(ctx context.Context, _ monitor.Config) (int64, error) {
ctx, cancel := context.WithTimeout(ctx, 3*time.Second)
defer cancel()
<-ctx.Done()
return 0, ctx.Err()
})
defer pm.Stop()
start := time.Now()
result, err := pm.UpsertMonitor(monitor.Config{ID: "test", Interval: 3600}, true)
require.NoError(t, err)
require.NotNil(t, result)
assert.Equal(t, 3*time.Second, time.Since(start))
assert.Equal(t, 100.0, result.PacketLoss)
assert.NoError(t, pm.monitors["test"].ctx.Err())
})
}
-116
View File
@@ -1,116 +0,0 @@
package agent
import (
"context"
"log/slog"
"sync"
"time"
"github.com/henrygd/beszel/internal/entities/monitor"
)
const monitorFailureLogInterval = 5 * time.Minute
// monitorTask coordinates a probe and its history for one immutable configuration.
type monitorTask struct {
config monitor.Config
ctx context.Context
cancel context.CancelFunc
history *monitorHistory
resumeGuard *monitorResumeGuard
runMu sync.Mutex
inflight *monitorRun
lastFailureLog int64 // Unix nanoseconds
}
type monitorRun struct {
done chan struct{}
result *monitor.Result // published by closing done; never mutated afterwards
}
func newMonitorTask(config monitor.Config) *monitorTask {
ctx, cancel := context.WithCancel(context.Background())
task := &monitorTask{config: config, ctx: ctx, history: newMonitorHistory()}
// Serialize cancellation with publication, so canceled probes cannot enter
// history copied into a replacement task.
task.cancel = func() {
task.runMu.Lock()
cancel()
task.runMu.Unlock()
}
return task
}
func newMonitorTaskFromExisting(config monitor.Config, existing *monitorTask) *monitorTask {
task := newMonitorTask(config)
if existing != nil {
task.history = existing.history.clone()
}
return task
}
// runProbe shares an in-flight check between scheduled and immediate requests.
// Every completed check contributes exactly one sample, regardless of how many
// callers were waiting for it. No task or history lock is held during network I/O.
func (task *monitorTask) runProbe(probe monitorProbe) *monitor.Result {
task.runMu.Lock()
if task.ctx.Err() != nil {
task.runMu.Unlock()
return nil
}
if run := task.inflight; run != nil {
task.runMu.Unlock()
select {
case <-task.ctx.Done():
return nil
case <-run.done:
if task.ctx.Err() != nil {
return nil
}
return copyMonitorResult(run.result)
}
}
run := &monitorRun{done: make(chan struct{})}
task.inflight = run
task.runMu.Unlock()
generation, _ := task.resumeGuard.snapshot()
responseUs, err := probe(task.ctx, task.config)
var logFailure bool
task.runMu.Lock()
currentGeneration, _ := task.resumeGuard.snapshot()
if task.ctx.Err() == nil && generation == currentGeneration {
now := time.Now()
if err != nil {
responseUs = -1
logAt := now.UnixNano()
if task.lastFailureLog == 0 || logAt < task.lastFailureLog || logAt-task.lastFailureLog >= int64(monitorFailureLogInterval) {
logFailure = true
task.lastFailureLog = logAt
}
} else {
task.lastFailureLog = 0
}
result := task.history.record(monitorSample{responseUs: responseUs, timestamp: now})
run.result = &result
}
task.inflight = nil
close(run.done)
task.runMu.Unlock()
if logFailure {
slog.Warn("monitor failed", "err", err, "target", task.config.Target, "protocol", task.config.Protocol)
}
if task.ctx.Err() != nil {
return nil
}
return copyMonitorResult(run.result)
}
func copyMonitorResult(result *monitor.Result) *monitor.Result {
if result == nil {
return nil
}
copy := *result
return &copy
}
-79
View File
@@ -1,79 +0,0 @@
//go:build testing
package agent
import (
"bytes"
"context"
"errors"
"log/slog"
"testing"
"testing/synctest"
"time"
"github.com/henrygd/beszel/internal/entities/monitor"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
func TestMonitorFailureLogCooldown(t *testing.T) {
var logs bytes.Buffer
previous := slog.Default()
slog.SetDefault(slog.New(slog.NewTextHandler(&logs, nil)))
t.Cleanup(func() { slog.SetDefault(previous) })
synctest.Test(t, func(t *testing.T) {
task := newMonitorTask(monitor.Config{ID: "test", Target: "example.test", Protocol: "tcp"})
defer task.cancel()
failure := errors.New("connection refused")
probe := func(context.Context, monitor.Config) (int64, error) { return 42, failure }
var samples int64
check := func(wantLog bool) {
t.Helper()
logs.Reset()
result := task.runProbe(probe)
require.NotNil(t, result)
samples++
assert.Equal(t, samples, result.SampleCount, "suppressed warnings must still record samples")
if !wantLog {
assert.Empty(t, logs.String())
} else {
assert.Contains(t, logs.String(), `msg="monitor failed"`)
assert.Equal(t, 1, bytes.Count(logs.Bytes(), []byte("\n")))
}
}
check(true)
check(false)
time.Sleep(5*time.Minute - time.Nanosecond)
check(false)
time.Sleep(time.Nanosecond)
check(true)
check(false)
time.Sleep(5 * time.Minute)
check(true)
check(false)
// Recovery clears the cooldown.
failure = nil
check(false)
failure = errors.New("connection refused again")
check(true)
// Another monitor has its own cooldown.
other := newMonitorTask(task.config)
defer other.cancel()
logs.Reset()
require.NotNil(t, other.runProbe(probe))
assert.Contains(t, logs.String(), `msg="monitor failed"`)
// A canceled probe must not publish a failure or emit a warning.
logs.Reset()
result := other.runProbe(func(context.Context, monitor.Config) (int64, error) {
other.cancel()
return -1, context.Canceled
})
assert.Nil(t, result)
assert.Empty(t, logs.String())
})
}

Some files were not shown because too many files have changed in this diff Show More