i18n: update locale files

Add a new "ContainerHealth" alert type that fires when a Docker container's
health check reports unhealthy, and resolves when it recovers. This mirrors
the existing Status (up/down) alert pattern: an alert can be armed per system
and honors the "min minutes" delay before firing.

When the alert fires, the notification (email and any configured webhook,
including Discord via shoutrrr) includes a log excerpt fetched live from the
agent for up to 2 of the unhealthy containers, prioritizing lines containing
"error" or "fatal" (falling back to the log tail if none match), capped to
keep the message well under Discord's size limit.

---------

Co-authored-by: hank <hank@henrygd.me>

.github/dependabot.yml

@@ -0,0 +1,12 @@
+version: 2
+updates:
+  - package-ecosystem: gomod
+    directory: /
+    schedule:
+      interval: weekly
+
+  - package-ecosystem: github-actions
+    directory: /
+    schedule:
+      interval: weekly
+

.github/workflows/docker-images.yml

@@ -41,7 +41,7 @@ jobs:
           # henrygd/beszel-agent-nvidia
           - image: henrygd/beszel-agent-nvidia
             dockerfile: ./internal/dockerfile_agent_nvidia
-            platforms: linux/amd64
+            platforms: linux/amd64,linux/arm64
             registry: docker.io
             username_secret: DOCKERHUB_USERNAME
             password_secret: DOCKERHUB_TOKEN
@@ -52,6 +52,19 @@ jobs:
               type=semver,pattern={{major}}
               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
           - image: henrygd/beszel-agent-intel
             dockerfile: ./internal/dockerfile_agent_intel
@@ -96,7 +109,7 @@ jobs:
           # ghcr.io/henrygd/beszel-agent-nvidia
           - image: ghcr.io/${{ github.repository }}/beszel-agent-nvidia
             dockerfile: ./internal/dockerfile_agent_nvidia
-            platforms: linux/amd64
+            platforms: linux/amd64,linux/arm64
             registry: ghcr.io
             username: ${{ github.actor }}
             password_secret: GITHUB_TOKEN
@@ -107,6 +120,19 @@ jobs:
               type=semver,pattern={{major}}
               type=raw,value={{sha}},enable=${{ 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: ${{ github.actor }}
+            password_secret: GITHUB_TOKEN
+            tags: |
+              type=raw,value=slim
+              type=semver,pattern={{version}}-slim
+              type=semver,pattern={{major}}.{{minor}}-slim
+              type=semver,pattern={{major}}-slim
+
           # ghcr.io/henrygd/beszel-agent-intel
           - image: ghcr.io/${{ github.repository }}/beszel-agent-intel
             dockerfile: ./internal/dockerfile_agent_intel
@@ -152,7 +178,7 @@ jobs:
 
     steps:
       - name: Checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v7
 
       - name: Set up bun
         uses: oven-sh/setup-bun@v2
@@ -164,14 +190,14 @@ jobs:
         run: bun run --cwd ./internal/site build
 
       - name: Set up QEMU
-        uses: docker/setup-qemu-action@v3
+        uses: docker/setup-qemu-action@v4
 
       - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v3
+        uses: docker/setup-buildx-action@v4
 
       - name: Docker metadata
         id: metadata
-        uses: docker/metadata-action@v5
+        uses: docker/metadata-action@v6
         with:
           images: ${{ matrix.image }}
           tags: ${{ matrix.tags }}
@@ -181,7 +207,7 @@ jobs:
         env:
           password_secret_exists: ${{ secrets[matrix.password_secret] != '' && 'true' || 'false' }}
         if: github.event_name != 'pull_request' && env.password_secret_exists == 'true'
-        uses: docker/login-action@v3
+        uses: docker/login-action@v4
         with:
           username: ${{ matrix.username || secrets[matrix.username_secret] }}
           password: ${{ secrets[matrix.password_secret] }}
@@ -190,11 +216,13 @@ jobs:
       # Build and push Docker image with Buildx (don't push on PR)
       # https://github.com/docker/build-push-action
       - name: Build and push Docker image
-        uses: docker/build-push-action@v5
+        uses: docker/build-push-action@v7
         with:
           context: ./
           file: ${{ matrix.dockerfile }}
-          platforms: ${{ matrix.platforms || 'linux/amd64,linux/arm64,linux/arm/v7' }}
+          platforms: ${{ matrix.platforms || 'linux/amd64,linux/arm64,linux/arm/v6,linux/arm/v7' }}
           push: ${{ github.ref_type == 'tag' && secrets[matrix.password_secret] != '' }}
+          provenance: mode=max
+          sbom: true
           tags: ${{ steps.metadata.outputs.tags }}
           labels: ${{ steps.metadata.outputs.labels }}

.github/workflows/helm-charts.yml

@@ -0,0 +1,109 @@
+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}"

.github/workflows/inactivity-actions.yml

@@ -15,7 +15,7 @@ jobs:
     name: Lock Inactive Issues
     runs-on: ubuntu-24.04
     steps:
-      - uses: klaasnicolaas/action-inactivity-lock@v1.1.3
+      - uses: klaasnicolaas/action-inactivity-lock@v2.0.1
         id: lock
         with:
           days-inactive-issues: 14
@@ -29,7 +29,7 @@ jobs:
     runs-on: ubuntu-24.04
     steps:
       - name: Close Stale Issues
-        uses: actions/stale@v10
+        uses: actions/stale@v11
         with:
           repo-token: ${{ secrets.GITHUB_TOKEN }}
 

.github/workflows/release.yml

@@ -13,7 +13,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v7
         with:
           fetch-depth: 0
 
@@ -27,12 +27,12 @@ jobs:
         run: bun run --cwd ./internal/site build
 
       - name: Set up Go
-        uses: actions/setup-go@v5
+        uses: actions/setup-go@v7
         with:
-          go-version: "^1.22.1"
+          go-version: stable
 
       - name: Set up .NET
-        uses: actions/setup-dotnet@v4
+        uses: actions/setup-dotnet@v6
         with:
           dotnet-version: "9.0.x"
 
@@ -42,7 +42,7 @@ jobs:
         shell: bash
 
       - name: GoReleaser beszel
-        uses: goreleaser/goreleaser-action@v6
+        uses: goreleaser/goreleaser-action@v7
         with:
           workdir: ./
           distribution: goreleaser

.github/workflows/update-helm-charts.yml

@@ -0,0 +1,101 @@
+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

.github/workflows/vulncheck.yml

@@ -2,10 +2,6 @@
 
 name: VulnCheck
 on:
-  pull_request:
-    branches:
-      - main
-
   push:
     branches:
       - main
@@ -19,11 +15,11 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Check out code into the Go module directory
-        uses: actions/checkout@v6
+        uses: actions/checkout@v7
       - name: Set up Go
-        uses: actions/setup-go@v6
+        uses: actions/setup-go@v7
         with:
-          go-version: 1.26.x
+          go-version: stable
           # cached: false
       - name: Get official govulncheck
         run: go install golang.org/x/vuln/cmd/govulncheck@latest

.gitignore

@@ -3,7 +3,6 @@ pb_data
 data
 temp
 .vscode
-beszel-agent
 beszel_data
 beszel_data*
 dist
@@ -21,3 +20,5 @@ __debug_*
 agent/lhm/obj
 agent/lhm/bin
 dockerfile_agent_dev
+.cr-release-packages
+.tmp

.goreleaser.yml

@@ -31,12 +31,16 @@ builds:
         goarch: arm64
       - goos: freebsd
         goarch: arm
+      - goos: darwin
+        goarch: arm
 
   - id: beszel-agent
     binary: beszel-agent
     main: internal/cmd/agent/agent.go
     env:
       - CGO_ENABLED=0
+    ldflags:
+      - -s -w -X github.com/henrygd/beszel/internal/ghupdate.buildGOARM={{ .Arm }}
     goos:
       - linux
       - darwin
@@ -52,6 +56,10 @@ builds:
       - mipsle
       - mips
       - ppc64le
+    goarm:
+      - "5"
+      - "6"
+      - "7"
     gomips:
       - hardfloat
       - softfloat
@@ -71,6 +79,8 @@ builds:
         gomips: hardfloat
       - goos: windows
         goarch: arm
+      - goos: darwin
+        goarch: arm
       - goos: darwin
         goarch: riscv64
       - goos: windows
@@ -97,6 +107,7 @@ archives:
       {{ .Binary }}_
       {{- .Os }}_
       {{- .Arch }}
+      {{- if ne .Arm "6" }}{{ with .Arm }}v{{ . }}{{ end }}{{ end }}
     format_overrides:
       - goos: windows
         formats: [zip]

Makefile

@@ -52,7 +52,7 @@ lint:
 	golangci-lint run
 
 test:
-	go test -tags=testing ./...
+	go test -tags='testing no_ui' ./...
 
 tidy:
 	go mod tidy

agent/agent.go

@@ -19,6 +19,8 @@ import (
 	gossh "golang.org/x/crypto/ssh"
 )
 
+const defaultDataCacheTimeMs uint16 = 60_000
+
 type Agent struct {
 	sync.Mutex                                                                      // Used to lock agent while collecting data
 	debug                     bool                                                  // true if LOG_LEVEL is set to debug
@@ -36,6 +38,7 @@ type Agent struct {
 	sensorConfig              *SensorConfig                                         // Sensors config
 	systemInfo                system.Info                                           // Host system info (dynamic)
 	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
 	cache                     *systemDataCache                                      // Cache for system stats based on cache time
 	connectionManager         *ConnectionManager                                    // Channel to signal connection events
@@ -45,6 +48,7 @@ type Agent struct {
 	keys                      []gossh.PublicKey                                     // SSH public keys
 	smartManager              *SmartManager                                         // Manages SMART data
 	systemdManager            *systemdManager                                       // Manages systemd services
+	zfsManager                *ZfsManager                                           // Manages ZFS pool and dataset data
 }
 
 // NewAgent creates a new agent with the given data directory for persisting data.
@@ -97,7 +101,7 @@ func NewAgent(dataDir ...string) (agent *Agent, err error) {
 	slog.Debug(beszel.Version)
 
 	// initialize docker manager
-	agent.dockerManager = newDockerManager()
+	agent.dockerManager = newDockerManager(agent)
 
 	// initialize system info
 	agent.refreshSystemDetails()
@@ -118,6 +122,19 @@ func NewAgent(dataDir ...string) (agent *Agent, err error) {
 	// initialize handler registry
 	agent.handlerRegistry = NewHandlerRegistry()
 
+	agent.zfsManager = newZfsManager()
+
+	// ZFS_INTERVAL env var to update ZFS detail data at this interval
+	if zfsIntervalEnv, exists := utils.GetEnv("ZFS_INTERVAL"); exists {
+		if duration, err := time.ParseDuration(zfsIntervalEnv); err == nil && duration > 0 {
+			agent.zfsManager.detailInterval = duration
+			agent.systemDetails.ZfsInterval = duration
+			slog.Info("ZFS_INTERVAL", "duration", duration)
+		} else {
+			slog.Warn("Invalid ZFS_INTERVAL", "err", err)
+		}
+	}
+
 	// initialize disk info
 	agent.initializeDiskInfo()
 
@@ -142,7 +159,7 @@ func NewAgent(dataDir ...string) (agent *Agent, err error) {
 
 	// if debugging, print stats
 	if agent.debug {
-		slog.Debug("Stats", "data", agent.gatherStats(common.DataRequestOptions{CacheTimeMs: 60_000, IncludeDetails: true}))
+		slog.Debug("Stats", "data", agent.gatherStats(common.DataRequestOptions{CacheTimeMs: defaultDataCacheTimeMs, IncludeDetails: true}))
 	}
 
 	return agent, nil
@@ -164,11 +181,6 @@ func (a *Agent) gatherStats(options common.DataRequestOptions) *system.CombinedD
 		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)
 
 	if a.dockerManager != nil {
@@ -181,7 +193,7 @@ func (a *Agent) gatherStats(options common.DataRequestOptions) *system.CombinedD
 	}
 
 	// skip updating systemd services if cache time is not the default 60sec interval
-	if a.systemdManager != nil && cacheTimeMs == 60_000 {
+	if a.systemdManager != nil && cacheTimeMs == defaultDataCacheTimeMs {
 		totalCount := uint16(a.systemdManager.getServiceStatsCount())
 		if totalCount > 0 {
 			numFailed := a.systemdManager.getFailedServiceCount()
@@ -189,13 +201,25 @@ func (a *Agent) gatherStats(options common.DataRequestOptions) *system.CombinedD
 		}
 		if a.systemdManager.hasFreshStats {
 			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.Info.ExtraFsPct = make(map[string]float64)
 	for name, stats := range a.fsStats {
-		if !stats.Root && stats.DiskTotal > 0 {
+		if stats.Root {
+			if stats.Name != "" {
+				data.Info.RootDiskName = stats.Name
+			}
+			continue
+		}
+		if stats.DiskTotal > 0 {
 			// Use custom name if available, otherwise use device name
 			key := name
 			if stats.Name != "" {
@@ -212,7 +236,8 @@ func (a *Agent) gatherStats(options common.DataRequestOptions) *system.CombinedD
 	slog.Debug("Extra FS", "data", data.Stats.ExtraFs)
 
 	a.cache.Set(data, cacheTimeMs)
-	return data
+
+	return a.attachSystemDetails(data, cacheTimeMs, options.IncludeDetails)
 }
 
 // Start initializes and starts the agent with optional WebSocket connection

agent/battery/battery.go

@@ -1,84 +1,73 @@
-//go:build !freebsd
-
-// Package battery provides functions to check if the system has a battery and to get the battery stats.
+// Package battery provides battery information for the host and connected devices.
 package battery
 
 import (
 	"errors"
-	"log/slog"
-	"math"
-
-	"github.com/distatus/battery"
+	"sort"
+	"strconv"
+	"strings"
 )
 
-var (
-	systemHasBattery   = false
-	haveCheckedBattery = false
+const (
+	stateUnknown uint8 = iota
+	stateEmpty
+	stateFull
+	stateCharging
+	stateDischarging
+	stateIdle
 )
 
-// HasReadableBattery checks if the system has a battery and returns true if it does.
-func HasReadableBattery() bool {
-	if haveCheckedBattery {
-		return systemHasBattery
-	}
-	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
+// Battery is a readable battery reported by the operating system.
+type Battery struct {
+	Name                  string
+	Percent               uint8
+	State                 uint8
+	FullChargeCapacity    uint64
+	HasFullChargeCapacity bool
+	System                bool
 }
 
-// GetBatteryStats returns the current battery percent and charge state
-// percent = (current charge of all batteries) / (sum of designed/full capacity of all batteries)
-func GetBatteryStats() (batteryPercent uint8, batteryState uint8, err error) {
-	if !HasReadableBattery() {
-		return batteryPercent, batteryState, errors.ErrUnsupported
+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
 	}
-	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.
+	return batteries
+}
+
+// Primary returns the representative battery. Reported full-charge capacity wins,
+// then system-scoped devices, then name for deterministic ties.
+func Primary(batteries []Battery) (Battery, bool) {
 	if len(batteries) == 0 {
-		return batteryPercent, batteryState, errors.New("no batteries")
+		return Battery{}, false
 	}
-
-	totalCapacity := float64(0)
-	totalCharge := float64(0)
-	errs, partialErrs := err.(battery.Errors)
-
-	batteryState = math.MaxUint8
-
-	for i, bat := range batteries {
-		if partialErrs && errs[i] != nil {
-			// if there were some errors, like missing data, skip it
-			continue
+	ordered := append([]Battery(nil), batteries...)
+	sort.SliceStable(ordered, func(i, j int) bool {
+		a, b := ordered[i], ordered[j]
+		if a.HasFullChargeCapacity != b.HasFullChargeCapacity {
+			return a.HasFullChargeCapacity
 		}
-		if bat == nil || bat.Full == 0 {
-			// 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.HasFullChargeCapacity && a.FullChargeCapacity != b.FullChargeCapacity {
+			return a.FullChargeCapacity > b.FullChargeCapacity
 		}
-		totalCapacity += bat.Full
-		totalCharge += min(bat.Current, bat.Full)
-		if bat.State.Raw >= 0 {
-			batteryState = uint8(bat.State.Raw)
+		if a.System != b.System {
+			return a.System
 		}
-	}
-
-	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
+		return a.Name < b.Name
+	})
+	return ordered[0], true
 }

agent/battery/battery_darwin.go

@@ -0,0 +1,76 @@
+//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
+}

agent/battery/battery_freebsd.go

@@ -1,13 +0,0 @@
-//go:build freebsd
-
-package battery
-
-import "errors"
-
-func HasReadableBattery() bool {
-	return false
-}
-
-func GetBatteryStats() (uint8, uint8, error) {
-	return 0, 0, errors.ErrUnsupported
-}

agent/battery/battery_linux.go

@@ -0,0 +1,85 @@
+//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
+}

agent/battery/battery_linux_test.go

@@ -0,0 +1,109 @@
+//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())
+}

agent/battery/battery_stub.go

@@ -0,0 +1,13 @@
+//go:build !darwin && !linux && !windows
+
+package battery
+
+import "errors"
+
+func HasReadableBattery() bool {
+	return false
+}
+
+func GetBatteryStats() ([]Battery, error) {
+	return nil, errors.ErrUnsupported
+}

agent/battery/battery_test.go

@@ -0,0 +1,48 @@
+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))
+	}
+}

agent/battery/battery_windows.go

@@ -0,0 +1,291 @@
+//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
+}

agent/client.go

@@ -2,6 +2,7 @@ package agent
 
 import (
 	"crypto/tls"
+	"crypto/x509"
 	"errors"
 	"fmt"
 	"log/slog"
@@ -20,12 +21,25 @@ import (
 	"github.com/fxamacker/cbor/v2"
 	"github.com/lxzan/gws"
 	"golang.org/x/crypto/ssh"
+	"golang.org/x/net/proxy"
 )
 
 const (
 	wsDeadline = 70 * time.Second
 )
 
+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.
 // It handles authentication, message routing, and connection lifecycle management.
 type WebSocketClient struct {
@@ -39,6 +53,7 @@ type WebSocketClient struct {
 	hubRequest         *common.HubRequest[cbor.RawMessage] // Reusable request structure for message parsing
 	lastConnectAttempt time.Time                           // Timestamp of last connection attempt
 	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.
@@ -52,14 +67,18 @@ func newWebSocketClient(agent *Agent) (client *WebSocketClient, err error) {
 	client = &WebSocketClient{}
 
 	client.hubURL, err = url.Parse(hubURLStr)
-	if err != nil {
-		return nil, errors.New("invalid hub URL")
+	if err != nil || client.hubURL.Host == "" {
+		return nil, fmt.Errorf("invalid HUB_URL %q: must include scheme and host (e.g. http://hub.example.com:8090)", hubURLStr)
 	}
 	// get registration token
 	client.token, err = getToken()
 	if err != nil {
 		return nil, err
 	}
+	client.tlsConfig, err = getTLSConfig()
+	if err != nil {
+		return nil, err
+	}
 
 	client.agent = agent
 	client.hubRequest = &common.HubRequest[cbor.RawMessage]{}
@@ -86,7 +105,52 @@ func getToken() (string, error) {
 	if err != nil {
 		return "", err
 	}
-	return strings.TrimSpace(string(tokenBytes)), nil
+	return parseTokenFile(string(tokenBytes), tokenFile)
+}
+
+// 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.
@@ -104,14 +168,22 @@ func (client *WebSocketClient) getOptions() *gws.ClientOption {
 	}
 	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{
 		Addr:      client.hubURL.String(),
-		TlsConfig: &tls.Config{InsecureSkipVerify: true},
+		TlsConfig: client.tlsConfig,
 		RequestHeader: http.Header{
 			"User-Agent": []string{getUserAgent()},
 			"X-Token":    []string{client.token},
 			"X-Beszel":   []string{beszel.Version},
 		},
+		NewDialer: func() (gws.Dialer, error) {
+			return proxy.FromEnvironment(), nil
+		},
 	}
 	return client.options
 }

agent/client_test.go

@@ -4,8 +4,19 @@ package agent
 
 import (
 	"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"
 	"os"
+	"path/filepath"
 	"strings"
 	"testing"
 	"time"
@@ -15,6 +26,7 @@ import (
 	"github.com/henrygd/beszel/internal/common"
 
 	"github.com/fxamacker/cbor/v2"
+	"github.com/lxzan/gws"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 	"golang.org/x/crypto/ssh"
@@ -51,11 +63,18 @@ func TestNewWebSocketClient(t *testing.T) {
 			errorMsg:    "HUB_URL environment variable not set",
 		},
 		{
-			name:        "invalid URL",
+			name:        "malformed URL",
 			hubURL:      "ht\ttp://invalid",
 			token:       "test-token",
 			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",
@@ -157,6 +176,155 @@ 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
 func TestWebSocketClient_VerifySignature(t *testing.T) {
 	agent := createTestAgent(t)
@@ -402,6 +570,41 @@ func TestGetToken(t *testing.T) {
 		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) {
 		// Create a temporary token file
 		expectedToken := "test-token-from-beszel-file"

agent/connection_manager.go

@@ -4,11 +4,15 @@ import (
 	"context"
 	"errors"
 	"log/slog"
+	"net"
+	"os"
 	"os/signal"
+	"strings"
 	"syscall"
 	"time"
 
 	"github.com/henrygd/beszel/agent/health"
+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/entities/system"
 )
 
@@ -83,6 +87,10 @@ func (c *ConnectionManager) Start(serverOptions ServerOptions) error {
 
 	wsClient, err := newWebSocketClient(c.agent)
 	if err != nil {
+		var caCertErr *caCertFileError
+		if errors.As(err, &caCertErr) {
+			return err
+		}
 		slog.Warn("Error creating WebSocket client", "err", err)
 	}
 	c.wsClient = wsClient
@@ -111,20 +119,45 @@ func (c *ConnectionManager) Start(serverOptions ServerOptions) error {
 			_ = health.Update()
 		case <-sigCtx.Done():
 			slog.Info("Shutting down", "cause", context.Cause(sigCtx))
-			_ = c.agent.StopServer()
-			c.closeWebSocket()
-			return health.CleanUp()
+			return c.stop()
 		}
 	}
 }
 
+// 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.closeWebSocket()
+	return health.CleanUp()
+}
+
 // handleEvent processes connection events and updates the connection state accordingly.
 func (c *ConnectionManager) handleEvent(event ConnectionEvent) {
 	switch event {
 	case WebSocketConnect:
 		c.handleStateChange(WebSocketConnected)
 	case SSHConnect:
-		c.handleStateChange(SSHConnected)
+		if c.State == Disconnected {
+			c.handleStateChange(SSHConnected)
+		}
 	case WebSocketDisconnect:
 		if c.State == WebSocketConnected {
 			c.handleStateChange(Disconnected)
@@ -185,9 +218,16 @@ func (c *ConnectionManager) connect() {
 
 	// Try WebSocket first, if it fails, start SSH server
 	err := c.startWebSocketConnection()
-	if err != nil && c.State == Disconnected {
-		c.startSSHServer()
-		c.startWsTicker()
+	if err != nil {
+		if shouldExitOnErr(err) {
+			time.Sleep(2 * time.Second) // prevent tight restart loop
+			_ = c.stop()
+			os.Exit(1)
+		}
+		if c.State == Disconnected {
+			c.startSSHServer()
+			c.startWsTicker()
+		}
 	}
 }
 
@@ -224,3 +264,14 @@ func (c *ConnectionManager) closeWebSocket() {
 		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
+}

agent/connection_manager_test.go

@@ -4,6 +4,7 @@ package agent
 
 import (
 	"crypto/ed25519"
+	"errors"
 	"fmt"
 	"net"
 	"net/url"
@@ -113,6 +114,12 @@ func TestConnectionManager_EventHandling(t *testing.T) {
 			event:         SSHConnect,
 			expectedState: SSHConnected,
 		},
+		{
+			name:          "SSH connect from WebSocket connected (no change)",
+			initialState:  WebSocketConnected,
+			event:         SSHConnect,
+			expectedState: WebSocketConnected,
+		},
 		{
 			name:          "WebSocket disconnect from connected",
 			initialState:  WebSocketConnected,
@@ -264,6 +271,19 @@ func TestConnectionManager_StartWithInvalidConfig(t *testing.T) {
 	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
 func TestConnectionManager_CloseWebSocket(t *testing.T) {
 	agent := createTestAgent(t)
@@ -298,3 +318,65 @@ func TestConnectionManager_ConnectFlow(t *testing.T) {
 		cm.connect()
 	}, "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)
+		})
+	}
+}

agent/data_dir_test.go

@@ -12,6 +12,14 @@ import (
 	"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) {
 	// Test with explicit dataDir parameter
 	t.Run("explicit data dir", func(t *testing.T) {
@@ -48,7 +56,7 @@ func TestGetDataDir(t *testing.T) {
 
 	// Test with invalid explicit dataDir
 	t.Run("invalid explicit data dir", func(t *testing.T) {
-		invalidPath := "/invalid/path/that/cannot/be/created"
+		invalidPath := invalidDataDir(t)
 		_, err := GetDataDir(invalidPath)
 		assert.Error(t, err)
 	})
@@ -78,7 +86,7 @@ func TestTestDataDirs(t *testing.T) {
 	// Test with multiple directories, first one valid
 	t.Run("multiple dirs - first valid", func(t *testing.T) {
 		tempDir := t.TempDir()
-		invalidDir := "/invalid/path"
+		invalidDir := invalidDataDir(t)
 		result, err := testDataDirs([]string{tempDir, invalidDir})
 		require.NoError(t, err)
 		assert.Equal(t, tempDir, result)
@@ -87,7 +95,7 @@ func TestTestDataDirs(t *testing.T) {
 	// Test with multiple directories, second one valid
 	t.Run("multiple dirs - second valid", func(t *testing.T) {
 		tempDir := t.TempDir()
-		invalidDir := "/invalid/path"
+		invalidDir := invalidDataDir(t)
 		result, err := testDataDirs([]string{invalidDir, tempDir})
 		require.NoError(t, err)
 		assert.Equal(t, tempDir, result)
@@ -109,7 +117,7 @@ func TestTestDataDirs(t *testing.T) {
 
 	// Test with no valid directories
 	t.Run("no valid directories", func(t *testing.T) {
-		invalidPaths := []string{"/invalid/path1", "/invalid/path2"}
+		invalidPaths := []string{invalidDataDir(t), invalidDataDir(t)}
 		_, err := testDataDirs(invalidPaths)
 		assert.Error(t, err)
 		assert.Contains(t, err.Error(), "data directory not found")

agent/disk.go

@@ -1,6 +1,7 @@
 package agent
 
 import (
+	"context"
 	"log/slog"
 	"os"
 	"path/filepath"
@@ -17,10 +18,11 @@ import (
 // fsRegistrationContext holds the shared lookup state needed to resolve a
 // filesystem into the tracked fsStats key and metadata.
 type fsRegistrationContext struct {
-	filesystem     string // value of optional FILESYSTEM env var
-	isWindows      bool
-	efPath         string // path to extra filesystems (default "/extra-filesystems")
-	diskIoCounters map[string]disk.IOCountersStat
+	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
@@ -33,6 +35,34 @@ type diskDiscovery struct {
 	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"
 // Returns the device/filesystem part and the custom name part
 func parseFilesystemEntry(entry string) (device, customName string) {
@@ -148,7 +178,7 @@ func (d *diskDiscovery) addConfiguredRootFs() bool {
 
 	for _, p := range d.partitions {
 		if filesystemMatchesPartitionSetting(d.ctx.filesystem, p) {
-			d.addFsStat(p.Device, p.Mountpoint, true, "")
+			d.addFsStat(p.Device, p.Mountpoint, true, d.ctx.filesystemName)
 			return true
 		}
 	}
@@ -156,7 +186,7 @@ func (d *diskDiscovery) addConfiguredRootFs() bool {
 	// 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}
+		d.agent.fsStats[ioKey] = &system.FsStats{Root: true, Mountpoint: d.rootMountPoint, Name: d.ctx.filesystemName}
 		return true
 	}
 
@@ -238,9 +268,11 @@ func (d *diskDiscovery) addConfiguredExtraFilesystems(extraFilesystems string) {
 
 // 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.
+// 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 !strings.HasPrefix(p.Mountpoint, d.ctx.efPath) {
+	if filepath.Dir(p.Mountpoint) != d.ctx.efPath {
 		return
 	}
 	device, customName := extraFilesystemPartitionInfo(p)
@@ -269,11 +301,12 @@ func (d *diskDiscovery) addExtraFilesystemFolders(folderNames []string) {
 
 // Sets up the filesystems to monitor for disk usage and I/O.
 func (a *Agent) initializeDiskInfo() {
-	filesystem, _ := utils.GetEnv("FILESYSTEM")
+	filesystemRaw, _ := utils.GetEnv("FILESYSTEM")
+	filesystem, filesystemName := parseFilesystemEntry(filesystemRaw)
 	hasRoot := false
 	isWindows := runtime.GOOS == "windows"
 
-	partitions, err := disk.Partitions(false)
+	partitions, err := disk.PartitionsWithContext(context.Background(), true)
 	if err != nil {
 		slog.Error("Error getting disk partitions", "err", err)
 	}
@@ -292,10 +325,11 @@ func (a *Agent) initializeDiskInfo() {
 	}
 	slog.Debug("Disk I/O", "diskstats", diskIoCounters)
 	ctx := fsRegistrationContext{
-		filesystem:     filesystem,
-		isWindows:      isWindows,
-		diskIoCounters: diskIoCounters,
-		efPath:         "/extra-filesystems",
+		filesystem:         filesystem,
+		filesystemName:     filesystemName,
+		isWindows:          isWindows,
+		diskIoCounters:     diskIoCounters,
+		efPath:             "/extra-filesystems",
 	}
 
 	// Get the appropriate root mount point for this system
@@ -503,7 +537,16 @@ func normalizeDeviceName(value string) string {
 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.zfsManager != nil {
+		zfsMountpoints = a.zfsManager.ZfsMountpoints()
+	}
 	for device, stats := range a.fsStats {
+		if zfsMountpoints[stats.Mountpoint] {
+			continue
+		}
 		// skip if not in diskIoCounters
 		d, exists := diskIoCounters[device]
 		if !exists {
@@ -528,20 +571,31 @@ func (a *Agent) updateDiskUsage(systemStats *system.Stats) {
 		!a.lastDiskUsageUpdate.IsZero() &&
 		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.zfsManager != nil {
+		zfsUsage = a.zfsManager.DatasetUsage()
+	}
+
 	// disk usage
 	for _, stats := range a.fsStats {
 		// Skip non-root filesystems if caching is active
 		if cacheExtraFs && !stats.Root {
 			continue
 		}
-		if d, err := disk.Usage(stats.Mountpoint); err == nil {
-			stats.DiskTotal = utils.BytesToGigabytes(d.Total)
-			stats.DiskUsed = utils.BytesToGigabytes(d.Used)
-			if stats.Root {
-				systemStats.DiskTotal = utils.BytesToGigabytes(d.Total)
-				systemStats.DiskUsed = utils.BytesToGigabytes(d.Used)
-				systemStats.DiskPct = utils.TwoDecimals(d.UsedPercent)
+		var total, used uint64
+		var usedPct float64
+		if u, ok := zfsUsage[stats.Mountpoint]; ok {
+			total = u.used + u.avail
+			used = u.used
+			if total > 0 {
+				usedPct = float64(used) / float64(total) * 100
 			}
+		} else if d, err := disk.Usage(stats.Mountpoint); err == nil {
+			total = d.Total
+			used = d.Used
+			usedPct = d.UsedPercent
 		} else {
 			// reset stats if error (likely unmounted)
 			slog.Error("Error getting disk stats", "name", stats.Mountpoint, "err", err)
@@ -549,6 +603,14 @@ func (a *Agent) updateDiskUsage(systemStats *system.Stats) {
 			stats.DiskUsed = 0
 			stats.TotalRead = 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)
 		}
 	}
 
@@ -578,16 +640,29 @@ func (a *Agent) updateDiskIo(cacheTimeMs uint16, systemStats *system.Stats) {
 			prev, hasPrev := a.diskPrev[cacheTimeMs][name]
 			if !hasPrev {
 				// Seed from agent-level fsStats if present, else seed from current
-				prev = prevDisk{readBytes: stats.TotalRead, writeBytes: stats.TotalWrite, at: stats.Time}
+				prev = prevDisk{
+					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() {
-					prev = prevDisk{readBytes: d.ReadBytes, writeBytes: d.WriteBytes, at: now}
+					prev = prevDiskFromCounter(d, now)
 				}
 			}
 
 			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 {
-				// 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
 			}
 
@@ -599,15 +674,38 @@ func (a *Agent) updateDiskIo(cacheTimeMs uint16, systemStats *system.Stats) {
 			// validate values
 			if readMbPerSecond > 50_000 || writeMbPerSecond > 50_000 {
 				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
 				a.initializeDiskIoStats(ioCounters)
 				continue
 			}
 
-			// Update per-interval snapshot
-			a.diskPrev[cacheTimeMs][name] = prevDisk{readBytes: d.ReadBytes, writeBytes: d.WriteBytes, at: now}
+			// These properties are calculated differently on different platforms,
+			// but generally represent cumulative time spent doing reads/writes on the device.
+			// 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
 			stats.Time = now
@@ -617,20 +715,42 @@ func (a *Agent) updateDiskIo(cacheTimeMs uint16, systemStats *system.Stats) {
 			stats.DiskWritePs = writeMbPerSecond
 			stats.DiskReadBytes = diskIORead
 			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 {
 				systemStats.DiskReadPs = stats.DiskReadPs
 				systemStats.DiskWritePs = stats.DiskWritePs
 				systemStats.DiskIO[0] = diskIORead
 				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 /
 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
 	if osReleaseContent, err := os.ReadFile("/etc/os-release"); err == nil {
 		content := string(osReleaseContent)

agent/disk_test.go

@@ -78,14 +78,7 @@ func TestParseFilesystemEntry(t *testing.T) {
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			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
-			}
+			fs, customName := parseFilesystemEntry(tt.input)
 
 			assert.Equal(t, tt.expectedFs, fs)
 			assert.Equal(t, tt.expectedName, customName)
@@ -287,8 +280,9 @@ func TestAddConfiguredRootFs(t *testing.T) {
 			rootMountPoint: "/",
 			partitions:     []disk.PartitionStat{{Device: "/dev/ada0p2", Mountpoint: "/"}},
 			ctx: fsRegistrationContext{
-				filesystem: "/dev/ada0p2",
-				isWindows:  false,
+				filesystem:     "/dev/ada0p2",
+				filesystemName: "root disk",
+				isWindows:      false,
 				diskIoCounters: map[string]disk.IOCountersStat{
 					"ada0": {Name: "ada0", ReadBytes: 1000, WriteBytes: 1000},
 				},
@@ -302,6 +296,7 @@ func TestAddConfiguredRootFs(t *testing.T) {
 		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) {
@@ -310,8 +305,9 @@ func TestAddConfiguredRootFs(t *testing.T) {
 			agent:          agent,
 			rootMountPoint: "/sysroot",
 			ctx: fsRegistrationContext{
-				filesystem: "zroot",
-				isWindows:  false,
+				filesystem:     "zroot",
+				filesystemName: "root pool",
+				isWindows:      false,
 				diskIoCounters: map[string]disk.IOCountersStat{
 					"nda0": {Name: "nda0", Label: "zroot", ReadBytes: 1000, WriteBytes: 1000},
 				},
@@ -325,6 +321,7 @@ func TestAddConfiguredRootFs(t *testing.T) {
 		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) {
@@ -530,6 +527,87 @@ func TestAddExtraFilesystemFolders(t *testing.T) {
 	})
 }
 
+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{

agent/disk_zfs_test.go

@@ -0,0 +1,109 @@
+//go:build testing
+
+package agent
+
+import (
+	"testing"
+
+	"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 := &ZfsManager{}
+	zm.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"},
+		},
+		zfsManager: 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 := &ZfsManager{}
+	zm.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: "/"},
+		},
+		zfsManager: 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 := &ZfsManager{}
+	zm.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"},
+		},
+		zfsManager: 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)
+}

agent/docker.go

@@ -25,6 +25,7 @@ import (
 	"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/system"
 
 	"github.com/blang/semver"
 )
@@ -52,20 +53,21 @@ const (
 )
 
 type dockerManager struct {
-	client              *http.Client                // Client to query Docker API
-	wg                  sync.WaitGroup              // WaitGroup to wait for all goroutines to finish
-	sem                 chan struct{}               // Semaphore to limit concurrent container requests
-	containerStatsMutex sync.RWMutex                // Mutex to prevent concurrent access to containerStatsMap
-	apiContainerList    []*container.ApiInfo        // List of containers from Docker API
-	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
-	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
-	buf                 *bytes.Buffer               // Buffer to store and read response bodies
-	decoder             *json.Decoder               // Reusable JSON decoder that reads from buf
-	apiStats            *container.ApiStats         // Reusable API stats object
-	excludeContainers   []string                    // Patterns to exclude containers by name
-	usingPodman         bool                        // Whether the Docker Engine API is running on Podman
+	agent                *Agent                      // Used to propagate system detail changes back to the agent
+	client               *http.Client                // Client to query Docker API
+	wg                   sync.WaitGroup              // WaitGroup to wait for all goroutines to finish
+	sem                  chan struct{}               // Semaphore to limit concurrent container requests
+	containerStatsMutex  sync.RWMutex                // Mutex to prevent concurrent access to containerStatsMap
+	apiContainerList     []*container.ApiInfo        // List of containers from Docker API
+	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
+	goodDockerVersion    bool                        // Whether docker version is at least 25.0.0 (one-shot works correctly)
+	dockerVersionChecked bool                        // Whether a version probe has completed successfully
+	isWindows            bool                        // Whether the Docker Engine API is running on Windows
+	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
+	usingPodman          bool                        // Whether the Docker Engine API is running on Podman
 
 	// Cache-time-aware tracking for CPU stats (similar to cpu.go)
 	// Maps cache time intervals to container-specific CPU usage tracking
@@ -78,7 +80,6 @@ type dockerManager struct {
 	networkSentTrackers 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
-	retrySleep          func(time.Duration)
 }
 
 // userAgentRoundTripper is a custom http.RoundTripper that adds a User-Agent header to all requests
@@ -87,6 +88,14 @@ type userAgentRoundTripper struct {
 	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
 func (u *userAgentRoundTripper) RoundTrip(req *http.Request) (*http.Response, error) {
 	req.Header.Set("User-Agent", u.userAgent)
@@ -134,7 +143,14 @@ func (dm *dockerManager) getDockerStats(cacheTimeMs uint16) ([]*container.Stats,
 		return nil, err
 	}
 
-	dm.isWindows = strings.Contains(resp.Header.Get("Server"), "windows")
+	// Detect Podman and Windows from Server header
+	serverHeader := resp.Header.Get("Server")
+	if !dm.usingPodman && detectPodmanFromHeader(serverHeader) {
+		dm.setIsPodman()
+	}
+	dm.isWindows = strings.Contains(serverHeader, "windows")
+
+	dm.ensureDockerVersionChecked()
 
 	containersLength := len(dm.apiContainerList)
 
@@ -357,16 +373,26 @@ func convertContainerPortsToString(ctr *container.ApiInfo) string {
 		return ""
 	}
 	sort.Slice(ctr.Ports, func(i, j int) bool {
-		return ctr.Ports[i].PublicPort < ctr.Ports[j].PublicPort
+		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
-	seenPorts := make(map[uint16]struct{})
+	seen := make(map[string]struct{})
 	for _, p := range ctr.Ports {
-		_, ok := seenPorts[p.PublicPort]
-		if p.PublicPort == 0 || ok {
+		if p.PublicPort == 0 {
 			continue
 		}
-		seenPorts[p.PublicPort] = struct{}{}
+		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(", ")
 		}
@@ -518,11 +544,18 @@ func (dm *dockerManager) updateContainerStats(ctr *container.ApiInfo, cacheTimeM
 	// Get previous CPU values
 	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
 	if dm.isWindows {
 		prevRead := dm.lastCpuReadTime[cacheTimeMs][ctr.IdShort]
 		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 {
 		cpuPct = res.CalculateCpuPercentLinux(prevCpuContainer, prevCpuSystem)
 	}
@@ -588,7 +621,7 @@ func (dm *dockerManager) deleteContainerStatsSync(id string) {
 }
 
 // Creates a new http client for Docker or Podman API
-func newDockerManager() *dockerManager {
+func newDockerManager(agent *Agent) *dockerManager {
 	dockerHost, exists := utils.GetEnv("DOCKER_HOST")
 	if exists {
 		// return nil if set to empty string
@@ -654,6 +687,7 @@ func newDockerManager() *dockerManager {
 	}
 
 	manager := &dockerManager{
+		agent: agent,
 		client: &http.Client{
 			Timeout:   timeout,
 			Transport: userAgentTransport,
@@ -671,51 +705,54 @@ func newDockerManager() *dockerManager {
 		networkSentTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
 		networkRecvTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
 		lastNetworkReadTime: make(map[uint16]map[string]time.Time),
-		retrySleep:          time.Sleep,
 	}
 
-	// If using podman, return client
-	if strings.Contains(dockerHost, "podman") {
-		manager.usingPodman = true
-		manager.goodDockerVersion = true
-		return manager
-	}
-
-	// run version check in goroutine to avoid blocking (server may not be ready and requires retries)
-	go manager.checkDockerVersion()
-
-	// give version check a chance to complete before returning
-	time.Sleep(50 * time.Millisecond)
+	// Best-effort startup probe. If the engine is not ready yet, getDockerStats will
+	// retry after the first successful /containers/json request.
+	_, _ = manager.checkDockerVersion()
 
 	return manager
 }
 
 // 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.
-func (dm *dockerManager) checkDockerVersion() {
-	var err error
-	var resp *http.Response
-	var versionInfo struct {
-		Version string `json:"Version"`
+func (dm *dockerManager) checkDockerVersion() (bool, error) {
+	resp, err := dm.client.Get("http://localhost/version")
+	if err != nil {
+		return false, err
 	}
-	const versionMaxTries = 2
-	for i := 1; i <= versionMaxTries; i++ {
-		resp, err = dm.client.Get("http://localhost/version")
-		if err == nil && resp.StatusCode == http.StatusOK {
-			break
-		}
-		if resp != nil {
-			resp.Body.Close()
-		}
-		if i < versionMaxTries {
-			slog.Debug("Failed to get Docker version; retrying", "attempt", i, "err", err, "response", resp)
-			dm.retrySleep(5 * time.Second)
-		}
+	if resp.StatusCode != http.StatusOK {
+		status := resp.Status
+		resp.Body.Close()
+		return false, fmt.Errorf("docker version request failed: %s", status)
 	}
-	if err != nil || resp.StatusCode != http.StatusOK {
+
+	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
 	}
-	if err := dm.decode(resp, &versionInfo); err != nil {
+	if _, err := dm.checkDockerVersion(); 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
 	}
 	// if version > 24, one-shot works correctly and we can limit concurrent operations
@@ -726,20 +763,18 @@ func (dm *dockerManager) checkDockerVersion() {
 	}
 }
 
-// Decodes Docker API JSON response using a reusable buffer and decoder. Not thread safe.
+// Decodes a Docker API JSON response using a reusable buffer. Not thread safe.
 func (dm *dockerManager) decode(resp *http.Response, d any) error {
 	if dm.buf == nil {
 		// initialize buffer with 256kb starting size
 		dm.buf = bytes.NewBuffer(make([]byte, 0, 1024*256))
-		dm.decoder = json.NewDecoder(dm.buf)
 	}
 	defer resp.Body.Close()
 	defer dm.buf.Reset()
-	_, err := dm.buf.ReadFrom(resp.Body)
-	if err != nil {
+	if _, err := dm.buf.ReadFrom(resp.Body); err != nil {
 		return err
 	}
-	return dm.decoder.Decode(d)
+	return json.Unmarshal(dm.buf.Bytes(), d)
 }
 
 // Test docker / podman sockets and return if one exists
@@ -941,3 +976,46 @@ func (dm *dockerManager) GetHostInfo() (info container.HostInfo, err error) {
 func (dm *dockerManager) IsPodman() bool {
 	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)
+}

agent/docker_test.go

@@ -539,59 +539,53 @@ func TestDockerManagerCreation(t *testing.T) {
 
 func TestCheckDockerVersion(t *testing.T) {
 	tests := []struct {
-		name      string
-		responses []struct {
-			statusCode int
-			body       string
-		}
-		expectedGood     bool
-		expectedRequests int
+		name            string
+		statusCode      int
+		body            string
+		server          string
+		expectSuccess   bool
+		expectedGood    bool
+		expectedPodman  bool
+		expectError     bool
+		expectedRequest string
 	}{
 		{
-			name: "200 with good version on first try",
-			responses: []struct {
-				statusCode int
-				body       string
-			}{
-				{http.StatusOK, `{"Version":"25.0.1"}`},
-			},
-			expectedGood:     true,
-			expectedRequests: 1,
+			name:            "good docker version",
+			statusCode:      http.StatusOK,
+			body:            `{"Version":"25.0.1"}`,
+			expectSuccess:   true,
+			expectedGood:    true,
+			expectedPodman:  false,
+			expectedRequest: "/version",
 		},
 		{
-			name: "200 with old version on first try",
-			responses: []struct {
-				statusCode int
-				body       string
-			}{
-				{http.StatusOK, `{"Version":"24.0.7"}`},
-			},
-			expectedGood:     false,
-			expectedRequests: 1,
+			name:            "old docker version",
+			statusCode:      http.StatusOK,
+			body:            `{"Version":"24.0.7"}`,
+			expectSuccess:   true,
+			expectedGood:    false,
+			expectedPodman:  false,
+			expectedRequest: "/version",
 		},
 		{
-			name: "non-200 then 200 with good version",
-			responses: []struct {
-				statusCode int
-				body       string
-			}{
-				{http.StatusServiceUnavailable, `"not ready"`},
-				{http.StatusOK, `{"Version":"25.1.0"}`},
-			},
-			expectedGood:     true,
-			expectedRequests: 2,
+			name:            "podman from server header",
+			statusCode:      http.StatusOK,
+			body:            `{"Version":"5.5.0"}`,
+			server:          "Libpod/5.5.0",
+			expectSuccess:   true,
+			expectedGood:    true,
+			expectedPodman:  true,
+			expectedRequest: "/version",
 		},
 		{
-			name: "non-200 on all retries",
-			responses: []struct {
-				statusCode int
-				body       string
-			}{
-				{http.StatusInternalServerError, `"error"`},
-				{http.StatusUnauthorized, `"error"`},
-			},
-			expectedGood:     false,
-			expectedRequests: 2,
+			name:            "non-200 response",
+			statusCode:      http.StatusServiceUnavailable,
+			body:            `"not ready"`,
+			expectSuccess:   false,
+			expectedGood:    false,
+			expectedPodman:  false,
+			expectError:     true,
+			expectedRequest: "/version",
 		},
 	}
 
@@ -599,13 +593,13 @@ func TestCheckDockerVersion(t *testing.T) {
 		t.Run(tt.name, func(t *testing.T) {
 			requestCount := 0
 			server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-				idx := requestCount
 				requestCount++
-				if idx >= len(tt.responses) {
-					idx = len(tt.responses) - 1
+				assert.Equal(t, tt.expectedRequest, r.URL.EscapedPath())
+				if tt.server != "" {
+					w.Header().Set("Server", tt.server)
 				}
-				w.WriteHeader(tt.responses[idx].statusCode)
-				fmt.Fprint(w, tt.responses[idx].body)
+				w.WriteHeader(tt.statusCode)
+				fmt.Fprint(w, tt.body)
 			}))
 			defer server.Close()
 
@@ -617,17 +611,24 @@ func TestCheckDockerVersion(t *testing.T) {
 						},
 					},
 				},
-				retrySleep: func(time.Duration) {},
 			}
 
-			dm.checkDockerVersion()
+			success, err := dm.checkDockerVersion()
 
+			assert.Equal(t, tt.expectSuccess, success)
+			assert.Equal(t, tt.expectSuccess, dm.dockerVersionChecked)
 			assert.Equal(t, tt.expectedGood, dm.goodDockerVersion)
-			assert.Equal(t, tt.expectedRequests, requestCount)
+			assert.Equal(t, tt.expectedPodman, dm.usingPodman)
+			assert.Equal(t, 1, requestCount)
+			if tt.expectError {
+				require.Error(t, err)
+			} else {
+				require.NoError(t, err)
+			}
 		})
 	}
 
-	t.Run("request error on all retries", func(t *testing.T) {
+	t.Run("request error", func(t *testing.T) {
 		requestCount := 0
 		dm := &dockerManager{
 			client: &http.Client{
@@ -638,16 +639,191 @@ func TestCheckDockerVersion(t *testing.T) {
 					},
 				},
 			},
-			retrySleep: func(time.Duration) {},
 		}
 
-		dm.checkDockerVersion()
+		success, err := dm.checkDockerVersion()
 
+		assert.False(t, success)
+		require.Error(t, err)
+		assert.False(t, dm.dockerVersionChecked)
 		assert.False(t, dm.goodDockerVersion)
-		assert.Equal(t, 2, requestCount)
+		assert.False(t, dm.usingPodman)
+		assert.Equal(t, 1, requestCount)
 	})
 }
 
+// newDockerManagerForVersionTest creates a dockerManager wired to a test server.
+func newDockerManagerForVersionTest(server *httptest.Server) *dockerManager {
+	return &dockerManager{
+		client: &http.Client{
+			Transport: &http.Transport{
+				DialContext: func(_ context.Context, network, _ string) (net.Conn, error) {
+					return net.Dial(network, server.Listener.Addr().String())
+				},
+			},
+		},
+		containerStatsMap:   make(map[string]*container.Stats),
+		lastCpuContainer:    make(map[uint16]map[string]uint64),
+		lastCpuSystem:       make(map[uint16]map[string]uint64),
+		lastCpuReadTime:     make(map[uint16]map[string]time.Time),
+		networkSentTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+		networkRecvTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+		lastNetworkReadTime: make(map[uint16]map[string]time.Time),
+	}
+}
+
+func TestGetDockerStatsChecksDockerVersionAfterContainerList(t *testing.T) {
+	tests := []struct {
+		name            string
+		containerServer string
+		versionServer   string
+		versionBody     string
+		expectedGood    bool
+		expectedPodman  bool
+	}{
+		{
+			name:           "200 with good version on first try",
+			versionBody:    `{"Version":"25.0.1"}`,
+			expectedGood:   true,
+			expectedPodman: false,
+		},
+		{
+			name:           "200 with old version on first try",
+			versionBody:    `{"Version":"24.0.7"}`,
+			expectedGood:   false,
+			expectedPodman: false,
+		},
+		{
+			name:            "podman detected from server header",
+			containerServer: "Libpod/5.5.0",
+			expectedGood:    true,
+			expectedPodman:  true,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			requestCounts := map[string]int{}
+			server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+				requestCounts[r.URL.EscapedPath()]++
+				switch r.URL.EscapedPath() {
+				case "/containers/json":
+					if tt.containerServer != "" {
+						w.Header().Set("Server", tt.containerServer)
+					}
+					w.WriteHeader(http.StatusOK)
+					fmt.Fprint(w, `[]`)
+				case "/version":
+					if tt.versionServer != "" {
+						w.Header().Set("Server", tt.versionServer)
+					}
+					w.WriteHeader(http.StatusOK)
+					fmt.Fprint(w, tt.versionBody)
+				default:
+					t.Fatalf("unexpected path: %s", r.URL.EscapedPath())
+				}
+			}))
+			defer server.Close()
+
+			dm := newDockerManagerForVersionTest(server)
+
+			stats, err := dm.getDockerStats(defaultCacheTimeMs)
+			require.NoError(t, err)
+			require.NotNil(t, stats, "A successful empty snapshot must remain distinguishable from a collection failure")
+			assert.Empty(t, stats)
+			assert.True(t, dm.dockerVersionChecked)
+			assert.Equal(t, tt.expectedGood, dm.goodDockerVersion)
+			assert.Equal(t, tt.expectedPodman, dm.usingPodman)
+			assert.Equal(t, 1, requestCounts["/containers/json"])
+			if tt.expectedPodman {
+				assert.Equal(t, 0, requestCounts["/version"])
+			} else {
+				assert.Equal(t, 1, requestCounts["/version"])
+			}
+
+			stats, err = dm.getDockerStats(defaultCacheTimeMs)
+			require.NoError(t, err)
+			require.NotNil(t, stats, "A successful empty snapshot must remain distinguishable from a collection failure")
+			assert.Empty(t, stats)
+			assert.Equal(t, tt.expectedGood, dm.goodDockerVersion)
+			assert.Equal(t, tt.expectedPodman, dm.usingPodman)
+			assert.Equal(t, 2, requestCounts["/containers/json"])
+			if tt.expectedPodman {
+				assert.Equal(t, 0, requestCounts["/version"])
+			} else {
+				assert.Equal(t, 1, requestCounts["/version"])
+			}
+		})
+	}
+
+}
+
+func TestGetDockerStatsRetriesVersionCheckUntilSuccess(t *testing.T) {
+	requestCounts := map[string]int{}
+	versionStatuses := []int{http.StatusServiceUnavailable, http.StatusOK}
+	versionBodies := []string{`"not ready"`, `{"Version":"25.1.0"}`}
+
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		requestCounts[r.URL.EscapedPath()]++
+		switch r.URL.EscapedPath() {
+		case "/containers/json":
+			w.WriteHeader(http.StatusOK)
+			fmt.Fprint(w, `[]`)
+		case "/version":
+			idx := requestCounts["/version"] - 1
+			if idx >= len(versionStatuses) {
+				idx = len(versionStatuses) - 1
+			}
+			w.WriteHeader(versionStatuses[idx])
+			fmt.Fprint(w, versionBodies[idx])
+		default:
+			t.Fatalf("unexpected path: %s", r.URL.EscapedPath())
+		}
+	}))
+	defer server.Close()
+
+	dm := newDockerManagerForVersionTest(server)
+
+	stats, err := dm.getDockerStats(defaultCacheTimeMs)
+	require.NoError(t, err)
+	assert.Empty(t, stats)
+	assert.False(t, dm.dockerVersionChecked)
+	assert.False(t, dm.goodDockerVersion)
+	assert.Equal(t, 1, requestCounts["/version"])
+
+	stats, err = dm.getDockerStats(defaultCacheTimeMs)
+	require.NoError(t, err)
+	assert.Empty(t, stats)
+	assert.True(t, dm.dockerVersionChecked)
+	assert.True(t, dm.goodDockerVersion)
+	assert.Equal(t, 2, requestCounts["/containers/json"])
+	assert.Equal(t, 2, requestCounts["/version"])
+
+	stats, err = dm.getDockerStats(defaultCacheTimeMs)
+	require.NoError(t, err)
+	assert.Empty(t, stats)
+	assert.Equal(t, 3, requestCounts["/containers/json"])
+	assert.Equal(t, 2, requestCounts["/version"])
+}
+
+// A failed decode must not break later decodes. Previously the reused json.Decoder
+// stayed desynced after one truncated response, breaking decode until restart.
+func TestDecodeRecoversFromError(t *testing.T) {
+	dm := &dockerManager{}
+
+	// truncated JSON: body reads fine, decode fails
+	var bad []container.ApiInfo
+	err := dm.decode(&http.Response{Body: io.NopCloser(strings.NewReader(`[{"Id":"abc`))}, &bad)
+	require.Error(t, err)
+
+	// the next decode must still succeed
+	var good []container.ApiInfo
+	err = dm.decode(&http.Response{Body: io.NopCloser(strings.NewReader(`[{"Id":"abcdef012345","Names":["/ok"]}]`))}, &good)
+	require.NoError(t, err)
+	require.Len(t, good, 1)
+	assert.Equal(t, "abcdef012345", good[0].Id)
+}
+
 func TestCycleCpuDeltas(t *testing.T) {
 	dm := &dockerManager{
 		lastCpuContainer: map[uint16]map[string]uint64{
@@ -847,6 +1023,200 @@ func TestCpuPercentageCalculationWithRealData(t *testing.T) {
 	assert.InDelta(t, expectedPct, actualPct, 0.01)
 }
 
+func TestCpuPercentageHandlesCounterRollback(t *testing.T) {
+	// If a stats response is processed after a newer one for the same container,
+	// or an accounting counter resets, the current total can be lower than the
+	// stored previous value. Unsigned subtraction wraps to ~2^64 instead of
+	// going negative, so the percentage explodes, validateCpuPercentage rejects
+	// the sample, and the whole collection is discarded - network stats too.
+	stats := &container.ApiStats{
+		CPUStats: container.CPUStats{
+			CPUUsage:    container.CPUUsage{TotalUsage: 1_000_000},
+			SystemUsage: 20_000_000,
+		},
+	}
+
+	// Container counter went backwards.
+	assert.Equal(t, 0.0, stats.CalculateCpuPercentLinux(2_000_000, 10_000_000))
+	// System counter went backwards.
+	assert.Equal(t, 0.0, stats.CalculateCpuPercentLinux(500_000, 30_000_000))
+	// A normal forward sample is unaffected: 500000 / 10000000 * 100 = 5%.
+	assert.InDelta(t, 5.0, stats.CalculateCpuPercentLinux(500_000, 10_000_000), 0.001)
+}
+
+func TestCpuPercentageWindowsHandlesCounterRollback(t *testing.T) {
+	now := time.Now()
+	stats := &container.ApiStats{
+		Read:     now,
+		NumProcs: 4,
+		CPUStats: container.CPUStats{
+			CPUUsage: container.CPUUsage{TotalUsage: 1_000_000},
+		},
+	}
+	prevRead := now.Add(-time.Second)
+
+	// Container counter went backwards.
+	assert.Equal(t, 0.0, stats.CalculateCpuPercentWindows(2_000_000, prevRead))
+	// A normal forward sample is unaffected.
+	assert.Greater(t, stats.CalculateCpuPercentWindows(500_000, prevRead), 0.0)
+}
+
+func TestCalculateCpuPercentPodman(t *testing.T) {
+	baseTime := time.Date(2026, 3, 15, 12, 0, 0, 0, time.UTC)
+
+	tests := []struct {
+		name             string
+		prevCpuContainer uint64
+		prevRead         time.Time
+		currentUsage     uint64
+		currentRead      time.Time
+		onlineCPUs       uint32
+		expectedPct      float64
+	}{
+		{
+			name: "normal calculation",
+			// container used 2ms of CPU over 1s with 2 CPUs → 0.1%
+			prevCpuContainer: 1_000_000_000,
+			prevRead:         baseTime,
+			currentUsage:     1_002_000_000, // +2ms CPU time
+			currentRead:      baseTime.Add(time.Second),
+			onlineCPUs:       2,
+			expectedPct:      0.1, // 2e6 / (1e9 * 2) * 100
+		},
+		{
+			name:             "first run returns zero",
+			prevCpuContainer: 0,
+			prevRead:         baseTime,
+			currentUsage:     5_000_000,
+			currentRead:      baseTime.Add(time.Second),
+			onlineCPUs:       4,
+			expectedPct:      0.0,
+		},
+		{
+			name:             "zero online cpus returns zero",
+			prevCpuContainer: 1_000_000_000,
+			prevRead:         baseTime,
+			currentUsage:     1_010_000_000,
+			currentRead:      baseTime.Add(time.Second),
+			onlineCPUs:       0,
+			expectedPct:      0.0,
+		},
+		{
+			name:             "same read time returns zero",
+			prevCpuContainer: 1_000_000_000,
+			prevRead:         baseTime,
+			currentUsage:     1_010_000_000,
+			currentRead:      baseTime, // no elapsed time
+			onlineCPUs:       2,
+			expectedPct:      0.0,
+		},
+		{
+			name:             "counter rollback returns zero",
+			prevCpuContainer: 2_000_000_000,
+			prevRead:         baseTime,
+			currentUsage:     1_000_000_000,
+			currentRead:      baseTime.Add(time.Second),
+			onlineCPUs:       2,
+			expectedPct:      0.0,
+		},
+		{
+			name: "100% single cpu",
+			// container consumed a full CPU-second over 1s on a 1-CPU host → 100%
+			prevCpuContainer: 1_000_000_000,
+			prevRead:         baseTime,
+			currentUsage:     2_000_000_000, // +1s CPU time
+			currentRead:      baseTime.Add(time.Second),
+			onlineCPUs:       1,
+			expectedPct:      100.0, // 1e9 / (1e9 * 1) * 100
+		},
+		{
+			name: "high utilization on multi-cpu host",
+			// container used 800ms on a 4-CPU host over 1s → 20%
+			prevCpuContainer: 10_000_000_000,
+			prevRead:         baseTime,
+			currentUsage:     10_800_000_000,
+			currentRead:      baseTime.Add(time.Second),
+			onlineCPUs:       4,
+			expectedPct:      20.0, // 800e6 / (1e9 * 4) * 100
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			s := &container.ApiStats{
+				Read: tt.currentRead,
+				CPUStats: container.CPUStats{
+					CPUUsage:   container.CPUUsage{TotalUsage: tt.currentUsage},
+					OnlineCPUs: tt.onlineCPUs,
+				},
+			}
+			got := s.CalculateCpuPercentPodman(tt.prevCpuContainer, tt.prevRead)
+			assert.InDelta(t, tt.expectedPct, got, 0.001, "test %q", tt.name)
+		})
+	}
+}
+
+func TestUpdateContainerStatsPodmanCpuCalculation(t *testing.T) {
+	// Verify that Podman containers use the time-based CPU calculation
+	// when online_cpus is provided in the stats response.
+	// container used 20ms CPU over 1s with 2 CPUs → 1%
+	prevReadTime := time.Date(2026, 3, 15, 21, 26, 58, 0, time.UTC) // 1 second before stats read
+	const prevCpuUsage = uint64(5_000_000_000)
+
+	dm := &dockerManager{
+		client: &http.Client{Transport: roundTripFunc(func(req *http.Request) (*http.Response, error) {
+			switch req.URL.EscapedPath() {
+			case "/containers/0123456789ab/stats":
+				return &http.Response{
+					StatusCode: http.StatusOK,
+					Status:     "200 OK",
+					Header:     make(http.Header),
+					Body: io.NopCloser(strings.NewReader(`{
+						"read":"2026-03-15T21:26:59Z",
+						"cpu_stats":{"cpu_usage":{"total_usage":5020000000},"system_cpu_usage":9999999,"online_cpus":2},
+						"memory_stats":{"usage":1048576,"stats":{"inactive_file":262144}},
+						"networks":{"eth0":{"rx_bytes":0,"tx_bytes":0}}
+					}`)),
+					Request: req,
+				}, nil
+			default:
+				return nil, fmt.Errorf("unexpected path: %s", req.URL.EscapedPath())
+			}
+		})},
+		containerStatsMap: make(map[string]*container.Stats),
+		apiStats:          &container.ApiStats{},
+		usingPodman:       true,
+		lastCpuContainer: map[uint16]map[string]uint64{
+			defaultCacheTimeMs: {"0123456789ab": prevCpuUsage},
+		},
+		lastCpuSystem: map[uint16]map[string]uint64{
+			defaultCacheTimeMs: {"0123456789ab": 1}, // intentionally tiny — should NOT be used
+		},
+		lastCpuReadTime: map[uint16]map[string]time.Time{
+			defaultCacheTimeMs: {"0123456789ab": prevReadTime},
+		},
+		networkSentTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+		networkRecvTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+		lastNetworkReadTime: make(map[uint16]map[string]time.Time),
+	}
+
+	ctr := &container.ApiInfo{
+		IdShort: "0123456789ab",
+		Names:   []string{"/myapp"},
+		Status:  "Up 5 minutes",
+		Image:   "myapp:latest",
+	}
+
+	err := dm.updateContainerStats(ctr, defaultCacheTimeMs)
+	require.NoError(t, err)
+
+	// cpu delta = 5020000000 - 5000000000 = 20000000 ns (20ms)
+	// elapsed = 1s = 1000000000 ns, online_cpus = 2
+	// expected = 20000000 / (1000000000 * 2) * 100 = 1.0%
+	expectedCpu := 1.0
+	assert.InDelta(t, expectedCpu, dm.containerStatsMap[ctr.IdShort].Cpu, 0.01)
+}
+
 func TestNetworkStatsCalculationWithRealData(t *testing.T) {
 	// Create synthetic test data to avoid timing issues
 	apiStats1 := &container.ApiStats{
@@ -1708,6 +2078,14 @@ func TestConvertContainerPortsToString(t *testing.T) {
 			},
 			expected: "80, 443",
 		},
+		{
+			name: "ipv4 and ipv6 wildcard bindings are deduplicated",
+			ports: []port{
+				{PublicPort: 80, IP: "0.0.0.0"},
+				{PublicPort: 80, IP: "::"},
+			},
+			expected: "80",
+		},
 		{
 			name: "multiple ports with different IPs",
 			ports: []port{
@@ -1716,6 +2094,22 @@ func TestConvertContainerPortsToString(t *testing.T) {
 			},
 			expected: "80, 1.2.3.4:443",
 		},
+		{
+			name: "same port bound to multiple IPs shows all entries",
+			ports: []port{
+				{PublicPort: 65533, IP: "172.16.151.72"},
+				{PublicPort: 65533, IP: "172.16.156.25"},
+			},
+			expected: "172.16.151.72:65533, 172.16.156.25:65533",
+		},
+		{
+			name: "same port bound to IPv4 and IPv6",
+			ports: []port{
+				{PublicPort: 65534, IP: "172.16.151.72"},
+				{PublicPort: 65534, IP: "fd04:38e2:98c6:3fd::72"},
+			},
+			expected: "172.16.151.72:65534, fd04:38e2:98c6:3fd::72:65534",
+		},
 		{
 			name: "ports slice is nilled after call",
 			ports: []port{

agent/fans.go

@@ -0,0 +1,117 @@
+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
+}

agent/fans_linux.go

@@ -0,0 +1,8 @@
+//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"

agent/fans_other.go

@@ -0,0 +1,7 @@
+//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 = ""

agent/fans_test.go

@@ -0,0 +1,105 @@
+//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)
+}

agent/fingerprint.go

@@ -50,6 +50,9 @@ func generateFingerprint(hostname, cpuModel string) string {
 			if info, err := cpu.Info(); err == nil && len(info) > 0 {
 				cpuModel = info[0].ModelName
 			}
+			if cpuModel == "" {
+				cpuModel = getCpuModelFromCpuinfo()
+			}
 		}
 		fingerprint = hostname + cpuModel
 	}

agent/gpu.go

@@ -48,6 +48,8 @@ type GPUManager struct {
 	// Per-cache-key tracking for delta calculations
 	// cacheKey -> gpuId -> snapshot of last count/usage/power values
 	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
@@ -90,6 +92,7 @@ const (
 	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)
@@ -106,6 +109,7 @@ func isValidCollectorSource(source collectorSource) bool {
 		collectorSourceNVML,
 		collectorSourceNvidiaSMI,
 		collectorSourceIntelGpuTop,
+		collectorSourceIntelSysfs,
 		collectorSourceAmdSysfs,
 		collectorSourceRocmSMI,
 		collectorSourceMacmon,
@@ -122,6 +126,8 @@ type gpuCapabilities struct {
 	hasAmdSysfs     bool
 	hasTegrastats   bool
 	hasIntelGpuTop  bool
+	hasXe           bool
+	hasIntelSysfs   bool
 	hasNvtop        bool
 	hasMacmon       bool
 	hasPowermetrics bool
@@ -355,12 +361,16 @@ func (gm *GPUManager) calculateGPUAverage(id string, gpu *system.GPUData, cacheK
 
 	// If no new data arrived
 	if deltaCount == 0 {
-		// If GPU appears suspended (instantaneous values are 0), return zero values
-		// Otherwise return last known average for temporary collection gaps
-		if gpu.Temperature == 0 && gpu.MemoryUsed == 0 {
+		// Only discrete GPUs report temp/memory, so treat all-zero as suspended (return zeros).
+		// Engine-based (Intel) GPUs don't, so carry the last average forward across sample gaps.
+		if gpu.Engines == nil && gpu.Temperature == 0 && gpu.MemoryUsed == 0 {
 			return system.GPUData{Name: gpu.Name}
 		}
-		return gm.lastAvgData[id] // zero value if not found
+		lastAvg := gm.lastAvgData[id] // zero value if not found
+		if lastAvg.Name == "" {
+			lastAvg.Name = gpu.Name
+		}
+		return lastAvg
 	}
 
 	// Calculate new average
@@ -369,12 +379,13 @@ func (gm *GPUManager) calculateGPUAverage(id string, gpu *system.GPUData, cacheK
 
 	gpuAvg.Power = utils.TwoDecimals(deltaPower / float64(deltaCount))
 
+	gpuAvg.PowerPkg = utils.TwoDecimals(deltaPowerPkg / float64(deltaCount))
+
 	if gpu.Engines != nil {
 		// make fresh map for averaged engine metrics to avoid mutating
 		// the accumulator map stored in gm.GpuDataMap
 		gpuAvg.Engines = make(map[string]float64, len(gpu.Engines))
 		gpuAvg.Usage = gm.calculateIntelGPUUsage(&gpuAvg, gpu, lastSnapshot, deltaCount)
-		gpuAvg.PowerPkg = utils.TwoDecimals(deltaPowerPkg / float64(deltaCount))
 	} else {
 		gpuAvg.Usage = utils.TwoDecimals(deltaUsage / float64(deltaCount))
 	}
@@ -444,6 +455,8 @@ func (gm *GPUManager) storeSnapshot(id string, gpu *system.GPUData, cacheKey uin
 func (gm *GPUManager) discoverGpuCapabilities() gpuCapabilities {
 	caps := gpuCapabilities{
 		hasAmdSysfs: gm.hasAmdSysfs(),
+		hasXe:       gm.hasXe(),
+		hasIntelSysfs: gm.hasIntelSysfs(),
 	}
 	if _, err := exec.LookPath(nvidiaSmiCmd); err == nil {
 		caps.hasNvidiaSmi = true
@@ -472,7 +485,7 @@ func (gm *GPUManager) discoverGpuCapabilities() gpuCapabilities {
 }
 
 func hasAnyGpuCollector(caps gpuCapabilities) bool {
-	return caps.hasNvidiaSmi || caps.hasRocmSmi || caps.hasAmdSysfs || caps.hasTegrastats || caps.hasIntelGpuTop || caps.hasNvtop || caps.hasMacmon || caps.hasPowermetrics
+	return caps.hasNvidiaSmi || caps.hasRocmSmi || caps.hasAmdSysfs || caps.hasTegrastats || caps.hasIntelGpuTop || caps.hasIntelSysfs || caps.hasNvtop || caps.hasMacmon || caps.hasPowermetrics
 }
 
 func (gm *GPUManager) startIntelCollector() {
@@ -542,7 +555,7 @@ func (gm *GPUManager) collectorDefinitions(caps gpuCapabilities) map[collectorSo
 	return map[collectorSource]collectorDefinition{
 		collectorSourceNVML: {
 			group:     collectorGroupNvidia,
-			available: caps.hasNvidiaSmi,
+			available: true,
 			start: func(_ func()) bool {
 				return gm.startNvmlCollector()
 			},
@@ -563,6 +576,13 @@ func (gm *GPUManager) collectorDefinitions(caps gpuCapabilities) map[collectorSo
 				return true
 			},
 		},
+		collectorSourceIntelSysfs: {
+			group:     collectorGroupIntel,
+			available: caps.hasIntelSysfs,
+			start: func(_ func()) bool {
+				return gm.startIntelSysfsCollector()
+			},
+		},
 		collectorSourceAmdSysfs: {
 			group:     collectorGroupAmd,
 			available: caps.hasAmdSysfs,
@@ -705,9 +725,12 @@ func (gm *GPUManager) resolveLegacyCollectorPriority(caps gpuCapabilities) []col
 		priorities = append(priorities, collectorSourceAmdSysfs)
 	}
 
-	if caps.hasIntelGpuTop {
+	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.
@@ -734,9 +757,6 @@ func NewGPUManager() (*GPUManager, error) {
 	}
 	var gm GPUManager
 	caps := gm.discoverGpuCapabilities()
-	if !hasAnyGpuCollector(caps) {
-		return nil, fmt.Errorf(noGPUFoundMsg)
-	}
 	gm.GpuDataMap = make(map[string]*system.GPUData)
 
 	// Jetson devices should always use tegrastats (ignore GPU_COLLECTOR).
@@ -745,7 +765,7 @@ func NewGPUManager() (*GPUManager, error) {
 		return &gm, nil
 	}
 
-	// if GPU_COLLECTOR is set, start user-defined collectors.
+	// Respect explicit collector selection before capability auto-detection.
 	if collectorConfig, ok := utils.GetEnv("GPU_COLLECTOR"); ok && strings.TrimSpace(collectorConfig) != "" {
 		priorities := parseCollectorPriority(collectorConfig)
 		if gm.startCollectorsByPriority(priorities, caps) == 0 {
@@ -754,6 +774,10 @@ func NewGPUManager() (*GPUManager, error) {
 		return &gm, nil
 	}
 
+	if !hasAnyGpuCollector(caps) {
+		return nil, fmt.Errorf(noGPUFoundMsg)
+	}
+
 	// auto-detect and start collectors when GPU_COLLECTOR is unset.
 	if gm.startCollectorsByPriority(gm.resolveLegacyCollectorPriority(caps), caps) == 0 {
 		return nil, fmt.Errorf(noGPUFoundMsg)

agent/gpu_amd_linux.go

@@ -156,6 +156,7 @@ func (gm *GPUManager) updateAmdGpuData(cardPath string) bool {
 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)

agent/gpu_intel_sysfs_linux.go

@@ -0,0 +1,280 @@
+//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))
+}

agent/gpu_intel_sysfs_linux_test.go

@@ -0,0 +1,217 @@
+//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)
+}

agent/gpu_intel_sysfs_unsupported.go

@@ -0,0 +1,13 @@
+//go:build !linux
+
+package agent
+
+type intelSysfsEnergySnapshot struct{}
+
+func (gm *GPUManager) hasIntelSysfs() bool {
+	return false
+}
+
+func (gm *GPUManager) startIntelSysfsCollector() bool {
+	return false
+}

agent/gpu_nvtop.go

@@ -5,6 +5,7 @@ import (
 	"io"
 	"log/slog"
 	"os/exec"
+	"path/filepath"
 	"strconv"
 	"strings"
 	"time"
@@ -48,9 +49,14 @@ func (gm *GPUManager) updateNvtopSnapshots(snapshots []nvtopSnapshot) bool {
 
 	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 == "" {
-			continue
+			if xeName == "" {
+				xeName = xeGpuName()
+			}
+			sample.DeviceName = xeName
 		}
 		indexID := "n" + strconv.Itoa(i)
 		id := indexID
@@ -158,3 +164,38 @@ func (gm *GPUManager) startNvtopCollector(interval string, onFailure func()) {
 		}
 	}()
 }
+
+// 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 + ")"
+}

agent/gpu_test.go

@@ -332,11 +332,12 @@ func TestUpdateNvtopSnapshotsKeepsDeviceAssociationWhenOrderChanges(t *testing.T
 }
 
 func TestParseCollectorPriority(t *testing.T) {
-	got := parseCollectorPriority(" nvml, nvidia-smi, intel_gpu_top, amd_sysfs, nvtop, rocm-smi, bad ")
+	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,
@@ -565,6 +566,42 @@ func TestGetCurrentData(t *testing.T) {
 		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) {
 		gm := &GPUManager{
 			GpuDataMap: map[string]*system.GPUData{
@@ -1461,6 +1498,25 @@ func TestNewGPUManagerConfiguredCollectorsMustStart(t *testing.T) {
 	})
 }
 
+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)

agent/handlers.go

@@ -51,6 +51,7 @@ func NewHandlerRegistry() *HandlerRegistry {
 	registry.Register(common.GetContainerInfo, &GetContainerInfoHandler{})
 	registry.Register(common.GetSmartData, &GetSmartDataHandler{})
 	registry.Register(common.GetSystemdInfo, &GetSystemdInfoHandler{})
+	registry.Register(common.GetZfsData, &GetZfsDataHandler{})
 
 	return registry
 }
@@ -166,14 +167,33 @@ type GetSmartDataHandler struct{}
 
 func (h *GetSmartDataHandler) Handle(hctx *HandlerContext) error {
 	if hctx.Agent.smartManager == nil {
-		// return empty map to indicate no data
-		return hctx.SendResponse(map[string]smart.SmartData{}, hctx.RequestID)
+		return hctx.SendResponse(smart.SmartDataResponse{Data: map[string]smart.SmartData{}}, hctx.RequestID)
 	}
-	if err := hctx.Agent.smartManager.Refresh(false); err != nil {
+	complete, err := hctx.Agent.smartManager.Refresh(false)
+	if err != nil {
 		slog.Debug("smart refresh failed", "err", err)
 	}
-	data := hctx.Agent.smartManager.GetCurrentData()
-	return hctx.SendResponse(data, hctx.RequestID)
+	return hctx.SendResponse(smart.SmartDataResponse{
+		Data:     hctx.Agent.smartManager.GetCurrentData(),
+		Complete: complete,
+	}, hctx.RequestID)
+}
+
+////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////
+
+// GetZfsDataHandler handles ZFS detail data requests
+type GetZfsDataHandler struct{}
+
+func (h *GetZfsDataHandler) Handle(hctx *HandlerContext) error {
+	if hctx.Agent.zfsManager == 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.zfsManager.GetDetail(req.Force), hctx.RequestID)
 }
 
 ////////////////////////////////////////////////////////////////////////////

agent/handlers_test.go

@@ -4,9 +4,12 @@ package agent
 
 import (
 	"testing"
+	"time"
 
 	"github.com/fxamacker/cbor/v2"
+	"github.com/henrygd/beszel/agent/zfs"
 	"github.com/henrygd/beszel/internal/common"
+	"github.com/henrygd/beszel/internal/entities/smart"
 	"github.com/stretchr/testify/assert"
 )
 
@@ -17,6 +20,44 @@ type MockHandler struct {
 	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 := &ZfsManager{detailInterval: time.Hour}
+	zm.poolStatsFn = func() ([]zfs.PoolStat, error) {
+		poolCalls++
+		return []zfs.PoolStat{{Name: "tank", Alloc: uint64(poolCalls)}}, nil
+	}
+	zm.poolStatusesFn = func() ([]zfs.PoolStatus, error) { return nil, nil }
+	zm.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{zfsManager: 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 {
 	if m.handleFunc != nil {
 		return m.handleFunc(ctx)

agent/lhm/beszel_lhm.csproj

@@ -8,6 +8,6 @@
   </PropertyGroup>
 
   <ItemGroup>
-    <PackageReference Include="LibreHardwareMonitorLib" Version="0.9.5" />
+    <PackageReference Include="LibreHardwareMonitorLib" Version="0.9.6" />
   </ItemGroup>
 </Project>

agent/mdraid_linux.go

@@ -17,15 +17,17 @@ import (
 var mdraidSysfsRoot = "/sys"
 
 type mdraidHealth struct {
-	level         string
-	arrayState    string
-	degraded      uint64
-	raidDisks     uint64
-	syncAction    string
-	syncCompleted string
-	syncSpeed     string
-	mismatchCnt   uint64
-	capacity      uint64
+	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.
@@ -92,6 +94,9 @@ func (sm *SmartManager) collectMdraidHealth(deviceInfo *DeviceInfo) (bool, error
 	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})
 	}
@@ -152,6 +157,7 @@ func readMdraidHealth(blockName string) (mdraidHealth, bool) {
 	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
 	}
@@ -177,11 +183,28 @@ func mdraidSmartStatus(health mdraidHealth) string {
 	case "resync", "recover", "reshape":
 		return "WARNING"
 	}
-	if health.degraded > 0 {
+	// 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 "check", "repair":
+	case "repair":
 		return "WARNING"
 	}
 	switch state {
@@ -191,6 +214,43 @@ func mdraidSmartStatus(health mdraidHealth) string {
 	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") {

agent/mdraid_linux_test.go

@@ -40,6 +40,15 @@ func TestMdraidMockSysfsScanAndCollect(t *testing.T) {
 	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")
 
@@ -81,19 +90,93 @@ func TestMdraidMockSysfsScanAndCollect(t *testing.T) {
 	}
 }
 
+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, syncAction: "recover"}); got != "WARNING" {
+	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}); got != "FAILED" {
-		t.Fatalf("mdraidSmartStatus(degraded) = %q, want FAILED", 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)
 	}

agent/response.go

@@ -21,6 +21,9 @@ func newAgentResponse(data any, requestID *uint32) common.AgentResponse {
 		response.String = &v
 	case map[string]smart.SmartData:
 		response.SmartData = v
+	case smart.SmartDataResponse:
+		response.SmartData = v.Data
+		response.SmartComplete = v.Complete
 	case systemd.ServiceDetails:
 		response.ServiceInfo = v
 	default:

agent/sensors.go

@@ -19,13 +19,20 @@ import (
 	"github.com/shirou/gopsutil/v4/sensors"
 )
 
+var errTemperatureFetchTimeout = errors.New("temperature collection timed out")
+
+// Matches sensors.TemperaturesWithContext to allow for panic recovery (gopsutil/issues/1832)
+type getTempsFn func(ctx context.Context) ([]sensors.TemperatureStat, error)
+
 type SensorConfig struct {
 	context        context.Context
 	sensors        map[string]struct{}
 	primarySensor  string
+	timeout        time.Duration
 	isBlacklist    bool
 	hasWildcards   bool
 	skipCollection bool
+	firstRun       bool
 }
 
 func (a *Agent) newSensorConfig() *SensorConfig {
@@ -33,25 +40,29 @@ func (a *Agent) newSensorConfig() *SensorConfig {
 	sysSensors, _ := utils.GetEnv("SYS_SENSORS")
 	sensorsEnvVal, sensorsSet := utils.GetEnv("SENSORS")
 	skipCollection := sensorsSet && sensorsEnvVal == ""
+	sensorsTimeout, _ := utils.GetEnv("SENSORS_TIMEOUT")
 
-	return a.newSensorConfigWithEnv(primarySensor, sysSensors, sensorsEnvVal, skipCollection)
+	return a.newSensorConfigWithEnv(primarySensor, sysSensors, sensorsEnvVal, sensorsTimeout, skipCollection)
 }
 
-// Matches sensors.TemperaturesWithContext to allow for panic recovery (gopsutil/issues/1832)
-type getTempsFn func(ctx context.Context) ([]sensors.TemperatureStat, error)
-
-var (
-	errTemperatureFetchTimeout = errors.New("temperature collection timed out")
-	temperatureFetchTimeout    = 2 * time.Second
-)
-
 // newSensorConfigWithEnv creates a SensorConfig with the provided environment variables
 // sensorsSet indicates if the SENSORS environment variable was explicitly set (even to empty string)
-func (a *Agent) newSensorConfigWithEnv(primarySensor, sysSensors, sensorsEnvVal string, skipCollection bool) *SensorConfig {
+func (a *Agent) newSensorConfigWithEnv(primarySensor, sysSensors, sensorsEnvVal, sensorsTimeout string, skipCollection bool) *SensorConfig {
+	timeout := 2 * time.Second
+	if sensorsTimeout != "" {
+		if d, err := time.ParseDuration(sensorsTimeout); err == nil {
+			timeout = d
+		} else {
+			slog.Warn("Invalid SENSORS_TIMEOUT", "value", sensorsTimeout)
+		}
+	}
+
 	config := &SensorConfig{
 		context:        context.Background(),
 		primarySensor:  primarySensor,
+		timeout:        timeout,
 		skipCollection: skipCollection,
+		firstRun:       true,
 		sensors:        make(map[string]struct{}),
 	}
 
@@ -167,6 +178,14 @@ func (a *Agent) getTempsWithTimeout(getTemps getTempsFn) ([]sensors.TemperatureS
 		err   error
 	}
 
+	// Use a longer timeout on the first run to allow for initialization
+	// (e.g. Windows LHM subprocess startup)
+	timeout := a.sensorConfig.timeout
+	if a.sensorConfig.firstRun {
+		a.sensorConfig.firstRun = false
+		timeout = 10 * time.Second
+	}
+
 	resultCh := make(chan result, 1)
 	go func() {
 		temps, err := a.getTempsWithPanicRecovery(getTemps)
@@ -176,7 +195,7 @@ func (a *Agent) getTempsWithTimeout(getTemps getTempsFn) ([]sensors.TemperatureS
 	select {
 	case res := <-resultCh:
 		return res.temps, res.err
-	case <-time.After(temperatureFetchTimeout):
+	case <-time.After(timeout):
 		return nil, errTemperatureFetchTimeout
 	}
 }

agent/sensors_default.go

@@ -1,4 +1,4 @@
-//go:build !windows
+//go:build !windows && !freebsd
 
 package agent
 

agent/sensors_freebsd.go

@@ -0,0 +1,14 @@
+//go:build freebsd
+
+package agent
+
+import (
+	"context"
+
+	"github.com/shirou/gopsutil/v4/sensors"
+	"golang.org/x/sys/unix"
+)
+
+var getSensorTemps = func(ctx context.Context) ([]sensors.TemperatureStat, error) {
+	return getFreeBSDSensorTemps(ctx, unix.SysctlUint32)
+}

agent/sensors_freebsd_common.go

@@ -0,0 +1,81 @@
+//go:build freebsd || testing
+
+package agent
+
+import (
+	"context"
+	"fmt"
+
+	"github.com/shirou/gopsutil/v4/sensors"
+)
+
+const (
+	freebsdZeroCelsiusDeciKelvin = 2731
+	freebsdAcpiThermalZoneCount  = 16
+)
+
+type freebsdSysctlUintReader func(name string) (uint32, error)
+
+func getFreeBSDSensorTemps(ctx context.Context, readSysctl freebsdSysctlUintReader) ([]sensors.TemperatureStat, error) {
+	cpuCount, err := readSysctl("hw.ncpu")
+	if err != nil {
+		return nil, err
+	}
+	temps := make([]sensors.TemperatureStat, 0, int(cpuCount)+freebsdAcpiThermalZoneCount)
+	for cpu := range cpuCount {
+		select {
+		case <-ctx.Done():
+			return temps, ctx.Err()
+		default:
+		}
+
+		sysctlName := fmt.Sprintf("dev.cpu.%d.temperature", cpu)
+		value, err := readSysctl(sysctlName)
+		if err != nil {
+			continue
+		}
+		temp, ok := freebsdDeciKelvinToCelsius(value)
+		if !ok {
+			continue
+		}
+		temps = append(temps, sensors.TemperatureStat{
+			SensorKey:   fmt.Sprintf("cpu.%d", cpu),
+			Temperature: temp,
+		})
+	}
+
+	for zone := 0; zone < freebsdAcpiThermalZoneCount; zone++ {
+		select {
+		case <-ctx.Done():
+			return temps, ctx.Err()
+		default:
+		}
+
+		sysctlName := fmt.Sprintf("hw.acpi.thermal.tz%d.temperature", zone)
+		value, err := readSysctl(sysctlName)
+		if err != nil {
+			continue
+		}
+		temp, ok := freebsdDeciKelvinToCelsius(value)
+		if !ok {
+			continue
+		}
+		temps = append(temps, sensors.TemperatureStat{
+			SensorKey:   fmt.Sprintf("acpi.thermal.tz%d", zone),
+			Temperature: temp,
+		})
+	}
+
+	return temps, nil
+}
+
+func freebsdDeciKelvinToCelsius(value uint32) (float64, bool) {
+	if value <= freebsdZeroCelsiusDeciKelvin {
+		return 0, false
+	}
+	temp := float64(int64(value)-freebsdZeroCelsiusDeciKelvin) / 10
+	if temp <= 0 || temp >= 200 {
+		return 0, false
+	}
+	return temp, true
+}

agent/sensors_freebsd_common_test.go

@@ -0,0 +1,167 @@
+//go:build testing
+
+package agent
+
+import (
+	"context"
+	"errors"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+var errFakeFreeBSDSysctlNotFound = errors.New("sysctl not found")
+
+type fakeFreeBSDSysctls struct {
+	values map[string]uint32
+	errs   map[string]error
+}
+
+func (f fakeFreeBSDSysctls) read(name string) (uint32, error) {
+	if err, ok := f.errs[name]; ok {
+		return 0, err
+	}
+	if value, ok := f.values[name]; ok {
+		return value, nil
+	}
+	return 0, errFakeFreeBSDSysctlNotFound
+}
+
+func TestFreeBSDDeciKelvinToCelsius(t *testing.T) {
+	tests := []struct {
+		name     string
+		value    uint32
+		expected float64
+		ok       bool
+	}{
+		{
+			name:     "45 Celsius",
+			value:    3181,
+			expected: 45,
+			ok:       true,
+		},
+		{
+			name:     "fractional Celsius",
+			value:    3186,
+			expected: 45.5,
+			ok:       true,
+		},
+		{
+			name:  "zero deci-Kelvin",
+			value: 0,
+			ok:    false,
+		},
+		{
+			name:  "zero Celsius",
+			value: freebsdZeroCelsiusDeciKelvin,
+			ok:    false,
+		},
+		{
+			name:  "below zero Celsius",
+			value: freebsdZeroCelsiusDeciKelvin - 1,
+			ok:    false,
+		},
+		{
+			name:  "invalid signed integer",
+			value: 1<<32 - 1,
+			ok:    false,
+		},
+		{
+			name:  "unreasonably high Celsius",
+			value: freebsdZeroCelsiusDeciKelvin + 2000,
+			ok:    false,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result, ok := freebsdDeciKelvinToCelsius(tt.value)
+			assert.Equal(t, tt.ok, ok)
+			assert.InDelta(t, tt.expected, result, 0.001)
+		})
+	}
+}
+
+func TestGetFreeBSDSensorTemps(t *testing.T) {
+	reader := fakeFreeBSDSysctls{
+		values: map[string]uint32{
+			"hw.ncpu":                         4,
+			"dev.cpu.0.temperature":           3231,
+			"dev.cpu.1.temperature":           3242,
+			"dev.cpu.3.temperature":           freebsdZeroCelsiusDeciKelvin,
+			"hw.acpi.thermal.tz0.temperature": 3101,
+			"hw.acpi.thermal.tz2.temperature": 3116,
+			"hw.acpi.thermal.tz3.temperature": freebsdZeroCelsiusDeciKelvin,
+			"unrelated.sensor.value":          9999,
+			"dev.cpu.99.temperature":          9999,
+			"dev.amdtemp.0.core0.foo":         9999,
+		},
+	}
+
+	temps, err := getFreeBSDSensorTemps(context.Background(), reader.read)
+
+	require.NoError(t, err)
+	require.Len(t, temps, 4)
+	assert.Equal(t, "cpu.0", temps[0].SensorKey)
+	assert.InDelta(t, 50.0, temps[0].Temperature, 0.001)
+	assert.Equal(t, "cpu.1", temps[1].SensorKey)
+	assert.InDelta(t, 51.1, temps[1].Temperature, 0.001)
+	assert.Equal(t, "acpi.thermal.tz0", temps[2].SensorKey)
+	assert.InDelta(t, 37.0, temps[2].Temperature, 0.001)
+	assert.Equal(t, "acpi.thermal.tz2", temps[3].SensorKey)
+	assert.InDelta(t, 38.5, temps[3].Temperature, 0.001)
+}
+
+func TestGetFreeBSDSensorTempsCpuCountError(t *testing.T) {
+	reader := fakeFreeBSDSysctls{
+		errs: map[string]error{
+			"hw.ncpu": errors.New("permission denied"),
+		},
+	}
+
+	temps, err := getFreeBSDSensorTemps(context.Background(), reader.read)
+
+	assert.Nil(t, temps)
+	assert.EqualError(t, err, "permission denied")
+}
+
+func TestGetFreeBSDSensorTempsNoTemperatureSysctls(t *testing.T) {
+	reader := fakeFreeBSDSysctls{
+		values: map[string]uint32{"hw.ncpu": 2},
+	}
+
+	temps, err := getFreeBSDSensorTemps(context.Background(), reader.read)
+
+	require.NoError(t, err)
+	assert.Empty(t, temps)
+}
+
+func TestGetFreeBSDSensorTempsAcpiOnly(t *testing.T) {
+	reader := fakeFreeBSDSysctls{
+		values: map[string]uint32{
+			"hw.ncpu":                         0,
+			"hw.acpi.thermal.tz0.temperature": 3081,
+		},
+	}
+
+	temps, err := getFreeBSDSensorTemps(context.Background(), reader.read)
+
+	require.NoError(t, err)
+	require.Len(t, temps, 1)
+	assert.Equal(t, "acpi.thermal.tz0", temps[0].SensorKey)
+	assert.InDelta(t, 35.0, temps[0].Temperature, 0.001)
+}
+
+func TestGetFreeBSDSensorTempsContextCancelled(t *testing.T) {
+	ctx, cancel := context.WithCancel(context.Background())
+	cancel()
+	reader := fakeFreeBSDSysctls{
+		values: map[string]uint32{"hw.ncpu": 2},
+	}
+
+	temps, err := getFreeBSDSensorTemps(ctx, reader.read)
+
+	assert.Empty(t, temps)
+	assert.ErrorIs(t, err, context.Canceled)
+}

agent/sensors_test.go

@@ -168,6 +168,7 @@ func TestNewSensorConfigWithEnv(t *testing.T) {
 		primarySensor  string
 		sysSensors     string
 		sensors        string
+		sensorsTimeout string
 		skipCollection bool
 		expectedConfig *SensorConfig
 	}{
@@ -179,12 +180,37 @@ func TestNewSensorConfigWithEnv(t *testing.T) {
 			expectedConfig: &SensorConfig{
 				context:        context.Background(),
 				primarySensor:  "",
+				timeout:        2 * time.Second,
 				sensors:        map[string]struct{}{},
 				isBlacklist:    false,
 				hasWildcards:   false,
 				skipCollection: false,
 			},
 		},
+		{
+			name:           "Custom timeout",
+			primarySensor:  "",
+			sysSensors:     "",
+			sensors:        "",
+			sensorsTimeout: "5s",
+			expectedConfig: &SensorConfig{
+				context: context.Background(),
+				timeout: 5 * time.Second,
+				sensors: map[string]struct{}{},
+			},
+		},
+		{
+			name:           "Invalid timeout falls back to default",
+			primarySensor:  "",
+			sysSensors:     "",
+			sensors:        "",
+			sensorsTimeout: "notaduration",
+			expectedConfig: &SensorConfig{
+				context: context.Background(),
+				timeout: 2 * time.Second,
+				sensors: map[string]struct{}{},
+			},
+		},
 		{
 			name:           "Explicitly set to empty string",
 			primarySensor:  "",
@@ -194,6 +220,7 @@ func TestNewSensorConfigWithEnv(t *testing.T) {
 			expectedConfig: &SensorConfig{
 				context:        context.Background(),
 				primarySensor:  "",
+				timeout:        2 * time.Second,
 				sensors:        map[string]struct{}{},
 				isBlacklist:    false,
 				hasWildcards:   false,
@@ -208,6 +235,7 @@ func TestNewSensorConfigWithEnv(t *testing.T) {
 			expectedConfig: &SensorConfig{
 				context:       context.Background(),
 				primarySensor: "cpu_temp",
+				timeout:       2 * time.Second,
 				sensors:       map[string]struct{}{},
 				isBlacklist:   false,
 				hasWildcards:  false,
@@ -221,6 +249,7 @@ func TestNewSensorConfigWithEnv(t *testing.T) {
 			expectedConfig: &SensorConfig{
 				context:       context.Background(),
 				primarySensor: "cpu_temp",
+				timeout:       2 * time.Second,
 				sensors: map[string]struct{}{
 					"cpu_temp": {},
 					"gpu_temp": {},
@@ -237,6 +266,7 @@ func TestNewSensorConfigWithEnv(t *testing.T) {
 			expectedConfig: &SensorConfig{
 				context:       context.Background(),
 				primarySensor: "cpu_temp",
+				timeout:       2 * time.Second,
 				sensors: map[string]struct{}{
 					"cpu_temp": {},
 					"gpu_temp": {},
@@ -253,6 +283,7 @@ func TestNewSensorConfigWithEnv(t *testing.T) {
 			expectedConfig: &SensorConfig{
 				context:       context.Background(),
 				primarySensor: "cpu_temp",
+				timeout:       2 * time.Second,
 				sensors: map[string]struct{}{
 					"cpu_*":    {},
 					"gpu_temp": {},
@@ -269,6 +300,7 @@ func TestNewSensorConfigWithEnv(t *testing.T) {
 			expectedConfig: &SensorConfig{
 				context:       context.Background(),
 				primarySensor: "cpu_temp",
+				timeout:       2 * time.Second,
 				sensors: map[string]struct{}{
 					"cpu_*":    {},
 					"gpu_temp": {},
@@ -284,6 +316,7 @@ func TestNewSensorConfigWithEnv(t *testing.T) {
 			sensors:       "cpu_temp",
 			expectedConfig: &SensorConfig{
 				primarySensor: "cpu_temp",
+				timeout:       2 * time.Second,
 				sensors: map[string]struct{}{
 					"cpu_temp": {},
 				},
@@ -295,7 +328,7 @@ func TestNewSensorConfigWithEnv(t *testing.T) {
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			result := agent.newSensorConfigWithEnv(tt.primarySensor, tt.sysSensors, tt.sensors, tt.skipCollection)
+			result := agent.newSensorConfigWithEnv(tt.primarySensor, tt.sysSensors, tt.sensors, tt.sensorsTimeout, tt.skipCollection)
 
 			// Check primary sensor
 			assert.Equal(t, tt.expectedConfig.primarySensor, result.primarySensor)
@@ -314,6 +347,7 @@ func TestNewSensorConfigWithEnv(t *testing.T) {
 			// Check flags
 			assert.Equal(t, tt.expectedConfig.isBlacklist, result.isBlacklist)
 			assert.Equal(t, tt.expectedConfig.hasWildcards, result.hasWildcards)
+			assert.Equal(t, tt.expectedConfig.timeout, result.timeout)
 
 			// Check context
 			if tt.sysSensors != "" {
@@ -333,12 +367,14 @@ func TestNewSensorConfig(t *testing.T) {
 	t.Setenv("BESZEL_AGENT_PRIMARY_SENSOR", "test_primary")
 	t.Setenv("BESZEL_AGENT_SYS_SENSORS", "/test/path")
 	t.Setenv("BESZEL_AGENT_SENSORS", "test_sensor1,test_*,test_sensor3")
+	t.Setenv("BESZEL_AGENT_SENSORS_TIMEOUT", "7s")
 
 	agent := &Agent{}
 	result := agent.newSensorConfig()
 
 	// Verify results
 	assert.Equal(t, "test_primary", result.primarySensor)
+	assert.Equal(t, 7*time.Second, result.timeout)
 	assert.NotNil(t, result.sensors)
 	assert.Equal(t, 3, len(result.sensors))
 	assert.True(t, result.hasWildcards)
@@ -532,15 +568,10 @@ func TestGetTempsWithTimeout(t *testing.T) {
 	agent := &Agent{
 		sensorConfig: &SensorConfig{
 			context: context.Background(),
+			timeout: 10 * time.Millisecond,
 		},
 	}
 
-	originalTimeout := temperatureFetchTimeout
-	t.Cleanup(func() {
-		temperatureFetchTimeout = originalTimeout
-	})
-	temperatureFetchTimeout = 10 * time.Millisecond
-
 	t.Run("returns temperatures before timeout", func(t *testing.T) {
 		temps, err := agent.getTempsWithTimeout(func(ctx context.Context) ([]sensors.TemperatureStat, error) {
 			return []sensors.TemperatureStat{{SensorKey: "cpu_temp", Temperature: 42}}, nil
@@ -567,15 +598,14 @@ func TestUpdateTemperaturesSkipsOnTimeout(t *testing.T) {
 		systemInfo: system.Info{DashboardTemp: 99},
 		sensorConfig: &SensorConfig{
 			context: context.Background(),
+			timeout: 10 * time.Millisecond,
 		},
 	}
 
-	originalTimeout := temperatureFetchTimeout
+	originalGetSensorTemps := getSensorTemps
 	t.Cleanup(func() {
-		temperatureFetchTimeout = originalTimeout
-		getSensorTemps = sensors.TemperaturesWithContext
+		getSensorTemps = originalGetSensorTemps
 	})
-	temperatureFetchTimeout = 10 * time.Millisecond
 	getSensorTemps = func(ctx context.Context) ([]sensors.TemperatureStat, error) {
 		time.Sleep(50 * time.Millisecond)
 		return nil, nil

agent/sensors_windows.go

@@ -214,9 +214,12 @@ func (lhm *lhmProcess) getTemps(ctx context.Context) (temps []sensors.Temperatur
 	return temps, nil
 }
 
-// getSensorTemps attempts to pull sensor temperatures from the embedded LHM process.
+// getSensorTemps is a variable so tests can replace the platform sensor collector.
+var getSensorTemps = getWindowsSensorTemps
+
+// getWindowsSensorTemps attempts to pull sensor temperatures from the embedded LHM process.
 // NB: LibreHardwareMonitorLib requires admin privileges to access all available sensors.
-func getSensorTemps(ctx context.Context) (temps []sensors.TemperatureStat, err error) {
+func getWindowsSensorTemps(ctx context.Context) (temps []sensors.TemperatureStat, err error) {
 	defer func() {
 		if err != nil {
 			slog.Debug("Error reading sensors", "err", err)

agent/server.go

@@ -29,9 +29,6 @@ type ServerOptions struct {
 	Keys    []gossh.PublicKey // SSH public keys for authentication
 }
 
-// hubVersions caches hub versions by session ID to avoid repeated parsing.
-var hubVersions map[string]semver.Version
-
 // StartServer starts the SSH server with the provided options.
 // It configures the server with secure defaults, sets up authentication,
 // and begins listening for connections. Returns an error if the server
@@ -99,24 +96,15 @@ func (a *Agent) StartServer(opts ServerOptions) error {
 	return a.server.Serve(ln)
 }
 
-// getHubVersion retrieves and caches the hub version for a given session.
-// It extracts the version from the SSH client version string and caches
-// it to avoid repeated parsing. Returns a zero version if parsing fails.
-func (a *Agent) getHubVersion(sessionId string, sessionCtx ssh.Context) semver.Version {
-	if hubVersions == nil {
-		hubVersions = make(map[string]semver.Version, 1)
-	}
-	hubVersion, ok := hubVersions[sessionId]
-	if ok {
-		return hubVersion
-	}
-	// Extract hub version from SSH client version
+// getHubVersion extracts the hub version from the SSH client version string
+// for a given session. Returns a zero version if parsing fails.
+func (a *Agent) getHubVersion(sessionCtx ssh.Context) semver.Version {
 	clientVersion := sessionCtx.Value(ssh.ContextKeyClientVersion)
 	if versionStr, ok := clientVersion.(string); ok {
-		hubVersion, _ = extractHubVersion(versionStr)
+		hubVersion, _ := extractHubVersion(versionStr)
+		return hubVersion
 	}
-	hubVersions[sessionId] = hubVersion
-	return hubVersion
+	return semver.Version{}
 }
 
 // handleSession handles an incoming SSH session by gathering system statistics
@@ -127,9 +115,8 @@ func (a *Agent) handleSession(s ssh.Session) {
 	a.connectionManager.eventChan <- SSHConnect
 
 	sessionCtx := s.Context()
-	sessionID := sessionCtx.SessionID()
 
-	hubVersion := a.getHubVersion(sessionID, sessionCtx)
+	hubVersion := a.getHubVersion(sessionCtx)
 
 	// Legacy one-shot behavior for older hubs
 	if hubVersion.LT(beszel.MinVersionAgentResponse) {
@@ -193,7 +180,7 @@ func (a *Agent) handleSSHRequest(w io.Writer, req *common.HubRequest[cbor.RawMes
 
 // handleLegacyStats serves the legacy one-shot stats payload for older hubs
 func (a *Agent) handleLegacyStats(w io.Writer, hubVersion semver.Version) error {
-	stats := a.gatherStats(common.DataRequestOptions{CacheTimeMs: 60_000})
+	stats := a.gatherStats(common.DataRequestOptions{CacheTimeMs: defaultDataCacheTimeMs})
 	return a.writeToSession(w, stats, hubVersion)
 }
 
@@ -278,6 +265,5 @@ func (a *Agent) StopServer() error {
 	slog.Info("Stopping SSH server")
 	_ = a.server.Close()
 	a.server = nil
-	a.connectionManager.eventChan <- SSHDisconnect
 	return nil
 }

agent/server_test.go

@@ -198,6 +198,28 @@ func TestStartServerDisableSSH(t *testing.T) {
 	assert.Contains(t, err.Error(), "SSH disabled")
 }
 
+func TestStopServerDoesNotBlockWhenEventQueueFull(t *testing.T) {
+	agent := createTestAgent(t)
+	agent.server = &ssh.Server{}
+	agent.connectionManager.eventChan = make(chan ConnectionEvent, 1)
+	agent.connectionManager.eventChan <- WebSocketConnect
+
+	done := make(chan error, 1)
+	go func() {
+		done <- agent.StopServer()
+	}()
+
+	select {
+	case err := <-done:
+		require.NoError(t, err)
+	case <-time.After(time.Second):
+		t.Fatal("StopServer blocked on the connection event queue")
+	}
+
+	assert.Nil(t, agent.server)
+	assert.Equal(t, WebSocketConnect, <-agent.connectionManager.eventChan)
+}
+
 /////////////////////////////////////////////////////////////////
 //////////////////// ParseKeys Tests ////////////////////////////
 /////////////////////////////////////////////////////////////////
@@ -404,27 +426,23 @@ func TestGetHubVersion(t *testing.T) {
 		clientVersion: "SSH-2.0-beszel_0.12.0",
 	}
 
-	// Test first call - should extract and cache version
-	version := agent.getHubVersion("test-session-123", mockCtx)
+	// Test first call - should extract version
+	version := agent.getHubVersion(mockCtx)
 	assert.Equal(t, "0.12.0", version.String())
 
-	// Test second call - should return cached version
-	mockCtx.clientVersion = "SSH-2.0-beszel_0.11.0" // Change version but should still return cached
-	version = agent.getHubVersion("test-session-123", mockCtx)
-	assert.Equal(t, "0.12.0", version.String()) // Should still be cached version
-
-	// Test different session - should extract new version
-	version = agent.getHubVersion("different-session", mockCtx)
+	// Test that version reflects the current client version (no stale caching)
+	mockCtx.clientVersion = "SSH-2.0-beszel_0.11.0"
+	version = agent.getHubVersion(mockCtx)
 	assert.Equal(t, "0.11.0", version.String())
 
 	// Test with invalid version string (non-beszel client)
 	mockCtx.clientVersion = "SSH-2.0-OpenSSH_8.0"
-	version = agent.getHubVersion("invalid-session", mockCtx)
+	version = agent.getHubVersion(mockCtx)
 	assert.Equal(t, "0.0.0", version.String()) // Should be empty version for non-beszel clients
 
 	// Test with no client version
 	mockCtx.clientVersion = ""
-	version = agent.getHubVersion("no-version-session", mockCtx)
+	version = agent.getHubVersion(mockCtx)
 	assert.True(t, version.EQ(semver.Version{})) // Should be empty version
 }
 
@@ -501,9 +519,6 @@ func TestWriteToSessionEncoding(t *testing.T) {
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			// Reset the global hubVersions map to ensure clean state for each test
-			hubVersions = nil
-
 			agent, err := NewAgent("")
 			require.NoError(t, err)
 
@@ -585,39 +600,28 @@ func createTestCombinedData() *system.CombinedData {
 	}
 }
 
-func TestHubVersionCaching(t *testing.T) {
-	// Reset the global hubVersions map to ensure clean state
-	hubVersions = nil
-
+// TestGetHubVersionConcurrent guards against a regression of the
+// "concurrent map writes" panic previously caused by a shared, unsynchronized
+// hubVersions cache (see https://github.com/henrygd/beszel/issues/2128).
+// getHubVersion no longer shares mutable state between sessions, so calling
+// it concurrently from many goroutines must be safe under `go test -race`.
+func TestGetHubVersionConcurrent(t *testing.T) {
 	agent, err := NewAgent("")
 	require.NoError(t, err)
 
-	ctx1 := &mockSSHContext{
-		sessionID:     "session1",
-		clientVersion: "SSH-2.0-beszel_0.12.0",
+	const goroutines = 50
+	var wg sync.WaitGroup
+	wg.Add(goroutines)
+	for i := 0; i < goroutines; i++ {
+		go func(i int) {
+			defer wg.Done()
+			ctx := &mockSSHContext{
+				sessionID:     fmt.Sprintf("session-%d", i),
+				clientVersion: "SSH-2.0-beszel_0.12.0",
+			}
+			version := agent.getHubVersion(ctx)
+			assert.Equal(t, "0.12.0", version.String())
+		}(i)
 	}
-	ctx2 := &mockSSHContext{
-		sessionID:     "session2",
-		clientVersion: "SSH-2.0-beszel_0.11.0",
-	}
-
-	// First calls should cache the versions
-	v1 := agent.getHubVersion("session1", ctx1)
-	v2 := agent.getHubVersion("session2", ctx2)
-
-	assert.Equal(t, "0.12.0", v1.String())
-	assert.Equal(t, "0.11.0", v2.String())
-
-	// Verify caching by changing context but keeping same session ID
-	ctx1.clientVersion = "SSH-2.0-beszel_0.10.0"
-	v1Cached := agent.getHubVersion("session1", ctx1)
-	assert.Equal(t, "0.12.0", v1Cached.String()) // Should still be cached version
-
-	// New session should get new version
-	ctx3 := &mockSSHContext{
-		sessionID:     "session3",
-		clientVersion: "SSH-2.0-beszel_0.13.0",
-	}
-	v3 := agent.getHubVersion("session3", ctx3)
-	assert.Equal(t, "0.13.0", v3.String())
+	wg.Wait()
 }

agent/smart.go

@@ -25,12 +25,15 @@ import (
 // SmartManager manages data collection for SMART devices
 type SmartManager struct {
 	sync.Mutex
-	SmartDataMap    map[string]*smart.SmartData
-	SmartDevices    []*DeviceInfo
-	refreshMutex    sync.Mutex
-	lastScanTime    time.Time
-	smartctlPath    string
-	excludedDevices map[string]struct{}
+	SmartDataMap       map[string]*smart.SmartData
+	SmartDevices       []*DeviceInfo
+	refreshMutex       sync.Mutex
+	lastScanTime       time.Time
+	smartctlPath       string
+	excludedDevices    map[string]struct{}
+	darwinNvmeOnce     sync.Once
+	darwinNvmeCapacity map[string]uint64      // serial → bytes cache, written once via darwinNvmeOnce
+	darwinNvmeProvider func() ([]byte, error) // overridable for testing
 }
 
 type scanOutput struct {
@@ -52,6 +55,11 @@ type DeviceInfo struct {
 	typeVerified bool
 	// parserType holds the parser type (nvme, sat, scsi) that last succeeded.
 	parserType string
+	// explicitType reports whether Type came from an explicit ":type" hint in
+	// SMART_DEVICES. Such a type is a deliberate user override and must always be
+	// passed to smartctl via -d, even for scsi/ata where a scan-detected type is
+	// otherwise left off (see smartctlArgs and issue #1345).
+	explicitType bool
 }
 
 // deviceKey is a composite key for a device, used to identify a device uniquely.
@@ -62,8 +70,9 @@ type deviceKey struct {
 
 var errNoValidSmartData = fmt.Errorf("no valid SMART data found") // Error for missing data
 
-// Refresh updates SMART data for all known devices
-func (sm *SmartManager) Refresh(forceScan bool) error {
+// Refresh updates SMART data for all known devices and reports whether every
+// discovered device was collected successfully.
+func (sm *SmartManager) Refresh(forceScan bool) (bool, error) {
 	sm.refreshMutex.Lock()
 	defer sm.refreshMutex.Unlock()
 
@@ -84,7 +93,7 @@ func (sm *SmartManager) Refresh(forceScan bool) error {
 		}
 	}
 
-	return sm.resolveRefreshError(scanErr, collectErr)
+	return scanErr == nil && collectErr == nil, sm.resolveRefreshError(scanErr, collectErr)
 }
 
 // devicesSnapshot returns a copy of the current device slice to avoid iterating
@@ -248,8 +257,9 @@ func (sm *SmartManager) parseConfiguredDevices(config string) ([]*DeviceInfo, er
 		}
 
 		devices = append(devices, &DeviceInfo{
-			Name: name,
-			Type: devType,
+			Name:         name,
+			Type:         devType,
+			explicitType: devType != "",
 		})
 	}
 
@@ -365,9 +375,15 @@ func (sm *SmartManager) parseSmartOutput(deviceInfo *DeviceInfo, output []byte)
 		Type  string
 		Parse func([]byte) (bool, int)
 	}{
-		{Type: "nvme", Parse: sm.parseSmartForNvme},
-		{Type: "sat", Parse: sm.parseSmartForSata},
-		{Type: "scsi", Parse: sm.parseSmartForScsi},
+		{Type: "nvme", Parse: func(output []byte) (bool, int) {
+			return sm.parseSmartForNvme(output, deviceInfo.Type)
+		}},
+		{Type: "sat", Parse: func(output []byte) (bool, int) {
+			return sm.parseSmartForSata(output, deviceInfo.Type)
+		}},
+		{Type: "scsi", Parse: func(output []byte) (bool, int) {
+			return sm.parseSmartForScsi(output, deviceInfo.Type)
+		}},
 	}
 
 	deviceType := normalizeParserType(deviceInfo.parserType)
@@ -476,10 +492,11 @@ func (sm *SmartManager) CollectSmart(deviceInfo *DeviceInfo) error {
 		return errNoValidSmartData
 	}
 
-	// slog.Info("collecting SMART data", "device", deviceInfo.Name, "type", deviceInfo.Type, "has_existing_data", sm.hasDataForDevice(deviceInfo.Name))
+	// slog.Info("collecting SMART data", "device", deviceInfo.Name, "type", deviceInfo.Type, "has_existing_data", sm.hasDataForDevice(deviceInfo))
 
-	// Check if we have any existing data for this device
-	hasExistingData := sm.hasDataForDevice(deviceInfo.Name)
+	// Check if we have existing data for this exact device identity. Multiple
+	// bridge slots can share a path, so a name-only match is not sufficient.
+	hasExistingData := sm.hasDataForDevice(deviceInfo)
 
 	ctx, cancel := context.WithTimeout(context.Background(), 15*time.Second)
 	defer cancel()
@@ -555,7 +572,9 @@ func (sm *SmartManager) smartctlArgs(deviceInfo *DeviceInfo, includeStandby bool
 		deviceType = strings.ToLower(deviceInfo.Type)
 		parserType = strings.ToLower(deviceInfo.parserType)
 		// types sometimes misidentified in scan; see github.com/henrygd/beszel/issues/1345
-		if deviceType != "" && deviceType != "scsi" && deviceType != "ata" {
+		// An explicit SMART_DEVICES ":type" hint is a deliberate override, so always
+		// pass it through; otherwise scsi/ata are left off so smartctl can auto-detect.
+		if deviceType != "" && (deviceInfo.explicitType || (deviceType != "scsi" && deviceType != "ata")) {
 			args = append(args, "-d", deviceInfo.Type)
 		}
 	}
@@ -580,14 +599,18 @@ func (sm *SmartManager) smartctlArgs(deviceInfo *DeviceInfo, includeStandby bool
 	return args
 }
 
-// hasDataForDevice checks if we have cached SMART data for a specific device
-func (sm *SmartManager) hasDataForDevice(deviceName string) bool {
+// hasDataForDevice checks if we have cached SMART data for a specific device identity.
+func (sm *SmartManager) hasDataForDevice(deviceInfo *DeviceInfo) bool {
+	if deviceInfo == nil {
+		return false
+	}
+
 	sm.Lock()
 	defer sm.Unlock()
 
-	// Check if any cached data has this device name
+	deviceKey := makeDeviceKey(deviceInfo.Name, deviceInfo.Type)
 	for _, data := range sm.SmartDataMap {
-		if data != nil && data.DiskName == deviceName {
+		if data != nil && makeDeviceKey(data.DiskName, data.DiskType) == deviceKey {
 			return true
 		}
 	}
@@ -660,6 +683,9 @@ func mergeDeviceLists(existing, scanned, configured []*DeviceInfo) []*DeviceInfo
 		target.Type = prev.Type
 		target.typeVerified = true
 		target.parserType = prev.parserType
+		if prev.explicitType {
+			target.explicitType = true
+		}
 	}
 
 	// applyConfiguredMetadata updates a matched device with any configured
@@ -673,6 +699,9 @@ func mergeDeviceLists(existing, scanned, configured []*DeviceInfo) []*DeviceInfo
 			existingDev.typeVerified = false
 			existingDev.parserType = normalizeParserType(newType)
 		}
+		if configuredDev.explicitType {
+			existingDev.explicitType = true
+		}
 		if configuredDev.InfoName != "" {
 			existingDev.InfoName = configuredDev.InfoName
 		}
@@ -729,7 +758,14 @@ func mergeDeviceLists(existing, scanned, configured []*DeviceInfo) []*DeviceInfo
 			continue
 		}
 		if existingDev := deviceIndexByName[configuredDevice.Name]; existingDev != nil {
+			oldKey := makeDeviceKey(existingDev.Name, existingDev.Type)
+			if prev := existingIndex[key]; prev != nil {
+				preserveVerifiedType(existingDev, prev)
+			}
 			applyConfiguredMetadata(existingDev, configuredDevice)
+			delete(deviceIndex, oldKey)
+			deviceIndex[makeDeviceKey(existingDev.Name, existingDev.Type)] = existingDev
+			delete(deviceIndexByName, configuredDevice.Name)
 			continue
 		}
 
@@ -833,9 +869,11 @@ func (sm *SmartManager) isVirtualDeviceFromStrings(fields ...string) bool {
 	return false
 }
 
-// parseSmartForSata parses the output of smartctl --all -j for SATA/ATA devices and updates the SmartDataMap
+// parseSmartForSata parses the output of smartctl --all -j for SATA/ATA devices and updates the SmartDataMap.
+// deviceType is the exact type used to identify and query the device; when set,
+// it takes precedence over the generic type reported by smartctl.
 // Returns hasValidData and exitStatus
-func (sm *SmartManager) parseSmartForSata(output []byte) (bool, int) {
+func (sm *SmartManager) parseSmartForSata(output []byte, deviceType string) (bool, int) {
 	var data smart.SmartInfoForSata
 
 	if err := json.Unmarshal(output, &data); err != nil {
@@ -874,6 +912,9 @@ func (sm *SmartManager) parseSmartForSata(output []byte) (bool, int) {
 	smartData.SmartStatus = getSmartStatus(smartData.Temperature, data.SmartStatus.Passed)
 	smartData.DiskName = data.Device.Name
 	smartData.DiskType = data.Device.Type
+	if deviceType != "" {
+		smartData.DiskType = deviceType
+	}
 
 	// get values from ata_device_statistics if necessary
 	var ataDeviceStats smart.AtaDeviceStatistics
@@ -890,6 +931,9 @@ func (sm *SmartManager) parseSmartForSata(output []byte) (bool, int) {
 		if parsed, ok := smart.ParseSmartRawValueString(attr.Raw.String); ok {
 			rawValue = parsed
 		}
+		if smartData.SmartStatus == "PASSED" && rawValue > 0 && (attr.ID == 5 || attr.ID == 197 || attr.ID == 198) {
+			smartData.SmartStatus = "WARNING"
+		}
 		smartAttr := &smart.SmartAttribute{
 			ID:         attr.ID,
 			Name:       attr.Name,
@@ -947,7 +991,7 @@ func findAtaDeviceStatisticsValue(data *smart.SmartInfoForSata, ataDeviceStats *
 	return nil
 }
 
-func (sm *SmartManager) parseSmartForScsi(output []byte) (bool, int) {
+func (sm *SmartManager) parseSmartForScsi(output []byte, deviceType string) (bool, int) {
 	var data smart.SmartInfoForScsi
 
 	if err := json.Unmarshal(output, &data); err != nil {
@@ -982,6 +1026,9 @@ func (sm *SmartManager) parseSmartForScsi(output []byte) (bool, int) {
 	smartData.SmartStatus = getSmartStatus(smartData.Temperature, data.SmartStatus.Passed)
 	smartData.DiskName = data.Device.Name
 	smartData.DiskType = data.Device.Type
+	if deviceType != "" {
+		smartData.DiskType = deviceType
+	}
 
 	attributes := make([]*smart.SmartAttribute, 0, 10)
 	attributes = append(attributes, &smart.SmartAttribute{Name: "PowerOnHours", RawValue: data.PowerOnTime.Hours})
@@ -1033,9 +1080,57 @@ func parseScsiGigabytesProcessed(value string) int64 {
 	return parsed
 }
 
-// parseSmartForNvme parses the output of smartctl --all -j /dev/nvmeX and updates the SmartDataMap
+// lookupDarwinNvmeCapacity returns the capacity in bytes for a given NVMe serial number on Darwin.
+// It uses system_profiler SPNVMeDataType to get capacity since Apple SSDs don't report user_capacity
+// via smartctl. Results are cached after the first call via sync.Once.
+func (sm *SmartManager) lookupDarwinNvmeCapacity(serial string) uint64 {
+	sm.darwinNvmeOnce.Do(func() {
+		sm.darwinNvmeCapacity = make(map[string]uint64)
+
+		provider := sm.darwinNvmeProvider
+		if provider == nil {
+			provider = func() ([]byte, error) {
+				ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+				defer cancel()
+				return exec.CommandContext(ctx, "system_profiler", "SPNVMeDataType", "-json").Output()
+			}
+		}
+
+		out, err := provider()
+		if err != nil {
+			slog.Debug("system_profiler NVMe lookup failed", "err", err)
+			return
+		}
+
+		var result struct {
+			SPNVMeDataType []struct {
+				Items []struct {
+					DeviceSerial string `json:"device_serial"`
+					SizeInBytes  uint64 `json:"size_in_bytes"`
+				} `json:"_items"`
+			} `json:"SPNVMeDataType"`
+		}
+		if err := json.Unmarshal(out, &result); err != nil {
+			slog.Debug("system_profiler NVMe parse failed", "err", err)
+			return
+		}
+
+		for _, controller := range result.SPNVMeDataType {
+			for _, item := range controller.Items {
+				if item.DeviceSerial != "" && item.SizeInBytes > 0 {
+					sm.darwinNvmeCapacity[item.DeviceSerial] = item.SizeInBytes
+				}
+			}
+		}
+	})
+	return sm.darwinNvmeCapacity[serial]
+}
+
+// parseSmartForNvme parses the output of smartctl --all -j /dev/nvmeX and updates the SmartDataMap.
+// deviceType is the exact type used to identify and query the device; when set,
+// it takes precedence over the generic type reported by smartctl.
 // Returns hasValidData and exitStatus
-func (sm *SmartManager) parseSmartForNvme(output []byte) (bool, int) {
+func (sm *SmartManager) parseSmartForNvme(output []byte, deviceType string) (bool, int) {
 	data := &smart.SmartInfoForNvme{}
 
 	if err := json.Unmarshal(output, &data); err != nil {
@@ -1069,10 +1164,19 @@ func (sm *SmartManager) parseSmartForNvme(output []byte) (bool, int) {
 	smartData.SerialNumber = data.SerialNumber
 	smartData.FirmwareVersion = data.FirmwareVersion
 	smartData.Capacity = data.UserCapacity.Bytes
+	if smartData.Capacity == 0 {
+		smartData.Capacity = data.NVMeTotalCapacity
+	}
+	if smartData.Capacity == 0 && (runtime.GOOS == "darwin" || sm.darwinNvmeProvider != nil) {
+		smartData.Capacity = sm.lookupDarwinNvmeCapacity(data.SerialNumber)
+	}
 	smartData.Temperature = data.NVMeSmartHealthInformationLog.Temperature
 	smartData.SmartStatus = getSmartStatus(smartData.Temperature, data.SmartStatus.Passed)
 	smartData.DiskName = data.Device.Name
 	smartData.DiskType = data.Device.Type
+	if deviceType != "" {
+		smartData.DiskType = deviceType
+	}
 
 	// nvme attributes does not follow the same format as ata attributes,
 	// so we manually map each field to SmartAttributes

agent/smart_test.go

@@ -4,8 +4,10 @@ package agent
 
 import (
 	"errors"
+	"fmt"
 	"os"
 	"path/filepath"
+	"strconv"
 	"testing"
 
 	"github.com/henrygd/beszel/internal/entities/smart"
@@ -24,7 +26,7 @@ func TestParseSmartForScsi(t *testing.T) {
 		SmartDataMap: make(map[string]*smart.SmartData),
 	}
 
-	hasData, exitStatus := sm.parseSmartForScsi(data)
+	hasData, exitStatus := sm.parseSmartForScsi(data, "")
 	if !hasData {
 		t.Fatalf("expected SCSI data to parse successfully")
 	}
@@ -69,7 +71,7 @@ func TestParseSmartForSata(t *testing.T) {
 		SmartDataMap: make(map[string]*smart.SmartData),
 	}
 
-	hasData, exitStatus := sm.parseSmartForSata(data)
+	hasData, exitStatus := sm.parseSmartForSata(data, "")
 	require.True(t, hasData)
 	assert.Equal(t, 64, exitStatus)
 
@@ -88,6 +90,52 @@ func TestParseSmartForSata(t *testing.T) {
 	}
 }
 
+func TestParseSmartForSataWarnsForCriticalAttributes(t *testing.T) {
+	for _, attrID := range []int{5, 197, 198} {
+		t.Run("attribute "+strconv.Itoa(attrID), func(t *testing.T) {
+			jsonPayload := []byte(fmt.Sprintf(`{
+				"smartctl": {"exit_status": 0},
+				"device": {"name": "/dev/sda", "type": "sat"},
+				"model_name": "Example",
+				"serial_number": "WARNING%d",
+				"smart_status": {"passed": true},
+				"temperature": {"current": 30},
+				"ata_smart_attributes": {"table": [{"id": %d, "raw": {"value": 1, "string": "1"}}]}
+			}`, attrID, attrID))
+
+			sm := &SmartManager{SmartDataMap: make(map[string]*smart.SmartData)}
+			hasData, _ := sm.parseSmartForSata(jsonPayload, "")
+			require.True(t, hasData)
+			assert.Equal(t, "WARNING", sm.SmartDataMap[fmt.Sprintf("WARNING%d", attrID)].SmartStatus)
+		})
+	}
+}
+
+func TestParseSmartForSataPreservesFailedAndUnknownStatus(t *testing.T) {
+	for _, test := range []struct {
+		name        string
+		temperature int
+		want        string
+	}{
+		{name: "failed", temperature: 30, want: "FAILED"},
+		{name: "unknown", want: "UNKNOWN"},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			jsonPayload := []byte(fmt.Sprintf(`{
+				"device": {"name": "/dev/sda", "type": "sat"},
+				"serial_number": "PRESERVE%s",
+				"temperature": {"current": %d},
+				"ata_smart_attributes": {"table": [{"id": 197, "raw": {"value": 1, "string": "1"}}]}
+			}`, test.name, test.temperature))
+
+			sm := &SmartManager{SmartDataMap: make(map[string]*smart.SmartData)}
+			hasData, _ := sm.parseSmartForSata(jsonPayload, "")
+			require.True(t, hasData)
+			assert.Equal(t, test.want, sm.SmartDataMap["PRESERVE"+test.name].SmartStatus)
+		})
+	}
+}
+
 func TestParseSmartForSataDeviceStatisticsTemperature(t *testing.T) {
 	jsonPayload := []byte(`{
 		"smartctl": {"exit_status": 0},
@@ -112,7 +160,7 @@ func TestParseSmartForSataDeviceStatisticsTemperature(t *testing.T) {
 	}`)
 
 	sm := &SmartManager{SmartDataMap: make(map[string]*smart.SmartData)}
-	hasData, exitStatus := sm.parseSmartForSata(jsonPayload)
+	hasData, exitStatus := sm.parseSmartForSata(jsonPayload, "")
 	require.True(t, hasData)
 	assert.Equal(t, 0, exitStatus)
 
@@ -147,7 +195,7 @@ func TestParseSmartForSataAtaDeviceStatistics(t *testing.T) {
 	}`)
 
 	sm := &SmartManager{SmartDataMap: make(map[string]*smart.SmartData)}
-	hasData, exitStatus := sm.parseSmartForSata(jsonPayload)
+	hasData, exitStatus := sm.parseSmartForSata(jsonPayload, "")
 	require.True(t, hasData)
 	assert.Equal(t, 0, exitStatus)
 
@@ -184,7 +232,7 @@ func TestParseSmartForSataNegativeDeviceStatistics(t *testing.T) {
 	}`)
 
 	sm := &SmartManager{SmartDataMap: make(map[string]*smart.SmartData)}
-	hasData, exitStatus := sm.parseSmartForSata(jsonPayload)
+	hasData, exitStatus := sm.parseSmartForSata(jsonPayload, "")
 	require.True(t, hasData)
 	assert.Equal(t, 0, exitStatus)
 
@@ -223,7 +271,7 @@ func TestParseSmartForSataParentheticalRawValue(t *testing.T) {
 
 	sm := &SmartManager{SmartDataMap: make(map[string]*smart.SmartData)}
 
-	hasData, exitStatus := sm.parseSmartForSata(jsonPayload)
+	hasData, exitStatus := sm.parseSmartForSata(jsonPayload, "")
 	require.True(t, hasData)
 	assert.Equal(t, 0, exitStatus)
 
@@ -245,7 +293,7 @@ func TestParseSmartForNvme(t *testing.T) {
 		SmartDataMap: make(map[string]*smart.SmartData),
 	}
 
-	hasData, exitStatus := sm.parseSmartForNvme(data)
+	hasData, exitStatus := sm.parseSmartForNvme(data, "")
 	require.True(t, hasData)
 	assert.Equal(t, 0, exitStatus)
 
@@ -268,13 +316,15 @@ func TestParseSmartForNvme(t *testing.T) {
 func TestHasDataForDevice(t *testing.T) {
 	sm := &SmartManager{
 		SmartDataMap: map[string]*smart.SmartData{
-			"serial-1": {DiskName: "/dev/sda"},
+			"serial-1": {DiskName: "/dev/sda", DiskType: "jms56x,0"},
 			"serial-2": nil,
 		},
 	}
 
-	assert.True(t, sm.hasDataForDevice("/dev/sda"))
-	assert.False(t, sm.hasDataForDevice("/dev/sdb"))
+	assert.True(t, sm.hasDataForDevice(&DeviceInfo{Name: "/dev/sda", Type: "jms56x,0"}))
+	assert.False(t, sm.hasDataForDevice(&DeviceInfo{Name: "/dev/sda", Type: "jms56x,1"}))
+	assert.False(t, sm.hasDataForDevice(&DeviceInfo{Name: "/dev/sdb", Type: "jms56x,0"}))
+	assert.False(t, sm.hasDataForDevice(nil))
 }
 
 func TestDevicesSnapshotReturnsCopy(t *testing.T) {
@@ -392,6 +442,81 @@ func TestSmartctlArgs(t *testing.T) {
 	)
 }
 
+// TestSmartctlArgsExplicitType verifies that an explicit SMART_DEVICES type hint
+// is always passed to smartctl via -d, while a scan-detected scsi/ata type is
+// still left off so smartctl can auto-detect it (see issue #1345).
+func TestSmartctlArgsExplicitType(t *testing.T) {
+	sm := &SmartManager{}
+
+	// Scan-detected scsi: -d is intentionally omitted.
+	scanScsi := &DeviceInfo{Name: "/dev/sda", Type: "scsi"}
+	assert.Equal(t,
+		[]string{"-a", "--json=c", "/dev/sda"},
+		sm.smartctlArgs(scanScsi, false),
+	)
+
+	// Explicit scsi from SMART_DEVICES: -d scsi must be passed.
+	explicitScsi := &DeviceInfo{Name: "/dev/sda", Type: "scsi", explicitType: true}
+	assert.Equal(t,
+		[]string{"-d", "scsi", "-a", "--json=c", "/dev/sda"},
+		sm.smartctlArgs(explicitScsi, false),
+	)
+
+	// Explicit ata from SMART_DEVICES: -d ata must be passed (devstat still added).
+	explicitAta := &DeviceInfo{Name: "/dev/sdb", Type: "ata", explicitType: true}
+	assert.Equal(t,
+		[]string{"-d", "ata", "-a", "--json=c", "-l", "devstat", "/dev/sdb"},
+		sm.smartctlArgs(explicitAta, false),
+	)
+}
+
+// TestSmartDevicesExplicitTypeFlowsToSmartctlArgs is a regression test for
+// issue #2072: an explicit SMART_DEVICES type (e.g. /dev/sda:scsi) must win over
+// a wrong scan-detected type (sat) and be handed to smartctl as -d scsi.
+func TestSmartDevicesExplicitTypeFlowsToSmartctlArgs(t *testing.T) {
+	sm := &SmartManager{}
+
+	configured, err := sm.parseConfiguredDevices("/dev/sda:scsi")
+	require.NoError(t, err)
+	require.Len(t, configured, 1)
+	assert.True(t, configured[0].explicitType)
+
+	// smartctl --scan misreports this USB drive as sat, which fails on it.
+	scanned := []*DeviceInfo{
+		{Name: "/dev/sda", Type: "sat", Protocol: "ATA"},
+	}
+
+	merged := mergeDeviceLists(nil, scanned, configured)
+	require.Len(t, merged, 1)
+
+	device := merged[0]
+	assert.Equal(t, "scsi", device.Type, "configured type should win over scan-detected sat")
+	assert.True(t, device.explicitType, "explicit hint must survive the merge")
+
+	assert.Equal(t,
+		[]string{"-d", "scsi", "-a", "--json=c", "/dev/sda"},
+		sm.smartctlArgs(device, false),
+		"explicit scsi type must be passed to smartctl, not dropped",
+	)
+}
+
+// TestMergeDeviceListsPreservesExplicitTypeAcrossRescan ensures a verified,
+// explicitly-typed device keeps its explicit flag when a later scan re-reports
+// it with a different auto-detected type.
+func TestMergeDeviceListsPreservesExplicitTypeAcrossRescan(t *testing.T) {
+	existing := []*DeviceInfo{
+		{Name: "/dev/sda", Type: "scsi", parserType: "scsi", typeVerified: true, explicitType: true},
+	}
+	scanned := []*DeviceInfo{
+		{Name: "/dev/sda", Type: "sat"},
+	}
+
+	merged := mergeDeviceLists(existing, scanned, nil)
+	require.Len(t, merged, 1)
+	assert.Equal(t, "scsi", merged[0].Type)
+	assert.True(t, merged[0].explicitType, "explicit type flag should survive a rescan")
+}
+
 func TestResolveRefreshError(t *testing.T) {
 	scanErr := errors.New("scan failed")
 	collectErr := errors.New("collect failed")
@@ -534,6 +659,74 @@ func TestMergeDeviceListsPrefersConfigured(t *testing.T) {
 	assert.Equal(t, "sat", byName["/dev/sdb"].Type)
 }
 
+func TestMergeDeviceListsExpandsConfiguredDevicesWithSamePath(t *testing.T) {
+	scanned := []*DeviceInfo{
+		{Name: "/dev/sdb", Type: "sat", InfoName: "scan-info", Protocol: "ATA"},
+	}
+	configured := []*DeviceInfo{
+		{Name: "/dev/sdb", Type: "jms56x,0", explicitType: true},
+		{Name: "/dev/sdb", Type: "jms56x,1", explicitType: true},
+	}
+
+	merged := mergeDeviceLists(nil, scanned, configured)
+	require.Len(t, merged, 2)
+
+	byKey := make(map[deviceKey]*DeviceInfo, len(merged))
+	for _, device := range merged {
+		byKey[makeDeviceKey(device.Name, device.Type)] = device
+	}
+
+	first := byKey[makeDeviceKey("/dev/sdb", "jms56x,0")]
+	require.NotNil(t, first)
+	assert.Equal(t, "scan-info", first.InfoName)
+	assert.Equal(t, "ATA", first.Protocol)
+	assert.True(t, first.explicitType)
+
+	second := byKey[makeDeviceKey("/dev/sdb", "jms56x,1")]
+	require.NotNil(t, second)
+	assert.True(t, second.explicitType)
+	assert.NotContains(t, byKey, makeDeviceKey("/dev/sdb", "sat"))
+}
+
+func TestMergeDeviceListsPreservesSamePathVerificationAcrossRescan(t *testing.T) {
+	existing := []*DeviceInfo{
+		{Name: "/dev/sdb", Type: "jms56x,0", parserType: "sat", typeVerified: true, explicitType: true},
+		{Name: "/dev/sdb", Type: "jms56x,1", parserType: "sat", typeVerified: true, explicitType: true},
+	}
+	scanned := []*DeviceInfo{
+		{Name: "/dev/sdb", Type: "sat", Protocol: "ATA"},
+	}
+	configured := []*DeviceInfo{
+		{Name: "/dev/sdb", Type: "jms56x,0", explicitType: true},
+		{Name: "/dev/sdb", Type: "jms56x,1", explicitType: true},
+	}
+
+	merged := mergeDeviceLists(existing, scanned, configured)
+	require.Len(t, merged, 2)
+	byKey := make(map[deviceKey]*DeviceInfo, len(merged))
+	for _, device := range merged {
+		byKey[makeDeviceKey(device.Name, device.Type)] = device
+		assert.True(t, device.typeVerified, device.Type)
+		assert.Equal(t, "sat", device.parserType, device.Type)
+		assert.True(t, device.explicitType, device.Type)
+	}
+	assert.Contains(t, byKey, makeDeviceKey("/dev/sdb", "jms56x,0"))
+	assert.Contains(t, byKey, makeDeviceKey("/dev/sdb", "jms56x,1"))
+}
+
+func TestMergeDeviceListsDeduplicatesConfiguredIdentityAfterRekey(t *testing.T) {
+	scanned := []*DeviceInfo{{Name: "/dev/sdb", Type: "sat"}}
+	configured := []*DeviceInfo{
+		{Name: "/dev/sdb", Type: "jms56x,0", explicitType: true},
+		{Name: "/dev/sdb", Type: "jms56x,0", explicitType: true},
+	}
+
+	merged := mergeDeviceLists(nil, scanned, configured)
+	require.Len(t, merged, 1)
+	assert.Equal(t, "/dev/sdb", merged[0].Name)
+	assert.Equal(t, "jms56x,0", merged[0].Type)
+}
+
 func TestMergeDeviceListsPreservesVerification(t *testing.T) {
 	existing := []*DeviceInfo{
 		{Name: "/dev/sda", Type: "sat+megaraid", parserType: "sat", typeVerified: true},
@@ -678,6 +871,20 @@ func TestParseSmartOutputKeepsCustomType(t *testing.T) {
 	assert.Equal(t, "sat+megaraid", device.Type)
 	assert.Equal(t, "sat", device.parserType)
 	assert.True(t, device.typeVerified)
+	assert.Equal(t, "sat+megaraid", sm.SmartDataMap["9C40918040082"].DiskType)
+}
+
+func TestParseSmartOutputDoesNotNormalizeDeviceIdentity(t *testing.T) {
+	fixturePath := filepath.Join("test-data", "smart", "sda.json")
+	data, err := os.ReadFile(fixturePath)
+	require.NoError(t, err)
+
+	sm := &SmartManager{SmartDataMap: make(map[string]*smart.SmartData)}
+	device := &DeviceInfo{Name: "/dev/sda", Type: "ata", explicitType: true}
+
+	require.True(t, sm.parseSmartOutput(device, data))
+	assert.Equal(t, "sat", device.parserType)
+	assert.Equal(t, "ata", sm.SmartDataMap["9C40918040082"].DiskType)
 }
 
 func TestParseSmartOutputResetsVerificationOnFailure(t *testing.T) {
@@ -1199,3 +1406,81 @@ func TestIsNvmeControllerPath(t *testing.T) {
 		})
 	}
 }
+
+func TestParseSmartForNvmeAppleSSD(t *testing.T) {
+	// Apple SSDs don't report user_capacity via smartctl; capacity should be fetched
+	// from system_profiler via the darwinNvmeProvider fallback.
+	fixturePath := filepath.Join("test-data", "smart", "apple_nvme.json")
+	data, err := os.ReadFile(fixturePath)
+	require.NoError(t, err)
+
+	providerCalls := 0
+	fakeProvider := func() ([]byte, error) {
+		providerCalls++
+		return []byte(`{
+			"SPNVMeDataType": [{
+				"_items": [{
+					"device_serial": "0ba0147940253c15",
+					"size_in_bytes": 251000193024
+				}]
+			}]
+		}`), nil
+	}
+
+	sm := &SmartManager{
+		SmartDataMap:       make(map[string]*smart.SmartData),
+		darwinNvmeProvider: fakeProvider,
+	}
+
+	hasData, _ := sm.parseSmartForNvme(data, "")
+	require.True(t, hasData)
+
+	deviceData, ok := sm.SmartDataMap["0ba0147940253c15"]
+	require.True(t, ok)
+	assert.Equal(t, "APPLE SSD AP0256Q", deviceData.ModelName)
+	assert.Equal(t, uint64(251000193024), deviceData.Capacity)
+	assert.Equal(t, uint8(42), deviceData.Temperature)
+	assert.Equal(t, "PASSED", deviceData.SmartStatus)
+	assert.Equal(t, 1, providerCalls, "system_profiler should be called once")
+
+	// Second parse: provider should NOT be called again (cache hit)
+	_, _ = sm.parseSmartForNvme(data, "")
+	assert.Equal(t, 1, providerCalls, "system_profiler should not be called again after caching")
+}
+
+func TestLookupDarwinNvmeCapacityMultipleDisks(t *testing.T) {
+	fakeProvider := func() ([]byte, error) {
+		return []byte(`{
+			"SPNVMeDataType": [
+				{
+					"_items": [
+						{"device_serial": "serial-disk0", "size_in_bytes": 251000193024},
+						{"device_serial": "serial-disk1", "size_in_bytes": 1000204886016}
+					]
+				},
+				{
+					"_items": [
+						{"device_serial": "serial-disk2", "size_in_bytes": 512110190592}
+					]
+				}
+			]
+		}`), nil
+	}
+
+	sm := &SmartManager{darwinNvmeProvider: fakeProvider}
+	assert.Equal(t, uint64(251000193024), sm.lookupDarwinNvmeCapacity("serial-disk0"))
+	assert.Equal(t, uint64(1000204886016), sm.lookupDarwinNvmeCapacity("serial-disk1"))
+	assert.Equal(t, uint64(512110190592), sm.lookupDarwinNvmeCapacity("serial-disk2"))
+	assert.Equal(t, uint64(0), sm.lookupDarwinNvmeCapacity("unknown-serial"))
+}
+
+func TestLookupDarwinNvmeCapacityProviderError(t *testing.T) {
+	fakeProvider := func() ([]byte, error) {
+		return nil, errors.New("system_profiler not found")
+	}
+
+	sm := &SmartManager{darwinNvmeProvider: fakeProvider}
+	assert.Equal(t, uint64(0), sm.lookupDarwinNvmeCapacity("any-serial"))
+	// Cache should be initialized even on error so we don't retry (Once already fired)
+	assert.NotNil(t, sm.darwinNvmeCapacity)
+}

agent/system.go

@@ -4,11 +4,11 @@ import (
 	"bufio"
 	"errors"
 	"fmt"
+	"io"
 	"log/slog"
 	"os"
 	"runtime"
 	"strings"
-	"time"
 
 	"github.com/henrygd/beszel"
 	"github.com/henrygd/beszel/agent/battery"
@@ -23,13 +23,6 @@ import (
 	"github.com/shirou/gopsutil/v4/mem"
 )
 
-// prevDisk stores previous per-device disk counters for a given cache interval
-type prevDisk struct {
-	readBytes  uint64
-	writeBytes uint64
-	at         time.Time
-}
-
 // Sets initial / non-changing values about the host system
 func (a *Agent) refreshSystemDetails() {
 	a.systemInfo.AgentVersion = beszel.Version
@@ -39,7 +32,11 @@ func (a *Agent) refreshSystemDetails() {
 
 	if a.dockerManager != nil {
 		a.systemDetails.Podman = a.dockerManager.IsPodman()
-		hostInfo, _ = a.dockerManager.GetHostInfo()
+		// Docker's host info describes the machine its daemon runs on. On macOS and
+		// Windows that is a Linux VM, so its CPU and memory totals are not this host's.
+		if runtime.GOOS != "darwin" && runtime.GOOS != "windows" {
+			hostInfo, _ = a.dockerManager.GetHostInfo()
+		}
 	}
 
 	a.systemDetails.Hostname, _ = os.Hostname()
@@ -86,6 +83,12 @@ func (a *Agent) refreshSystemDetails() {
 	if info, err := cpu.Info(); err == nil && len(info) > 0 {
 		a.systemDetails.CpuModel = info[0].ModelName
 	}
+	// gopsutil doesn't parse the "cpu model" field from /proc/cpuinfo, which
+	// is the only source of the CPU model name on MIPS. Fall back to reading
+	// it directly when ModelName is empty.
+	if a.systemDetails.CpuModel == "" {
+		a.systemDetails.CpuModel = getCpuModelFromCpuinfo()
+	}
 	// cores / threads
 	cores, _ := cpu.Counts(false)
 	threads := hostInfo.NCPU
@@ -115,14 +118,39 @@ func (a *Agent) refreshSystemDetails() {
 	}
 }
 
+// attachSystemDetails returns details only for fresh default-interval responses.
+func (a *Agent) attachSystemDetails(data *system.CombinedData, cacheTimeMs uint16, includeRequested bool) *system.CombinedData {
+	if cacheTimeMs != defaultDataCacheTimeMs || (!includeRequested && !a.detailsDirty) {
+		return data
+	}
+
+	// copy data to avoid adding details to the original cached struct
+	response := *data
+	response.Details = &a.systemDetails
+	a.detailsDirty = false
+	return &response
+}
+
+// updateSystemDetails applies a mutation to the static details payload and marks
+// it for inclusion on the next fresh default-interval response.
+func (a *Agent) updateSystemDetails(updateFunc func(details *system.Details)) {
+	updateFunc(&a.systemDetails)
+	a.detailsDirty = true
+}
+
 // Returns current info, stats about the host system
 func (a *Agent) getSystemStats(cacheTimeMs uint16) system.Stats {
 	var systemStats system.Stats
 
 	// battery
-	if batteryPercent, batteryState, err := battery.GetBatteryStats(); err == nil {
-		systemStats.Battery[0] = batteryPercent
-		systemStats.Battery[1] = batteryState
+	if batteries, err := battery.GetBatteryStats(); err == nil {
+		systemStats.Batteries = make(map[string]uint8, len(batteries))
+		for _, device := range batteries {
+			systemStats.Batteries[device.Name] = device.Percent
+		}
+		if primary, ok := battery.Primary(batteries); ok {
+			systemStats.Battery = [2]uint8{primary.Percent, primary.State}
+		}
 	}
 
 	// cpu metrics
@@ -147,9 +175,9 @@ func (a *Agent) getSystemStats(cacheTimeMs uint16) system.Stats {
 
 	// load average
 	if avgstat, err := load.Avg(); err == nil {
-		systemStats.LoadAvg[0] = avgstat.Load1
-		systemStats.LoadAvg[1] = avgstat.Load5
-		systemStats.LoadAvg[2] = avgstat.Load15
+		systemStats.LoadAvg[0] = utils.TwoDecimals(avgstat.Load1)
+		systemStats.LoadAvg[1] = utils.TwoDecimals(avgstat.Load5)
+		systemStats.LoadAvg[2] = utils.TwoDecimals(avgstat.Load15)
 		slog.Debug("Load average", "5m", avgstat.Load5, "15m", avgstat.Load15)
 	} else {
 		slog.Error("Error getting load average", "err", err)
@@ -157,21 +185,11 @@ func (a *Agent) getSystemStats(cacheTimeMs uint16) system.Stats {
 
 	// memory
 	if v, err := mem.VirtualMemory(); err == nil {
+		used, cacheBuff, swapUsed := calculateHostMemoryUsage(v, a.memCalc == "htop")
 		// swap
 		systemStats.Swap = utils.BytesToGigabytes(v.SwapTotal)
-		systemStats.SwapUsed = utils.BytesToGigabytes(v.SwapTotal - v.SwapFree - v.SwapCached)
-		// cache + buffers value for default mem calculation
-		// note: gopsutil automatically adds SReclaimable to v.Cached
-		cacheBuff := v.Cached + v.Buffers - v.Shared
-		if cacheBuff <= 0 {
-			cacheBuff = max(v.Total-v.Free-v.Used, 0)
-		}
-		// htop memory calculation overrides (likely outdated as of mid 2025)
-		if a.memCalc == "htop" {
-			// cacheBuff = v.Cached + v.Buffers - v.Shared
-			v.Used = v.Total - (v.Free + cacheBuff)
-			v.UsedPercent = float64(v.Used) / float64(v.Total) * 100.0
-		}
+		systemStats.SwapUsed = utils.BytesToGigabytes(swapUsed)
+		v.Used = used
 		// if a.memCalc == "legacy" {
 		// 	v.Used = v.Total - v.Free - v.Buffers - v.Cached
 		// 	cacheBuff = v.Total - v.Free - v.Used
@@ -181,10 +199,14 @@ func (a *Agent) getSystemStats(cacheTimeMs uint16) system.Stats {
 		if a.zfs {
 			if arcSize, _ := zfs.ARCSize(); arcSize > 0 && arcSize < v.Used {
 				v.Used = v.Used - arcSize
-				v.UsedPercent = float64(v.Used) / float64(v.Total) * 100.0
 				systemStats.MemZfsArc = utils.BytesToGigabytes(arcSize)
 			}
 		}
+		if v.Total > 0 {
+			v.UsedPercent = float64(v.Used) / float64(v.Total) * 100.0
+		} else {
+			v.UsedPercent = 0
+		}
 		systemStats.Mem = utils.BytesToGigabytes(v.Total)
 		systemStats.MemBuffCache = utils.BytesToGigabytes(cacheBuff)
 		systemStats.MemUsed = utils.BytesToGigabytes(v.Used)
@@ -197,6 +219,9 @@ func (a *Agent) getSystemStats(cacheTimeMs uint16) system.Stats {
 	// disk i/o (cache-aware per interval)
 	a.updateDiskIo(cacheTimeMs, &systemStats)
 
+	// zfs pool stats
+	a.zfsManager.Update(&systemStats)
+
 	// network stats (per cache interval)
 	a.updateNetworkStats(cacheTimeMs, &systemStats)
 
@@ -204,6 +229,9 @@ func (a *Agent) getSystemStats(cacheTimeMs uint16) system.Stats {
 	// TODO: maybe refactor to methods on systemStats
 	a.updateTemperatures(&systemStats)
 
+	// fan speeds (Linux-only; sysfs hwmon)
+	a.updateFans(&systemStats)
+
 	// GPU data
 	if a.gpuManager != nil {
 		// reset high gpu percent
@@ -244,13 +272,99 @@ func (a *Agent) getSystemStats(cacheTimeMs uint16) system.Stats {
 	a.systemInfo.MemPct = systemStats.MemPct
 	a.systemInfo.DiskPct = systemStats.DiskPct
 	a.systemInfo.Battery = systemStats.Battery
-	a.systemInfo.Uptime, _ = host.Uptime()
+	a.systemInfo.Uptime, _ = getUptime()
 	a.systemInfo.BandwidthBytes = systemStats.Bandwidth[0] + systemStats.Bandwidth[1]
 	a.systemInfo.Threads = a.systemDetails.Threads
 
 	return systemStats
 }
 
+// cpuModelFallbackKeys are the field names to look for in /proc/cpuinfo when
+// gopsutil fails to return a ModelName. The "cpu model" key is used on MIPS
+// (e.g. "MIPS 1004Kc V2.15"), while "system type" provides SoC information
+// on various embedded architectures.
+var cpuModelFallbackKeys = []string{"cpu model", "system type"}
+
+// getCpuModelFromCpuinfo reads /proc/cpuinfo and returns a CPU model string.
+// This is a fallback for architectures where gopsutil's cpu.Info() does not
+// populate ModelName, most notably MIPS.
+func getCpuModelFromCpuinfo() string {
+	file, err := os.Open("/proc/cpuinfo")
+	if err != nil {
+		return ""
+	}
+	defer file.Close()
+	return parseCpuModel(file)
+}
+
+// parseCpuModel scans r (expected to be /proc/cpuinfo content) and returns
+// a combined CPU model string. It collects values from all matching keys
+// and joins them with " / " when multiple are found.
+func parseCpuModel(r io.Reader) string {
+	lines := readLines(r)
+	var parts []string
+	for _, key := range cpuModelFallbackKeys {
+		for _, line := range lines {
+			after, found := strings.CutPrefix(line, key)
+			if !found {
+				continue
+			}
+			after = strings.TrimSpace(after)
+			if len(after) < 2 || after[0] != ':' {
+				continue
+			}
+			if value := strings.TrimSpace(after[1:]); value != "" {
+				parts = append(parts, value)
+				break
+			}
+		}
+	}
+	return strings.Join(parts, " / ")
+}
+
+// readLines reads all lines from r into a slice.
+func readLines(r io.Reader) []string {
+	scanner := bufio.NewScanner(r)
+	var lines []string
+	for scanner.Scan() {
+		lines = append(lines, scanner.Text())
+	}
+	return lines
+}
+
+// calculateHostMemoryUsage derives counters defensively because /proc/meminfo may
+// change while gopsutil reads it. Invalid unsigned subtractions saturate at zero.
+func calculateHostMemoryUsage(v *mem.VirtualMemoryStat, htop bool) (used, cacheBuff, swapUsed uint64) {
+	used = v.Used
+	if used > v.Total {
+		used = saturatingSub(v.Total, v.Available)
+	}
+
+	// gopsutil automatically adds SReclaimable to Cached.
+	cacheBuff = min(v.Cached, v.Total)
+	cacheBuff += min(v.Buffers, v.Total-cacheBuff)
+	cacheBuff = saturatingSub(cacheBuff, min(v.Shared, v.Total))
+	if v.Cached == 0 && v.Buffers == 0 {
+		cacheBuff = saturatingSub(v.Total, v.Free, used)
+	}
+	if htop {
+		used = saturatingSub(v.Total, v.Free, cacheBuff)
+	}
+	// Cached swap pages still occupy swap slots and are included in `free`'s used value.
+	return used, cacheBuff, saturatingSub(v.SwapTotal, v.SwapFree)
+}
+
+// saturatingSub subtracts each value, returning zero on underflow.
+func saturatingSub(value uint64, subtrahends ...uint64) uint64 {
+	for _, subtrahend := range subtrahends {
+		if subtrahend > value {
+			return 0
+		}
+		value -= subtrahend
+	}
+	return value
+}
+
 // getOsPrettyName attempts to get the pretty OS name from /etc/os-release on Linux systems
 func getOsPrettyName() (string, error) {
 	file, err := os.Open("/etc/os-release")

agent/system_test.go

@@ -0,0 +1,194 @@
+package agent
+
+import (
+	"strings"
+	"testing"
+
+	"github.com/henrygd/beszel/internal/common"
+	"github.com/henrygd/beszel/internal/entities/system"
+	"github.com/shirou/gopsutil/v4/mem"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestGatherStatsDoesNotAttachDetailsToCachedRequests(t *testing.T) {
+	agent := &Agent{
+		cache:         NewSystemDataCache(),
+		systemDetails: system.Details{Hostname: "updated-host", Podman: true},
+		detailsDirty:  true,
+	}
+	cached := &system.CombinedData{
+		Info: system.Info{Hostname: "cached-host"},
+	}
+	agent.cache.Set(cached, defaultDataCacheTimeMs)
+
+	response := agent.gatherStats(common.DataRequestOptions{CacheTimeMs: defaultDataCacheTimeMs})
+
+	assert.Same(t, cached, response)
+	assert.Nil(t, response.Details)
+	assert.True(t, agent.detailsDirty)
+	assert.Equal(t, "cached-host", response.Info.Hostname)
+	assert.Nil(t, cached.Details)
+
+	secondResponse := agent.gatherStats(common.DataRequestOptions{CacheTimeMs: defaultDataCacheTimeMs})
+	assert.Same(t, cached, secondResponse)
+	assert.Nil(t, secondResponse.Details)
+}
+
+func TestCalculateHostMemoryUsage(t *testing.T) {
+	tests := []struct {
+		name                      string
+		memory                    mem.VirtualMemoryStat
+		htop                      bool
+		used, cacheBuff, swapUsed uint64
+	}{
+		{
+			name:      "normal",
+			memory:    mem.VirtualMemoryStat{Total: 100, Available: 40, Used: 60, Free: 20, Cached: 25, Buffers: 10, Shared: 5, SwapTotal: 20, SwapFree: 8, SwapCached: 2},
+			used:      60,
+			cacheBuff: 30,
+			swapUsed:  12,
+		},
+		{
+			name:      "inconsistent counters saturate",
+			memory:    mem.VirtualMemoryStat{Total: 100, Available: 110, Used: ^uint64(0) - 9, Free: 90, Cached: 5, Buffers: 10, Shared: 20, SwapTotal: 10, SwapFree: 9, SwapCached: 2},
+			used:      0,
+			cacheBuff: 0,
+			swapUsed:  1,
+		},
+		{
+			name:      "htop subtraction saturates",
+			memory:    mem.VirtualMemoryStat{Total: 100, Available: 20, Used: 80, Free: 90, Cached: 20, Buffers: 5, SwapTotal: 30, SwapFree: 10, SwapCached: 5},
+			htop:      true,
+			used:      0,
+			cacheBuff: 25,
+			swapUsed:  20,
+		},
+		{
+			name:      "zero cache from shared cancellation does not fall back",
+			memory:    mem.VirtualMemoryStat{Total: 100, Used: 60, Free: 10, Cached: 20, Buffers: 10, Shared: 30},
+			used:      60,
+			cacheBuff: 0,
+		},
+		{
+			name:      "absent cache counters use fallback",
+			memory:    mem.VirtualMemoryStat{Total: 100, Used: 60, Free: 10},
+			used:      60,
+			cacheBuff: 30,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			used, cacheBuff, swapUsed := calculateHostMemoryUsage(&tt.memory, tt.htop)
+			assert.Equal(t, tt.used, used)
+			assert.Equal(t, tt.cacheBuff, cacheBuff)
+			assert.Equal(t, tt.swapUsed, swapUsed)
+		})
+	}
+}
+
+func TestUpdateSystemDetailsMarksDetailsDirty(t *testing.T) {
+	agent := &Agent{}
+
+	agent.updateSystemDetails(func(details *system.Details) {
+		details.Hostname = "updated-host"
+		details.Podman = true
+	})
+
+	assert.True(t, agent.detailsDirty)
+	assert.Equal(t, "updated-host", agent.systemDetails.Hostname)
+	assert.True(t, agent.systemDetails.Podman)
+
+	original := &system.CombinedData{}
+	realTimeResponse := agent.attachSystemDetails(original, 1000, true)
+	assert.Same(t, original, realTimeResponse)
+	assert.Nil(t, realTimeResponse.Details)
+	assert.True(t, agent.detailsDirty)
+
+	response := agent.attachSystemDetails(original, defaultDataCacheTimeMs, false)
+	require.NotNil(t, response.Details)
+	assert.NotSame(t, original, response)
+	assert.Equal(t, "updated-host", response.Details.Hostname)
+	assert.True(t, response.Details.Podman)
+	assert.False(t, agent.detailsDirty)
+	assert.Nil(t, original.Details)
+}
+
+func TestParseCpuModel(t *testing.T) {
+	tests := []struct {
+		name     string
+		input    string
+		expected string
+	}{
+		{
+			name: "MIPS with both cpu model and system type",
+			input: `system type             : MediaTek MT7621 ver:1 eco:3
+machine                 : ASUS RT-AX53U
+processor               : 0
+cpu model               : MIPS 1004Kc V2.15
+BogoMIPS                : 586.13
+wait instruction        : yes`,
+			expected: "MIPS 1004Kc V2.15 / MediaTek MT7621 ver:1 eco:3",
+		},
+		{
+			name: "MIPS with different SoC",
+			input: `system type             : Atheros AR7161 rev 2
+machine                 : NETGEAR WNDR3700
+processor               : 0
+cpu model               : MIPS 24Kc V7.4
+BogoMIPS                : 452.19`,
+			expected: "MIPS 24Kc V7.4 / Atheros AR7161 rev 2",
+		},
+		{
+			name: "only system type when cpu model missing",
+			input: `system type             : Broadcom BCM47xx
+processor               : 0
+BogoMIPS                : 296.11`,
+			expected: "Broadcom BCM47xx",
+		},
+		{
+			name: "only cpu model when system type missing",
+			input: `processor               : 0
+cpu model               : MIPS 34Kc V2.15
+BogoMIPS                : 300.00`,
+			expected: "MIPS 34Kc V2.15",
+		},
+		{
+			name: "x86 cpuinfo returns empty",
+			input: `processor	: 0
+vendor_id	: GenuineIntel
+cpu family	: 6
+model		: 142
+model name	: Intel(R) Core(TM) i5-8250U CPU @ 1.60GHz
+stepping	: 10`,
+			expected: "",
+		},
+		{
+			name:     "empty input",
+			input:    "",
+			expected: "",
+		},
+		{
+			name: "cpu model with extra whitespace",
+			input: `processor               : 0
+cpu model               :   MIPS 34Kc V2.15
+BogoMIPS                : 300.00`,
+			expected: "MIPS 34Kc V2.15",
+		},
+		{
+			name: "cpu model without value",
+			input: `processor               : 0
+cpu model               :
+BogoMIPS                : 300.00`,
+			expected: "",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := parseCpuModel(strings.NewReader(tt.input))
+			assert.Equal(t, tt.expected, result)
+		})
+	}
+}

agent/test-data/smart/apple_nvme.json

@@ -0,0 +1,51 @@
+{
+  "json_format_version": [1, 0],
+  "smartctl": {
+    "version": [7, 4],
+    "argv": ["smartctl", "-aix", "-j", "IOService:/AppleARMPE/arm-io@10F00000/AppleT810xIO/ans@77400000/AppleASCWrapV4/iop-ans-nub/RTBuddy(ANS2)/RTBuddyService/AppleANS3NVMeController/NS_01@1"],
+    "exit_status": 4
+  },
+  "device": {
+    "name": "IOService:/AppleARMPE/arm-io@10F00000/AppleT810xIO/ans@77400000/AppleASCWrapV4/iop-ans-nub/RTBuddy(ANS2)/RTBuddyService/AppleANS3NVMeController/NS_01@1",
+    "info_name": "IOService:/AppleARMPE/arm-io@10F00000/AppleT810xIO/ans@77400000/AppleASCWrapV4/iop-ans-nub/RTBuddy(ANS2)/RTBuddyService/AppleANS3NVMeController/NS_01@1",
+    "type": "nvme",
+    "protocol": "NVMe"
+  },
+  "model_name": "APPLE SSD AP0256Q",
+  "serial_number": "0ba0147940253c15",
+  "firmware_version": "555",
+  "smart_support": {
+    "available": true,
+    "enabled": true
+  },
+  "smart_status": {
+    "passed": true,
+    "nvme": {
+      "value": 0
+    }
+  },
+  "nvme_smart_health_information_log": {
+    "critical_warning": 0,
+    "temperature": 42,
+    "available_spare": 100,
+    "available_spare_threshold": 99,
+    "percentage_used": 1,
+    "data_units_read": 270189386,
+    "data_units_written": 166753862,
+    "host_reads": 7543766995,
+    "host_writes": 3761621926,
+    "controller_busy_time": 0,
+    "power_cycles": 366,
+    "power_on_hours": 2850,
+    "unsafe_shutdowns": 195,
+    "media_errors": 0,
+    "num_err_log_entries": 0
+  },
+  "temperature": {
+    "current": 42
+  },
+  "power_cycle_count": 366,
+  "power_on_time": {
+    "hours": 2850
+  }
+}

agent/test-data/zfs/zfs_list.txt

@@ -0,0 +1,9 @@
+tank	12000000000000	11999000000000	/tank
+tank/apps	1000000000000	11999000000000	/tank/apps
+tank/backup	2000000000000	11999000000000	/tank/backup
+tank/media	1000000000000	11999000000000	/tank/my media
+rpool	900000000000	300000000000	-
+rpool/ROOT	1000000000	300000000000	-
+rpool/ROOT/pve-1	890000000000	300000000000	/
+rpool/data	9000000000	300000000000	-
+rpool/data/subvol-100-disk-0	400000000000	300000000000	/subvol-100-disk-0

agent/test-data/zfs/zpool_list.txt

@@ -0,0 +1,2 @@
+tank	23999000000000	12000000000000	11999000000000	ONLINE
+rpool	1200000000000	900000000000	300000000000	DEGRADED

agent/test-data/zfs/zpool_status.txt

@@ -0,0 +1,29 @@
+  pool: tank
+ state: ONLINE
+  scan: scrub repaired 0B in 00:05:12 with 0 errors on Sun Jun  1 02:00:12 2025
+config:
+
+	NAME        STATE     READ WRITE CKSUM
+	tank        ONLINE       0     0     0
+	  mirror-0  ONLINE       0     0     0
+	    sda     ONLINE       0     0     0
+	    sdb     ONLINE       0     0     0
+
+errors: No known data errors
+
+  pool: rpool
+ state: DEGRADED
+ status: One or more devices could not be used because the label is missing or
+	invalid.  Sufficient replicas exist for the pool to continue functioning in a
+	degraded state.
+  scan: scrub in progress since Sun Jun  8 01:00:00 2025
+	10.00% done, 01:30:00 to go, 0.00/s
+config:
+
+	NAME        STATE     READ WRITE CKSUM
+	rpool       DEGRADED     0     0     0
+	  mirror-0  DEGRADED     0     0     0
+	    sda     ONLINE       0     0     0
+	    sdb     FAULTED      1     2     3
+
+errors: 1 data errors, use '-v' for a list

agent/uptime_linux.go

@@ -0,0 +1,44 @@
+//go:build linux
+
+package agent
+
+import (
+	"math"
+	"os"
+	"strconv"
+	"strings"
+
+	"github.com/shirou/gopsutil/v4/host"
+)
+
+// uptimeFilePath is a variable so tests can point it at a fixture.
+var uptimeFilePath = "/proc/uptime"
+
+// getUptime returns the system uptime in seconds.
+//
+// This reads /proc/uptime instead of using host.Uptime(), which calls the
+// sysinfo(2) syscall. Inside an LXC container lxcfs virtualizes /proc/uptime
+// but cannot intercept a syscall, so sysinfo(2) reports the host's uptime
+// rather than the container's.
+//
+// Falls back to host.Uptime() if /proc/uptime is missing or unparseable, so
+// behavior is unchanged anywhere the file isn't available.
+func getUptime() (uint64, error) {
+	data, err := os.ReadFile(uptimeFilePath)
+	if err != nil {
+		return host.Uptime()
+	}
+	fields := strings.Fields(string(data))
+	if len(fields) == 0 {
+		return host.Uptime()
+	}
+	seconds, err := strconv.ParseFloat(fields[0], 64)
+	if err != nil ||
+		math.IsNaN(seconds) ||
+		math.IsInf(seconds, 0) ||
+		seconds < 0 ||
+		seconds >= 1<<64 {
+		return host.Uptime()
+	}
+	return uint64(seconds), nil
+}

agent/uptime_linux_test.go

@@ -0,0 +1,101 @@
+//go:build linux
+
+package agent
+
+import (
+	"os"
+	"path/filepath"
+	"testing"
+)
+
+func TestGetUptimeFromProc(t *testing.T) {
+	tests := []struct {
+		name     string
+		contents string
+		want     uint64
+	}{
+		{"typical", "12345.67 98765.43\n", 12345},
+		{"zero", "0.00 0.00\n", 0},
+		{"no trailing newline", "42.99 7.00", 42},
+		{"single field", "600.5", 600},
+		{"large value", "266030.12 1000000.00\n", 266030},
+	}
+
+	prev := uptimeFilePath
+	t.Cleanup(func() { uptimeFilePath = prev })
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			path := filepath.Join(t.TempDir(), "uptime")
+			if err := os.WriteFile(path, []byte(tt.contents), 0o644); err != nil {
+				t.Fatal(err)
+			}
+			uptimeFilePath = path
+
+			got, err := getUptime()
+			if err != nil {
+				t.Fatalf("getUptime() returned error: %v", err)
+			}
+			if got != tt.want {
+				t.Errorf("getUptime() = %d, want %d", got, tt.want)
+			}
+		})
+	}
+}
+
+func writeUptime(contents string) func(t *testing.T) string {
+	return func(t *testing.T) string {
+		path := filepath.Join(t.TempDir(), "uptime")
+		if err := os.WriteFile(path, []byte(contents), 0o644); err != nil {
+			t.Fatal(err)
+		}
+		return path
+	}
+}
+
+// Malformed, missing, or out-of-range input must fall back to host.Uptime()
+// rather than returning a bogus value, so the agent still reports something sane.
+func TestGetUptimeFallsBack(t *testing.T) {
+	prev := uptimeFilePath
+	t.Cleanup(func() { uptimeFilePath = prev })
+
+	for _, tt := range []struct {
+		name    string
+		prepare func(t *testing.T) string
+	}{
+		{"missing file", func(t *testing.T) string {
+			return filepath.Join(t.TempDir(), "does-not-exist")
+		}},
+		{"empty file", func(t *testing.T) string {
+			path := filepath.Join(t.TempDir(), "uptime")
+			if err := os.WriteFile(path, nil, 0o644); err != nil {
+				t.Fatal(err)
+			}
+			return path
+		}},
+		{"unparseable", func(t *testing.T) string {
+			path := filepath.Join(t.TempDir(), "uptime")
+			if err := os.WriteFile(path, []byte("not-a-number 1.0\n"), 0o644); err != nil {
+				t.Fatal(err)
+			}
+			return path
+		}},
+		{"NaN", writeUptime("NaN 1.0\n")},
+		{"positive infinity", writeUptime("+Inf 1.0\n")},
+		{"negative infinity", writeUptime("-Inf 1.0\n")},
+		{"negative", writeUptime("-42.5 1.0\n")},
+		{"exceeds uint64 range", writeUptime("1e20 1.0\n")},
+	} {
+		t.Run(tt.name, func(t *testing.T) {
+			uptimeFilePath = tt.prepare(t)
+
+			got, err := getUptime()
+			if err != nil {
+				t.Fatalf("getUptime() returned error: %v", err)
+			}
+			if got == 0 {
+				t.Error("getUptime() = 0, expected fallback to host.Uptime()")
+			}
+		})
+	}
+}

agent/uptime_stub.go

@@ -0,0 +1,10 @@
+//go:build !linux
+
+package agent
+
+import "github.com/shirou/gopsutil/v4/host"
+
+// getUptime returns the system uptime in seconds.
+func getUptime() (uint64, error) {
+	return host.Uptime()
+}

agent/utils/utils.go

@@ -1,6 +1,8 @@
+// Package utils provides utility functions for the agent.
 package utils
 
 import (
+	"fmt"
 	"io"
 	"math"
 	"os"
@@ -68,6 +70,9 @@ func ReadStringFileLimited(path string, maxSize int) (string, error) {
 	if err != nil && err != io.EOF {
 		return "", err
 	}
+	if n < 0 {
+		return "", fmt.Errorf("%s returned negative bytes: %d", path, n)
+	}
 	return strings.TrimSpace(string(buf[:n])), nil
 }
 

agent/zfs/zfs.go

@@ -0,0 +1,160 @@
+// Package zfs provides functions to read ZFS statistics.
+package zfs
+
+import (
+	"bufio"
+	"bytes"
+	"context"
+	"errors"
+	"fmt"
+	"os"
+	"os/exec"
+	"strconv"
+	"strings"
+	"time"
+)
+
+var commandTimeout = 10 * time.Second
+
+var commandOutput = func(name string, args ...string) ([]byte, error) {
+	ctx, cancel := context.WithTimeout(context.Background(), commandTimeout)
+	defer cancel()
+	cmd := exec.CommandContext(ctx, name, args...)
+	cmd.Env = append(os.Environ(), "LC_ALL=C", "LANG=C")
+	out, err := cmd.Output()
+	if ctx.Err() != nil {
+		return nil, fmt.Errorf("%s timed out after %s: %w", name, commandTimeout, ctx.Err())
+	}
+	return out, err
+}
+
+// ErrNoZfs is returned when the ZFS utilities or kernel interfaces are unavailable.
+var ErrNoZfs = errors.New("zfs utilities unavailable")
+
+// PoolStat is a snapshot of a ZFS pool's capacity and health.
+type PoolStat struct {
+	Name   string
+	Size   uint64 // total capacity in bytes
+	Alloc  uint64 // allocated bytes
+	Free   uint64 // free bytes
+	Health string // ONLINE, DEGRADED, FAULTED, ...
+}
+
+// PoolKernelStat is the inexpensive pool telemetry exposed by the ZFS kernel.
+// NRead and NWrite are cumulative byte counters since the pool was imported.
+type PoolKernelStat struct {
+	Name   string
+	Health string
+	NRead  uint64
+	NWrite uint64
+}
+
+// PoolIoStats holds calculated per-second I/O rates for a pool.
+type PoolIoStats struct {
+	NRead  uint64
+	NWrite uint64
+}
+
+// Dataset is a single ZFS dataset with usage information.
+type Dataset struct {
+	Name       string
+	Used       uint64
+	Avail      uint64
+	Mountpoint string
+}
+
+// PoolStats returns capacity and health for all pools on the system using
+// `zpool list`. Frequent health and I/O sampling uses PoolKernelStats instead.
+func PoolStats() ([]PoolStat, error) {
+	out, err := commandOutput("zpool", "list", "-Hp", "-o", "name,size,alloc,free,health")
+	if err != nil {
+		var exitErr *exec.ExitError
+		if errors.As(err, &exitErr) && strings.Contains(string(exitErr.Stderr), "no pools available") {
+			return nil, nil
+		}
+		return nil, fmt.Errorf("zpool list: %w", err)
+	}
+	return parseZpoolListOutput(out)
+}
+
+// Datasets returns all datasets on the system with usage and mountpoint
+// information using `zfs list` (recursive by default).
+func Datasets() ([]Dataset, error) {
+	out, err := commandOutput("zfs", "list", "-Hp", "-o", "name,used,avail,mountpoint")
+	if err != nil {
+		return nil, fmt.Errorf("zfs list: %w", err)
+	}
+	return parseZfsListOutput(out)
+}
+
+// parseZpoolListOutput parses `zpool list -Hp -o name,size,alloc,free,health` output.
+// Columns are tab-separated; numeric columns are raw bytes.
+func parseZpoolListOutput(out []byte) ([]PoolStat, error) {
+	var pools []PoolStat
+	scanner := bufio.NewScanner(bytes.NewReader(out))
+	for scanner.Scan() {
+		line := strings.TrimSpace(scanner.Text())
+		if line == "" {
+			continue
+		}
+		if line == "no pools available" && len(pools) == 0 {
+			return nil, nil
+		}
+		fields := strings.Split(line, "\t")
+		if len(fields) < 5 {
+			return nil, fmt.Errorf("unexpected zpool list line: %q", line)
+		}
+		size, err := strconv.ParseUint(fields[1], 10, 64)
+		if err != nil {
+			return nil, fmt.Errorf("parsing size for pool %q: %w", fields[0], err)
+		}
+		alloc, err := strconv.ParseUint(fields[2], 10, 64)
+		if err != nil {
+			return nil, fmt.Errorf("parsing alloc for pool %q: %w", fields[0], err)
+		}
+		free, err := strconv.ParseUint(fields[3], 10, 64)
+		if err != nil {
+			return nil, fmt.Errorf("parsing free for pool %q: %w", fields[0], err)
+		}
+		pools = append(pools, PoolStat{
+			Name:   fields[0],
+			Size:   size,
+			Alloc:  alloc,
+			Free:   free,
+			Health: fields[4],
+		})
+	}
+	return pools, scanner.Err()
+}
+
+// parseZfsListOutput parses `zfs list -Hp -o name,used,avail,mountpoint` output.
+// The mountpoint column may contain spaces, so it is split on tabs only.
+func parseZfsListOutput(out []byte) ([]Dataset, error) {
+	var datasets []Dataset
+	scanner := bufio.NewScanner(bytes.NewReader(out))
+	for scanner.Scan() {
+		line := strings.TrimSpace(scanner.Text())
+		if line == "" {
+			continue
+		}
+		fields := strings.SplitN(line, "\t", 4)
+		if len(fields) < 4 {
+			return nil, fmt.Errorf("unexpected zfs list line: %q", line)
+		}
+		used, err := strconv.ParseUint(fields[1], 10, 64)
+		if err != nil {
+			return nil, fmt.Errorf("parsing used for dataset %q: %w", fields[0], err)
+		}
+		avail, err := strconv.ParseUint(fields[2], 10, 64)
+		if err != nil {
+			return nil, fmt.Errorf("parsing avail for dataset %q: %w", fields[0], err)
+		}
+		datasets = append(datasets, Dataset{
+			Name:       fields[0],
+			Used:       used,
+			Avail:      avail,
+			Mountpoint: fields[3],
+		})
+	}
+	return datasets, scanner.Err()
+}

agent/zfs/zfs_freebsd.go

@@ -3,9 +3,17 @@
 package zfs
 
 import (
+	"errors"
+
 	"golang.org/x/sys/unix"
 )
 
 func ARCSize() (uint64, error) {
 	return unix.SysctlUint64("kstat.zfs.misc.arcstats.size")
 }
+
+// FreeBSD does not expose Linux's per-pool procfs kstats. Capacity, health,
+// and detail collection still work through the cached utilities.
+func PoolKernelStats() ([]PoolKernelStat, error) {
+	return nil, errors.ErrUnsupported
+}

agent/zfs/zfs_linux.go

@@ -5,14 +5,18 @@ package zfs
 
 import (
 	"bufio"
+	"errors"
 	"fmt"
 	"os"
+	"path/filepath"
 	"strconv"
 	"strings"
 )
 
+var procZfsPath = "/proc/spl/kstat/zfs"
+
 func ARCSize() (uint64, error) {
-	file, err := os.Open("/proc/spl/kstat/zfs/arcstats")
+	file, err := os.Open(filepath.Join(procZfsPath, "arcstats"))
 	if err != nil {
 		return 0, err
 	}
@@ -29,6 +33,167 @@ func ARCSize() (uint64, error) {
 			return strconv.ParseUint(fields[2], 10, 64)
 		}
 	}
+	if err := scanner.Err(); err != nil {
+		return 0, err
+	}
 
 	return 0, fmt.Errorf("size field not found in arcstats")
 }
+
+// PoolKernelStats reads pool state and cumulative I/O counters directly from
+// procfs. These kstats are the same interfaces used by node_exporter's Linux
+// ZFS collector and avoid keeping a `zpool iostat` subprocess alive.
+func PoolKernelStats() ([]PoolKernelStat, error) {
+	poolDirs := make(map[string]struct{})
+	for _, filename := range []string{"state", "io", "objset-*"} {
+		paths, err := filepath.Glob(filepath.Join(procZfsPath, "*", filename))
+		if err != nil {
+			return nil, err
+		}
+		for _, path := range paths {
+			poolDirs[filepath.Dir(path)] = struct{}{}
+		}
+	}
+	if len(poolDirs) == 0 {
+		return nil, ErrNoZfs
+	}
+	pools := make([]PoolKernelStat, 0, len(poolDirs))
+	for poolDir := range poolDirs {
+		nread, nwrite, err := readPoolCounters(poolDir)
+		if err != nil {
+			if errors.Is(err, os.ErrNotExist) {
+				continue // pool may have been exported after the glob
+			}
+			return nil, err
+		}
+		state, err := os.ReadFile(filepath.Join(poolDir, "state"))
+		if err != nil && !errors.Is(err, os.ErrNotExist) {
+			return nil, err
+		}
+		pools = append(pools, PoolKernelStat{
+			Name: filepath.Base(poolDir), Health: strings.ToUpper(strings.TrimSpace(string(state))),
+			NRead: nread, NWrite: nwrite,
+		})
+	}
+	if len(pools) == 0 {
+		return nil, ErrNoZfs
+	}
+	return pools, nil
+}
+
+// readPoolCounters supports both ZFS kernel interfaces. OpenZFS through 2.3
+// exposes aggregate vdev counters in "io". When that file is unavailable, sum
+// the logical I/O counters exposed for each dataset in the pool.
+func readPoolCounters(poolDir string) (uint64, uint64, error) {
+	nread, nwrite, err := readPoolIO(filepath.Join(poolDir, "io"))
+	if err == nil || !errors.Is(err, os.ErrNotExist) {
+		return nread, nwrite, err
+	}
+	return readPoolObjsets(poolDir)
+}
+
+func readPoolIO(path string) (uint64, uint64, error) {
+	file, err := os.Open(path)
+	if err != nil {
+		return 0, 0, err
+	}
+	defer file.Close()
+
+	scanner := bufio.NewScanner(file)
+	for scanner.Scan() {
+		fields := strings.Fields(scanner.Text())
+		if len(fields) < 2 || fields[0] != "nread" {
+			continue
+		}
+		if !scanner.Scan() {
+			break
+		}
+		values := strings.Fields(scanner.Text())
+		if len(values) < 2 {
+			break
+		}
+		nread, err := strconv.ParseUint(values[0], 10, 64)
+		if err != nil {
+			return 0, 0, fmt.Errorf("parsing nread in %s: %w", path, err)
+		}
+		nwrite, err := strconv.ParseUint(values[1], 10, 64)
+		if err != nil {
+			return 0, 0, fmt.Errorf("parsing nwritten in %s: %w", path, err)
+		}
+		return nread, nwrite, nil
+	}
+	if err := scanner.Err(); err != nil {
+		return 0, 0, err
+	}
+	return 0, 0, fmt.Errorf("I/O counters not found in %s", path)
+}
+
+func readPoolObjsets(poolDir string) (uint64, uint64, error) {
+	paths, err := filepath.Glob(filepath.Join(poolDir, "objset-*"))
+	if err != nil {
+		return 0, 0, err
+	}
+	if len(paths) == 0 {
+		return 0, 0, fmt.Errorf("dataset I/O counters not found in %s", poolDir)
+	}
+
+	var totalRead, totalWrite uint64
+	objsetsRead := 0
+	for _, path := range paths {
+		nread, nwrite, err := readObjsetIO(path)
+		if errors.Is(err, os.ErrNotExist) {
+			continue // dataset may have been destroyed after the glob
+		}
+		if err != nil {
+			return 0, 0, err
+		}
+		totalRead += nread
+		totalWrite += nwrite
+		objsetsRead++
+	}
+	if objsetsRead == 0 {
+		return 0, 0, fmt.Errorf("dataset I/O counters not found in %s", poolDir)
+	}
+	return totalRead, totalWrite, nil
+}
+
+func readObjsetIO(path string) (uint64, uint64, error) {
+	file, err := os.Open(path)
+	if err != nil {
+		return 0, 0, err
+	}
+	defer file.Close()
+
+	var nread, nwrite uint64
+	var foundRead, foundWrite bool
+	scanner := bufio.NewScanner(file)
+	for scanner.Scan() {
+		fields := strings.Fields(scanner.Text())
+		if len(fields) < 3 {
+			continue
+		}
+		var target *uint64
+		switch fields[0] {
+		case "nread":
+			target = &nread
+			foundRead = true
+		case "nwritten":
+			target = &nwrite
+			foundWrite = true
+		default:
+			continue
+		}
+		value, err := strconv.ParseUint(fields[2], 10, 64)
+		if err != nil {
+			return 0, 0, fmt.Errorf("parsing %s in %s: %w", fields[0], path, err)
+		}
+		*target = value
+	}
+	if err := scanner.Err(); err != nil {
+		return 0, 0, err
+	}
+	if !foundRead || !foundWrite {
+		return 0, 0, fmt.Errorf("incomplete I/O counters in %s", path)
+	}
+	return nread, nwrite, nil
+}

agent/zfs/zfs_linux_test.go

@@ -0,0 +1,90 @@
+//go:build testing && linux
+
+package zfs
+
+import (
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestPoolKernelStats(t *testing.T) {
+	root := t.TempDir()
+	oldPath := procZfsPath
+	procZfsPath = root
+	t.Cleanup(func() { procZfsPath = oldPath })
+
+	poolDir := filepath.Join(root, "tank")
+	require.NoError(t, os.MkdirAll(poolDir, 0o755))
+	require.NoError(t, os.WriteFile(filepath.Join(poolDir, "io"), []byte(
+		"11 3 0x00 1 80 0 0\n"+
+			"nread nwritten reads writes wtime wlentime wupdate rtime rlentime rupdate wcnt rcnt\n"+
+			"1884160 6450688 22 978 0 0 0 0 0 0 0 0\n",
+	), 0o644))
+	require.NoError(t, os.WriteFile(filepath.Join(poolDir, "state"), []byte("DEGRADED\n"), 0o644))
+
+	stats, err := PoolKernelStats()
+	require.NoError(t, err)
+	require.Len(t, stats, 1)
+	assert.Equal(t, PoolKernelStat{
+		Name: "tank", Health: "DEGRADED", NRead: 1884160, NWrite: 6450688,
+	}, stats[0])
+}
+
+func TestPoolKernelStatsOpenZfs24(t *testing.T) {
+	root := t.TempDir()
+	oldPath := procZfsPath
+	procZfsPath = root
+	t.Cleanup(func() { procZfsPath = oldPath })
+
+	poolDir := filepath.Join(root, "tank")
+	require.NoError(t, os.MkdirAll(poolDir, 0o755))
+	require.NoError(t, os.WriteFile(filepath.Join(poolDir, "state"), []byte("ONLINE\n"), 0o644))
+	require.NoError(t, os.WriteFile(filepath.Join(poolDir, "objset-0x1"), []byte(
+		"34 1 0x01 28 7872 0 0\n"+
+			"name type data\n"+
+			"dataset_name 7 tank\n"+
+			"nwritten 4 2000\n"+
+			"nread 4 1000\n",
+	), 0o644))
+	require.NoError(t, os.WriteFile(filepath.Join(poolDir, "objset-0x2"), []byte(
+		"34 1 0x01 28 7872 0 0\n"+
+			"name type data\n"+
+			"dataset_name 7 tank/videos\n"+
+			"nwritten 4 400\n"+
+			"nread 4 300\n",
+	), 0o644))
+
+	stats, err := PoolKernelStats()
+	require.NoError(t, err)
+	require.Len(t, stats, 1)
+	assert.Equal(t, PoolKernelStat{
+		Name: "tank", Health: "ONLINE", NRead: 1300, NWrite: 2400,
+	}, stats[0])
+}
+
+func TestPoolKernelStatsNoZfs(t *testing.T) {
+	oldPath := procZfsPath
+	procZfsPath = t.TempDir()
+	t.Cleanup(func() { procZfsPath = oldPath })
+
+	_, err := PoolKernelStats()
+	assert.ErrorIs(t, err, ErrNoZfs)
+}
+
+func TestReadPoolIORejectsMalformedCounters(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "io")
+	require.NoError(t, os.WriteFile(path, []byte("nread nwritten\nnope 10\n"), 0o644))
+	_, _, err := readPoolIO(path)
+	require.Error(t, err)
+}
+
+func TestReadObjsetIORequiresAllCounters(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "objset-0x1")
+	require.NoError(t, os.WriteFile(path, []byte("nread 4 10\n"), 0o644))
+	_, _, err := readObjsetIO(path)
+	require.Error(t, err)
+}

agent/zfs/zfs_status.go

@@ -0,0 +1,150 @@
+package zfs
+
+import (
+	"bufio"
+	"bytes"
+	"fmt"
+	"regexp"
+	"strconv"
+	"strings"
+)
+
+// PoolStatus holds parsed `zpool status` information for one pool.
+type PoolStatus struct {
+	Name  string
+	State string // ONLINE, DEGRADED, FAULTED, ...
+	Scrub ScrubStatus
+	Vdevs []VdevStatus
+}
+
+// ScrubStatus holds the scrub (or resilver) status parsed from the scan line.
+type ScrubStatus struct {
+	State    string // NONE, SCANNING, FINISHED, CANCELED
+	Progress string // e.g. "10.00%" while scanning
+	Errors   uint64
+}
+
+// VdevStatus is a single vdev row (mirror, raidz, or leaf disk).
+type VdevStatus struct {
+	Name         string
+	State        string
+	ReadErrs     uint64
+	WriteErrs    uint64
+	ChecksumErrs uint64
+}
+
+var (
+	progressRe = regexp.MustCompile(`(\d+\.\d+)%\s+done`)
+	errorsRe   = regexp.MustCompile(`with\s+(\d+)\s+errors`)
+)
+
+// PoolStatuses runs `zpool status` and parses per-pool state, scrub, and vdev
+// information. The human-readable format has been stable across OpenZFS
+// releases; rows are matched by their tabular shape rather than position.
+func PoolStatuses() ([]PoolStatus, error) {
+	out, err := commandOutput("zpool", "status")
+	if err != nil {
+		return nil, fmt.Errorf("zpool status: %w", err)
+	}
+	return parseZpoolStatusOutput(out)
+}
+
+// parseZpoolStatusOutput parses the output of `zpool status`.
+func parseZpoolStatusOutput(out []byte) ([]PoolStatus, error) {
+	var pools []PoolStatus
+	var current *PoolStatus
+	inConfig := false
+	scanContinuation := false // next non-blank line continues the scan line (progress)
+
+	scanner := bufio.NewScanner(bytes.NewReader(out))
+	for scanner.Scan() {
+		line := scanner.Text()
+		trimmed := strings.TrimSpace(line)
+
+		switch {
+		case strings.HasPrefix(trimmed, "pool:"):
+			pools = append(pools, PoolStatus{Name: strings.TrimSpace(strings.TrimPrefix(trimmed, "pool:"))})
+			current = &pools[len(pools)-1]
+			inConfig = false
+			scanContinuation = false
+		case current == nil:
+			continue
+		case strings.HasPrefix(trimmed, "state:"):
+			current.State = strings.TrimSpace(strings.TrimPrefix(trimmed, "state:"))
+		case strings.HasPrefix(trimmed, "scan:"):
+			current.Scrub = parseScanLine(trimmed)
+			// zpool status prints the progress percentage on the line after scan.
+			scanContinuation = true
+		case trimmed == "config:":
+			inConfig = true
+		case scanContinuation:
+			// The line after scan: may be an indented progress continuation.
+			if m := progressRe.FindStringSubmatch(trimmed); m != nil {
+				current.Scrub.Progress = m[1] + "%"
+			}
+			scanContinuation = false
+		case inConfig && (line == "" || strings.HasPrefix(line, " ") || strings.HasPrefix(line, "\t")):
+			// Table rows are indented; blank lines separate sections. The
+			// column header and the pool's own row are skipped.
+			if trimmed != "" && !strings.HasPrefix(trimmed, "NAME") {
+				if vdev, ok := parseVdevLine(trimmed, current.Name); ok {
+					current.Vdevs = append(current.Vdevs, vdev)
+				}
+			}
+		case inConfig:
+			// unindented line (errors:, status:, next pool:) ends the table
+			inConfig = false
+		}
+	}
+	return pools, scanner.Err()
+}
+
+// parseScanLine maps a `scan:` line to a ScrubStatus.
+func parseScanLine(line string) ScrubStatus {
+	var scrub ScrubStatus
+	switch {
+	case strings.Contains(line, "in progress"):
+		scrub.State = "SCANNING"
+	case strings.Contains(line, "canceled"):
+		scrub.State = "CANCELED"
+	case strings.Contains(line, "repaired"), strings.Contains(line, "resilvered"):
+		scrub.State = "FINISHED"
+	default:
+		scrub.State = "NONE"
+	}
+	if m := progressRe.FindStringSubmatch(line); m != nil {
+		scrub.Progress = m[1] + "%"
+	}
+	if m := errorsRe.FindStringSubmatch(line); m != nil {
+		if n, err := strconv.ParseUint(m[1], 10, 64); err == nil {
+			scrub.Errors = n
+		}
+	}
+	return scrub
+}
+
+// parseVdevLine parses one row of the config table. Rows have the shape
+// "NAME STATE READ WRITE CKSUM [extra...]". The first data row is the pool
+// itself and is skipped since it duplicates pool-level info.
+func parseVdevLine(line, poolName string) (VdevStatus, bool) {
+	fields := strings.Fields(line)
+	if len(fields) < 5 {
+		return VdevStatus{}, false
+	}
+	if fields[0] == poolName {
+		return VdevStatus{}, false
+	}
+	read, err1 := strconv.ParseUint(fields[2], 10, 64)
+	write, err2 := strconv.ParseUint(fields[3], 10, 64)
+	cksum, err3 := strconv.ParseUint(fields[4], 10, 64)
+	if err1 != nil || err2 != nil || err3 != nil {
+		return VdevStatus{}, false
+	}
+	return VdevStatus{
+		Name:         fields[0],
+		State:        fields[1],
+		ReadErrs:     read,
+		WriteErrs:    write,
+		ChecksumErrs: cksum,
+	}, true
+}

agent/zfs/zfs_test.go

@@ -0,0 +1,143 @@
+//go:build testing
+
+package zfs
+
+import (
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func fixturePath(name string) string {
+	return filepath.Join("..", "test-data", "zfs", name)
+}
+
+func TestParseZpoolListOutput(t *testing.T) {
+	data, err := os.ReadFile(fixturePath("zpool_list.txt"))
+	require.NoError(t, err)
+
+	pools, err := parseZpoolListOutput(data)
+	require.NoError(t, err)
+	require.Len(t, pools, 2)
+
+	assert.Equal(t, PoolStat{Name: "tank", Size: 23999000000000, Alloc: 12000000000000, Free: 11999000000000, Health: "ONLINE"}, pools[0])
+	assert.Equal(t, PoolStat{Name: "rpool", Size: 1200000000000, Alloc: 900000000000, Free: 300000000000, Health: "DEGRADED"}, pools[1])
+}
+
+func TestParseZpoolListOutputIgnoresEmptyLines(t *testing.T) {
+	pools, err := parseZpoolListOutput([]byte("tank\t100\t50\t50\tONLINE\n\n"))
+	require.NoError(t, err)
+	require.Len(t, pools, 1)
+	assert.Equal(t, "tank", pools[0].Name)
+}
+
+func TestParseZpoolListOutputNoPools(t *testing.T) {
+	pools, err := parseZpoolListOutput([]byte("no pools available\n"))
+	require.NoError(t, err)
+	assert.Empty(t, pools)
+}
+
+func TestParseZpoolListOutputRejectsMalformedLine(t *testing.T) {
+	_, err := parseZpoolListOutput([]byte("tank\t100\t50\n"))
+	require.Error(t, err)
+
+	_, err = parseZpoolListOutput([]byte("tank\tnotanumber\t50\t50\tONLINE\n"))
+	require.Error(t, err)
+}
+
+func TestParseZfsListOutput(t *testing.T) {
+	data, err := os.ReadFile(fixturePath("zfs_list.txt"))
+	require.NoError(t, err)
+
+	datasets, err := parseZfsListOutput(data)
+	require.NoError(t, err)
+	require.Len(t, datasets, 9)
+
+	// Mountpoint with a space must be kept intact (tab-split only).
+	assert.Equal(t, "/tank/my media", datasets[3].Mountpoint)
+	// Unmounted datasets/zvols report "-".
+	assert.Equal(t, "-", datasets[4].Mountpoint)
+	assert.Equal(t, uint64(12000000000000), datasets[0].Used)
+	assert.Equal(t, uint64(11999000000000), datasets[0].Avail)
+}
+
+func TestParseZpoolStatusOutput(t *testing.T) {
+	data, err := os.ReadFile(fixturePath("zpool_status.txt"))
+	require.NoError(t, err)
+
+	pools, err := parseZpoolStatusOutput(data)
+	require.NoError(t, err)
+	require.Len(t, pools, 2)
+
+	tank := pools[0]
+	assert.Equal(t, "tank", tank.Name)
+	assert.Equal(t, "ONLINE", tank.State)
+	assert.Equal(t, "FINISHED", tank.Scrub.State)
+	assert.Equal(t, "", tank.Scrub.Progress)
+	assert.Equal(t, uint64(0), tank.Scrub.Errors)
+	// Pool row itself is skipped; mirror + 2 disks remain.
+	require.Len(t, tank.Vdevs, 3)
+	assert.Equal(t, "mirror-0", tank.Vdevs[0].Name)
+	assert.Equal(t, "sda", tank.Vdevs[1].Name)
+	assert.Equal(t, "sdb", tank.Vdevs[2].Name)
+
+	rpool := pools[1]
+	assert.Equal(t, "rpool", rpool.Name)
+	assert.Equal(t, "DEGRADED", rpool.State)
+	assert.Equal(t, "SCANNING", rpool.Scrub.State)
+	assert.Equal(t, "10.00%", rpool.Scrub.Progress)
+	require.Len(t, rpool.Vdevs, 3)
+	assert.Equal(t, "FAULTED", rpool.Vdevs[2].State)
+	assert.Equal(t, uint64(1), rpool.Vdevs[2].ReadErrs)
+	assert.Equal(t, uint64(2), rpool.Vdevs[2].WriteErrs)
+	assert.Equal(t, uint64(3), rpool.Vdevs[2].ChecksumErrs)
+}
+
+func TestParseScanLine(t *testing.T) {
+	assert.Equal(t, "FINISHED", parseScanLine("scan: scrub repaired 0B in 00:05:12 with 0 errors on Sun Jun  1 02:00:12 2025").State)
+	assert.Equal(t, uint64(3), parseScanLine("scan: scrub repaired 10G in 01:00:00 with 3 errors on Sun Jun  1 02:00:12 2025").Errors)
+	assert.Equal(t, "SCANNING", parseScanLine("scan: scrub in progress since Sun Jun  8 01:00:00 2025").State)
+	assert.Equal(t, "CANCELED", parseScanLine("scan: scrub canceled on Sun Jun  1 02:00:12 2025").State)
+	assert.Equal(t, "FINISHED", parseScanLine("scan: resilvered 1.23G in 00:01:00 with 0 errors on Sun Jun  1 02:00:12 2025").State)
+	assert.Equal(t, "NONE", parseScanLine("scan: none requested").State)
+}
+
+func TestCommandOutputForcesLocaleAndTimesOut(t *testing.T) {
+	t.Setenv("BESZEL_ZFS_COMMAND_HELPER", "1")
+	out, err := commandOutput(os.Args[0], "-test.run=TestZfsCommandHelperProcess", "--", "locale")
+	require.NoError(t, err)
+	assert.Equal(t, "C/C", string(out))
+
+	oldTimeout := commandTimeout
+	commandTimeout = 20 * time.Millisecond
+	t.Cleanup(func() { commandTimeout = oldTimeout })
+	_, err = commandOutput(os.Args[0], "-test.run=TestZfsCommandHelperProcess", "--", "sleep")
+	require.Error(t, err)
+	assert.Contains(t, err.Error(), "timed out")
+}
+
+func TestZfsCommandHelperProcess(t *testing.T) {
+	if os.Getenv("BESZEL_ZFS_COMMAND_HELPER") != "1" {
+		return
+	}
+	mode := ""
+	for i, arg := range os.Args {
+		if arg == "--" && i+1 < len(os.Args) {
+			mode = os.Args[i+1]
+			break
+		}
+	}
+	switch strings.TrimSpace(mode) {
+	case "locale":
+		_, _ = fmt.Printf("%s/%s", os.Getenv("LC_ALL"), os.Getenv("LANG"))
+	case "sleep":
+		time.Sleep(time.Second)
+	}
+	os.Exit(0)
+}

agent/zfs/zfs_unsupported.go

@@ -7,3 +7,7 @@ import "errors"
 func ARCSize() (uint64, error) {
 	return 0, errors.ErrUnsupported
 }
+
+func PoolKernelStats() ([]PoolKernelStat, error) {
+	return nil, errors.ErrUnsupported
+}

agent/zfs_pool.go

@@ -0,0 +1,320 @@
+package agent
+
+import (
+	"log/slog"
+	"strings"
+	"sync"
+	"time"
+
+	"github.com/henrygd/beszel/agent/zfs"
+	"github.com/henrygd/beszel/internal/entities/system"
+	zfsentity "github.com/henrygd/beszel/internal/entities/zfs"
+)
+
+// zfsDatasetUsage holds usage values for a ZFS dataset mountpoint.
+type zfsDatasetUsage struct {
+	used  uint64
+	avail uint64
+}
+
+// datasetUsageRefreshInterval controls how often `zfs list` is re-run for the
+// mountpoint usage map. Dataset inventory changes rarely.
+const datasetUsageRefreshInterval = 5 * time.Minute
+
+// poolStatsRefreshInterval controls how often `zpool list` is re-run for pool
+// capacity. Health and I/O are read from procfs on Linux, so the utility only
+// needs to refresh slow-moving space accounting.
+const poolStatsRefreshInterval = time.Minute
+
+type poolKernelSample struct {
+	nread  uint64
+	nwrite uint64
+	at     time.Time
+}
+
+// ZfsManager collects ZFS pool and dataset statistics. Collection functions
+// are fields so unit tests can substitute them (same pattern as
+// diskDiscovery.usageFn). It is safe for concurrent use by a single goroutine
+// only; callers must hold the agent lock like updateDiskUsage does.
+type ZfsManager struct {
+	poolStatsFn    func() ([]zfs.PoolStat, error)       // capacity/health source
+	datasetsFn     func() ([]zfs.Dataset, error)        // dataset inventory source
+	kernelStatsFn  func() ([]zfs.PoolKernelStat, error) // procfs pool state/I/O source
+	poolStatusesFn func() ([]zfs.PoolStatus, error)     // scrub/vdev detail source
+
+	poolData      []zfs.PoolStat // cached pool inventory (TTL below)
+	lastPoolStats time.Time
+	kernelSamples map[string]poolKernelSample
+
+	datasetUsage     map[string]zfsDatasetUsage // mountpoint -> usage
+	lastUsageRefresh time.Time
+
+	// Detail data (pools, vdevs, scrub, datasets) is cached and refreshed on
+	// an interval. Accessed from handler goroutines, so it is mutex-protected.
+	detailMu          sync.Mutex
+	detail            *zfsentity.ZfsData
+	lastDetailRefresh time.Time
+	detailInterval    time.Duration
+}
+
+// newZfsManager creates a ZfsManager wired to the system's ZFS utilities.
+func newZfsManager() *ZfsManager {
+	return &ZfsManager{
+		poolStatsFn:    zfs.PoolStats,
+		datasetsFn:     zfs.Datasets,
+		kernelStatsFn:  zfs.PoolKernelStats,
+		poolStatusesFn: zfs.PoolStatuses,
+		detailInterval: time.Hour,
+	}
+}
+
+// Update refreshes systemStats.ZfsPools with the latest pool data. I/O
+// throughput and health come from inexpensive kernel kstats on Linux. Pool
+// capacity and dataset usage come from separately cached utility calls. It is
+// a no-op when ZFS is absent.
+func (zm *ZfsManager) Update(systemStats *system.Stats) {
+	pools := zm.poolStats()
+	if len(pools) == 0 {
+		return
+	}
+
+	kernelStats, ioRates := zm.kernelStats()
+
+	if systemStats.ZfsPools == nil {
+		systemStats.ZfsPools = make(map[string]*system.ZfsPool, len(pools))
+	}
+	for i := range pools {
+		pool := &pools[i]
+		// Full precision, matching the dataset values below; the frontend
+		// formats any magnitude.
+		stats := &system.ZfsPool{
+			Total:  float64(pool.Size) / (1024 * 1024 * 1024),
+			Used:   float64(pool.Alloc) / (1024 * 1024 * 1024),
+			Health: pool.Health,
+		}
+		if kernel, exists := kernelStats[pool.Name]; exists && kernel.Health != "" {
+			stats.Health = kernel.Health
+		}
+		if io, exists := ioRates[pool.Name]; exists {
+			stats.ReadBytes = io.NRead
+			stats.WriteBytes = io.NWrite
+		}
+		slog.Debug("ZFS pool sample", "pool", pool.Name, "health", stats.Health, "used_gb", stats.Used, "read_bps", stats.ReadBytes, "write_bps", stats.WriteBytes)
+		systemStats.ZfsPools[pool.Name] = stats
+	}
+
+}
+
+// poolStats returns the cached pool inventory, re-running `zpool list` at most
+// every poolStatsRefreshInterval. On failure the previous inventory is
+// retained and the refresh is retried on the next cadence.
+func (zm *ZfsManager) poolStats() []zfs.PoolStat {
+	if zm.lastPoolStats.IsZero() || time.Since(zm.lastPoolStats) >= poolStatsRefreshInterval {
+		pools, err := zm.poolStatsFn()
+		if err != nil {
+			slog.Debug("ZFS pool stats unavailable", "err", err)
+		} else {
+			zm.poolData = pools
+		}
+		zm.lastPoolStats = time.Now()
+	}
+	return zm.poolData
+}
+
+// kernelStats reads cumulative pool counters and converts them to per-second
+// rates. Counter decreases indicate a pool export/import and reset the
+// baseline instead of producing an underflow spike.
+func (zm *ZfsManager) kernelStats() (map[string]zfs.PoolKernelStat, map[string]zfs.PoolIoStats) {
+	if zm.kernelStatsFn == nil {
+		return nil, nil
+	}
+	stats, err := zm.kernelStatsFn()
+	if err != nil {
+		slog.Debug("ZFS kernel stats unavailable", "err", err)
+		return nil, nil
+	}
+	now := time.Now()
+	byName := make(map[string]zfs.PoolKernelStat, len(stats))
+	rates := make(map[string]zfs.PoolIoStats, len(stats))
+	nextSamples := make(map[string]poolKernelSample, len(stats))
+	for _, stat := range stats {
+		byName[stat.Name] = stat
+		if previous, ok := zm.kernelSamples[stat.Name]; ok && now.After(previous.at) &&
+			stat.NRead >= previous.nread && stat.NWrite >= previous.nwrite {
+			seconds := now.Sub(previous.at).Seconds()
+			rates[stat.Name] = zfs.PoolIoStats{
+				NRead:  uint64(float64(stat.NRead-previous.nread) / seconds),
+				NWrite: uint64(float64(stat.NWrite-previous.nwrite) / seconds),
+			}
+		}
+		nextSamples[stat.Name] = poolKernelSample{nread: stat.NRead, nwrite: stat.NWrite, at: now}
+	}
+	zm.kernelSamples = nextSamples
+	return byName, rates
+}
+
+// refreshDatasetUsage re-runs `zfs list` when the refresh window has elapsed
+// and rebuilds the mountpoint-keyed usage map.
+func (zm *ZfsManager) refreshDatasetUsage() {
+	if !zm.lastUsageRefresh.IsZero() && time.Since(zm.lastUsageRefresh) < datasetUsageRefreshInterval {
+		return
+	}
+	datasets, err := zm.datasetsFn()
+	if err != nil {
+		slog.Debug("ZFS dataset usage unavailable", "err", err)
+	} else {
+		usage := make(map[string]zfsDatasetUsage, len(datasets))
+		for _, ds := range datasets {
+			if ds.Mountpoint != "" && ds.Mountpoint != "-" {
+				usage[ds.Mountpoint] = zfsDatasetUsage{used: ds.Used, avail: ds.Avail}
+			}
+		}
+		zm.datasetUsage = usage
+	}
+	zm.lastUsageRefresh = time.Now()
+}
+
+// DatasetUsage returns ZFS dataset usage keyed by mountpoint, refreshed at
+// most every datasetUsageRefreshInterval. On failure the previous map is
+// retained and a debug log is emitted.
+func (zm *ZfsManager) DatasetUsage() map[string]zfsDatasetUsage {
+	zm.refreshDatasetUsage()
+	return zm.datasetUsage
+}
+
+// GetDetail returns ZFS detail data (pool health, scrub, vdevs, datasets).
+// Scheduled requests use the cached snapshot until stale; manual requests can
+// force collection. On failure the previous snapshot is retained.
+func (zm *ZfsManager) GetDetail(force bool) *zfsentity.ZfsData {
+	zm.detailMu.Lock()
+	defer zm.detailMu.Unlock()
+
+	if force || zm.detail == nil || time.Since(zm.lastDetailRefresh) >= zm.detailInterval {
+		if data, err := zm.collectDetail(zm.detail); err != nil {
+			slog.Debug("ZFS detail collection failed", "err", err)
+			if zm.detail == nil {
+				return &zfsentity.ZfsData{}
+			}
+			return &zfsentity.ZfsData{Pools: zm.detail.Pools}
+		} else {
+			zm.detail = data
+			zm.lastDetailRefresh = time.Now()
+		}
+	}
+	if zm.detail == nil {
+		return &zfsentity.ZfsData{}
+	}
+	return zm.detail
+}
+
+// collectDetail builds a ZfsData payload from the current system state.
+func (zm *ZfsManager) collectDetail(previous *zfsentity.ZfsData) (*zfsentity.ZfsData, error) {
+	pools, err := zm.poolStatsFn()
+	if err != nil {
+		return nil, err
+	}
+	if len(pools) == 0 {
+		return &zfsentity.ZfsData{Pools: []*zfsentity.PoolDetail{}, Complete: true}, nil
+	}
+
+	statuses, statusErr := zm.poolStatusesFn()
+	if statusErr != nil {
+		slog.Debug("ZFS pool status unavailable", "err", statusErr)
+	}
+	datasets, datasetsErr := zm.datasetsFn()
+	if datasetsErr != nil {
+		slog.Debug("ZFS datasets unavailable", "err", datasetsErr)
+	}
+
+	statusByPool := make(map[string]zfs.PoolStatus, len(statuses))
+	for _, st := range statuses {
+		statusByPool[st.Name] = st
+	}
+
+	previousByPool := make(map[string]*zfsentity.PoolDetail)
+	if previous != nil {
+		for _, pool := range previous.Pools {
+			if pool != nil {
+				previousByPool[pool.Name] = pool
+			}
+		}
+	}
+
+	data := &zfsentity.ZfsData{Pools: make([]*zfsentity.PoolDetail, 0, len(pools)), Complete: true}
+	for i := range pools {
+		p := &pools[i]
+		detail := &zfsentity.PoolDetail{
+			Name:   p.Name,
+			Health: p.Health,
+			Size:   p.Size,
+			Alloc:  p.Alloc,
+			Free:   p.Free,
+		}
+		if st, ok := statusByPool[p.Name]; statusErr == nil && ok {
+			if st.Scrub.State != "" && st.Scrub.State != "NONE" {
+				detail.Scrub = &zfsentity.Scrub{
+					State:    st.Scrub.State,
+					Progress: st.Scrub.Progress,
+					Errors:   st.Scrub.Errors,
+				}
+			}
+			for _, v := range st.Vdevs {
+				detail.Vdevs = append(detail.Vdevs, &zfsentity.Vdev{
+					Name:         v.Name,
+					State:        v.State,
+					ReadErrs:     v.ReadErrs,
+					WriteErrs:    v.WriteErrs,
+					ChecksumErrs: v.ChecksumErrs,
+				})
+			}
+		} else {
+			if cached := previousByPool[p.Name]; cached != nil {
+				detail.Scrub = cached.Scrub
+				detail.Vdevs = cached.Vdevs
+			}
+		}
+		if datasetsErr == nil {
+			foundDataset := false
+			for _, ds := range datasets {
+				if poolOfDataset(ds.Name) == p.Name {
+					foundDataset = true
+					detail.Datasets = append(detail.Datasets, &zfsentity.Dataset{
+						Name:       ds.Name,
+						Used:       ds.Used,
+						Avail:      ds.Avail,
+						Mountpoint: ds.Mountpoint,
+					})
+				}
+			}
+			if !foundDataset {
+				if cached := previousByPool[p.Name]; cached != nil {
+					detail.Datasets = cached.Datasets
+				}
+			}
+		} else if cached := previousByPool[p.Name]; cached != nil {
+			detail.Datasets = cached.Datasets
+		}
+		data.Pools = append(data.Pools, detail)
+	}
+	return data, nil
+}
+
+// poolOfDataset returns the pool name for a dataset name (everything before
+// the first '/'). Datasets without a separator belong to a pool of the same
+// name.
+func poolOfDataset(name string) string {
+	if idx := strings.IndexByte(name, '/'); idx >= 0 {
+		return name[:idx]
+	}
+	return name
+}
+
+// ZfsMountpoints returns the set of mountpoints backed by ZFS datasets.
+func (zm *ZfsManager) ZfsMountpoints() map[string]bool {
+	usage := zm.DatasetUsage()
+	mountpoints := make(map[string]bool, len(usage))
+	for mountpoint := range usage {
+		mountpoints[mountpoint] = true
+	}
+	return mountpoints
+}

agent/zfs_pool_test.go

@@ -0,0 +1,245 @@
+//go:build testing
+
+package agent
+
+import (
+	"testing"
+	"time"
+
+	"github.com/henrygd/beszel/agent/zfs"
+	"github.com/henrygd/beszel/internal/entities/system"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestUpdatePopulatesZfsPools(t *testing.T) {
+	zm := &ZfsManager{}
+	zm.poolStatsFn = func() ([]zfs.PoolStat, error) {
+		return []zfs.PoolStat{{Name: "tank", Size: 23999000000000, Alloc: 12000000000000, Free: 11999000000000, Health: "DEGRADED"}}, nil
+	}
+	zm.datasetsFn = func() ([]zfs.Dataset, error) {
+		return []zfs.Dataset{
+			{Name: "tank/apps", Used: 5000000000000, Avail: 11999000000000, Mountpoint: "/tank/apps"},
+			{Name: "tank/backup", Used: 6000000000000, Avail: 11999000000000, Mountpoint: "/tank/backup"},
+			// Small zvol (Proxmox VM EFI disk): must not round to zero.
+			{Name: "rpool/vm-100-disk-2", Used: 4194304, Avail: 0, Mountpoint: "-"},
+		}, nil
+	}
+	var kernelCalls int
+	zm.kernelStatsFn = func() ([]zfs.PoolKernelStat, error) {
+		kernelCalls++
+		return []zfs.PoolKernelStat{{
+			Name: "tank", Health: "ONLINE",
+			NRead: uint64(kernelCalls-1) * 1250, NWrite: uint64(kernelCalls-1) * 5120,
+		}}, nil
+	}
+
+	var stats system.Stats
+	// The first kernel sample establishes the cumulative-counter baseline.
+	zm.Update(&stats)
+	zm.kernelSamples["tank"] = poolKernelSample{at: time.Now().Add(-time.Second)}
+	zm.Update(&stats)
+	require.NotNil(t, stats.ZfsPools)
+	require.Contains(t, stats.ZfsPools, "tank")
+	assert.InDelta(t, 22350.8105, stats.ZfsPools["tank"].Total, 0.0001) // Size in GiB
+	assert.InDelta(t, 11175.8709, stats.ZfsPools["tank"].Used, 0.0001)  // Alloc in GiB
+	assert.Equal(t, "ONLINE", stats.ZfsPools["tank"].Health)
+	assert.InDelta(t, 1250, stats.ZfsPools["tank"].ReadBytes, 5)
+	assert.InDelta(t, 5120, stats.ZfsPools["tank"].WriteBytes, 5)
+
+}
+
+// TestUpdateKernelStatsMissing verifies pools without a kernel sample report zero
+// I/O instead of erroring.
+func TestUpdateKernelStatsMissing(t *testing.T) {
+	zm := &ZfsManager{}
+	zm.poolStatsFn = func() ([]zfs.PoolStat, error) {
+		return []zfs.PoolStat{{Name: "tank", Size: 1, Alloc: 1, Health: "ONLINE"}}, nil
+	}
+	zm.datasetsFn = func() ([]zfs.Dataset, error) { return nil, nil }
+	zm.kernelStatsFn = func() ([]zfs.PoolKernelStat, error) {
+		return nil, zfs.ErrNoZfs
+	}
+
+	var stats system.Stats
+	zm.Update(&stats)
+	require.NotNil(t, stats.ZfsPools)
+	assert.Equal(t, uint64(0), stats.ZfsPools["tank"].ReadBytes)
+	assert.Equal(t, uint64(0), stats.ZfsPools["tank"].WriteBytes)
+}
+
+func TestUpdateKernelCounterReset(t *testing.T) {
+	zm := &ZfsManager{}
+	zm.poolStatsFn = func() ([]zfs.PoolStat, error) {
+		return []zfs.PoolStat{{Name: "tank", Health: "ONLINE"}}, nil
+	}
+	zm.datasetsFn = func() ([]zfs.Dataset, error) { return nil, nil }
+	zm.kernelSamples = map[string]poolKernelSample{
+		"tank": {nread: 100, nwrite: 200, at: time.Now().Add(-time.Second)},
+	}
+	zm.kernelStatsFn = func() ([]zfs.PoolKernelStat, error) {
+		return []zfs.PoolKernelStat{{Name: "tank", Health: "ONLINE", NRead: 10, NWrite: 20}}, nil
+	}
+
+	var stats system.Stats
+	zm.Update(&stats)
+	assert.Equal(t, uint64(0), stats.ZfsPools["tank"].ReadBytes)
+	assert.Equal(t, uint64(0), stats.ZfsPools["tank"].WriteBytes)
+}
+
+func TestUpdateNoZfs(t *testing.T) {
+	zm := &ZfsManager{}
+	calls := 0
+	zm.poolStatsFn = func() ([]zfs.PoolStat, error) {
+		calls++
+		return nil, zfs.ErrNoZfs
+	}
+
+	var stats system.Stats
+	zm.Update(&stats)
+	zm.Update(&stats)
+	assert.Nil(t, stats.ZfsPools)
+	assert.Equal(t, 1, calls, "failed pool discovery should be cached until the next refresh interval")
+}
+
+func TestUpdateEmptyPools(t *testing.T) {
+	zm := &ZfsManager{}
+	calls := 0
+	zm.poolStatsFn = func() ([]zfs.PoolStat, error) {
+		calls++
+		return nil, nil
+	}
+
+	var stats system.Stats
+	zm.Update(&stats)
+	zm.Update(&stats)
+	assert.Nil(t, stats.ZfsPools)
+	assert.Equal(t, 1, calls, "an empty pool inventory should be cached until the next refresh interval")
+}
+
+func TestDatasetUsage(t *testing.T) {
+	zm := &ZfsManager{}
+	calls := 0
+	zm.datasetsFn = func() ([]zfs.Dataset, error) {
+		calls++
+		return []zfs.Dataset{
+			{Name: "tank", Used: 12000000000000, Avail: 11999000000000, Mountpoint: "/tank"},
+			{Name: "tank/apps", Used: 1000000000000, Avail: 11999000000000, Mountpoint: "/tank/apps"},
+			{Name: "rpool", Used: 900000000000, Avail: 300000000000, Mountpoint: "-"}, // zvol/unmounted: excluded
+		}, nil
+	}
+
+	usage := zm.DatasetUsage()
+	require.Len(t, usage, 2)
+	assert.Equal(t, zfsDatasetUsage{used: 12000000000000, avail: 11999000000000}, usage["/tank"])
+	assert.Equal(t, zfsDatasetUsage{used: 1000000000000, avail: 11999000000000}, usage["/tank/apps"])
+	assert.Equal(t, 1, calls)
+
+	// Second call within the refresh window must not re-run the collector.
+	zm.DatasetUsage()
+	assert.Equal(t, 1, calls)
+}
+
+func TestDatasetUsageRefreshOnErrorKeepsPrevious(t *testing.T) {
+	zm := &ZfsManager{}
+	zm.datasetsFn = func() ([]zfs.Dataset, error) {
+		return []zfs.Dataset{{Name: "tank", Used: 1, Avail: 1, Mountpoint: "/tank"}}, nil
+	}
+	assert.Len(t, zm.DatasetUsage(), 1)
+
+	// Force refresh window expiry, then a failing collector.
+	zm.lastUsageRefresh = time.Now().Add(-10 * time.Minute)
+	zm.datasetsFn = func() ([]zfs.Dataset, error) {
+		return nil, zfs.ErrNoZfs
+	}
+	usage := zm.DatasetUsage()
+	assert.Len(t, usage, 1, "previous usage should be retained on error")
+}
+
+func TestGetDetailForceRefresh(t *testing.T) {
+	zm := &ZfsManager{detailInterval: time.Hour}
+	poolCalls := 0
+	zm.poolStatsFn = func() ([]zfs.PoolStat, error) {
+		poolCalls++
+		return []zfs.PoolStat{{Name: "tank", Alloc: uint64(poolCalls)}}, nil
+	}
+	zm.poolStatusesFn = func() ([]zfs.PoolStatus, error) { return nil, nil }
+	zm.datasetsFn = func() ([]zfs.Dataset, error) { return nil, nil }
+
+	first := zm.GetDetail(false)
+	assert.True(t, first.Complete)
+	require.Len(t, first.Pools, 1)
+	assert.Equal(t, uint64(1), first.Pools[0].Alloc)
+
+	cached := zm.GetDetail(false)
+	require.Len(t, cached.Pools, 1)
+	assert.Equal(t, uint64(1), cached.Pools[0].Alloc)
+	assert.Equal(t, 1, poolCalls)
+
+	refreshed := zm.GetDetail(true)
+	assert.True(t, refreshed.Complete)
+	require.Len(t, refreshed.Pools, 1)
+	assert.Equal(t, uint64(2), refreshed.Pools[0].Alloc)
+	assert.Equal(t, 2, poolCalls)
+}
+
+func TestGetDetailSuccessfulEmptyInventoryClearsCache(t *testing.T) {
+	zm := &ZfsManager{detailInterval: time.Hour}
+	zm.poolStatsFn = func() ([]zfs.PoolStat, error) {
+		return []zfs.PoolStat{{Name: "tank"}}, nil
+	}
+	zm.poolStatusesFn = func() ([]zfs.PoolStatus, error) { return nil, nil }
+	zm.datasetsFn = func() ([]zfs.Dataset, error) { return nil, nil }
+
+	require.Len(t, zm.GetDetail(false).Pools, 1)
+	zm.poolStatsFn = func() ([]zfs.PoolStat, error) { return nil, nil }
+	empty := zm.GetDetail(true)
+	assert.True(t, empty.Complete)
+	assert.Empty(t, empty.Pools)
+}
+
+func TestGetDetailFailureReturnsIncompleteCachedInventory(t *testing.T) {
+	zm := &ZfsManager{detailInterval: time.Hour}
+	zm.poolStatsFn = func() ([]zfs.PoolStat, error) {
+		return []zfs.PoolStat{{Name: "tank"}}, nil
+	}
+	zm.poolStatusesFn = func() ([]zfs.PoolStatus, error) {
+		return []zfs.PoolStatus{{Name: "tank", Vdevs: []zfs.VdevStatus{{Name: "mirror-0"}}}}, nil
+	}
+	zm.datasetsFn = func() ([]zfs.Dataset, error) {
+		return []zfs.Dataset{{Name: "tank/data"}}, nil
+	}
+	first := zm.GetDetail(false)
+	require.True(t, first.Complete)
+	require.Len(t, first.Pools[0].Vdevs, 1)
+	require.Len(t, first.Pools[0].Datasets, 1)
+
+	zm.poolStatusesFn = func() ([]zfs.PoolStatus, error) { return nil, zfs.ErrNoZfs }
+	zm.datasetsFn = func() ([]zfs.Dataset, error) { return nil, zfs.ErrNoZfs }
+	partial := zm.GetDetail(true)
+	require.True(t, partial.Complete)
+	require.Len(t, partial.Pools[0].Vdevs, 1)
+	require.Len(t, partial.Pools[0].Datasets, 1)
+
+	zm.poolStatsFn = func() ([]zfs.PoolStat, error) { return nil, zfs.ErrNoZfs }
+	lastSuccessfulRefresh := zm.lastDetailRefresh
+	failed := zm.GetDetail(true)
+	assert.False(t, failed.Complete)
+	require.Len(t, failed.Pools, 1)
+	assert.Equal(t, "tank", failed.Pools[0].Name)
+	assert.Equal(t, lastSuccessfulRefresh, zm.lastDetailRefresh)
+}
+
+func TestZfsMountpoints(t *testing.T) {
+	zm := &ZfsManager{}
+	zm.datasetsFn = func() ([]zfs.Dataset, error) {
+		return []zfs.Dataset{
+			{Name: "tank", Mountpoint: "/tank"},
+			{Name: "rpool/ROOT/pve-1", Mountpoint: "/"},
+		}, nil
+	}
+	mountpoints := zm.ZfsMountpoints()
+	assert.Len(t, mountpoints, 2)
+	assert.True(t, mountpoints["/tank"])
+	assert.True(t, mountpoints["/"])
+}

beszel.go

@@ -6,7 +6,7 @@ import "github.com/blang/semver"
 
 const (
 	// Version is the current version of the application.
-	Version = "0.18.5"
+	Version = "0.19.0"
 	// AppName is the name of the application.
 	AppName = "beszel"
 )
@@ -16,3 +16,6 @@ var MinVersionCbor = semver.MustParse("0.12.0")
 
 // MinVersionAgentResponse is the minimum supported version for AgentResponse compatibility.
 var MinVersionAgentResponse = semver.MustParse("0.13.0")
+
+// MinVersionZfsData is the minimum agent version that supports ZFS detail requests.
+var MinVersionZfsData = semver.MustParse("0.18.9")

go.mod

@@ -1,69 +1,68 @@
 module github.com/henrygd/beszel
 
-go 1.26.1
+go 1.27.1
 
 require (
 	github.com/blang/semver v3.5.1+incompatible
 	github.com/coreos/go-systemd/v22 v22.7.0
-	github.com/distatus/battery v0.11.0
-	github.com/ebitengine/purego v0.10.0
-	github.com/fxamacker/cbor/v2 v2.9.0
+	github.com/ebitengine/purego v0.11.0
+	github.com/fxamacker/cbor/v2 v2.9.3
 	github.com/gliderlabs/ssh v0.3.8
 	github.com/google/uuid v1.6.0
-	github.com/lxzan/gws v1.9.1
-	github.com/nicholas-fedor/shoutrrr v0.14.1
+	github.com/lxzan/gws v1.10.1
+	github.com/nicholas-fedor/shoutrrr v0.19.0
 	github.com/pocketbase/dbx v1.12.0
-	github.com/pocketbase/pocketbase v0.36.7
-	github.com/shirou/gopsutil/v4 v4.26.2
+	github.com/pocketbase/pocketbase v0.40.2
+	github.com/shirou/gopsutil/v4 v4.26.8
 	github.com/spf13/cast v1.10.0
 	github.com/spf13/cobra v1.10.2
 	github.com/spf13/pflag v1.0.10
-	github.com/stretchr/testify v1.11.1
-	golang.org/x/crypto v0.49.0
-	golang.org/x/exp v0.0.0-20260312153236-7ab1446f8b90
-	golang.org/x/sys v0.42.0
+	github.com/stretchr/testify v1.12.1
+	golang.org/x/crypto v0.56.0
+	golang.org/x/exp v0.0.0-20260824195058-e88cd73687aa
+	golang.org/x/net v0.58.0
+	golang.org/x/sys v0.47.0
 	gopkg.in/yaml.v3 v3.0.1
+	howett.net/plist v1.0.1
 )
 
 require (
 	github.com/anmitsu/go-shlex v0.0.0-20200514113438-38f4b401e2be // indirect
 	github.com/asaskevich/govalidator v0.0.0-20230301143203-a9d515a09cc2 // indirect
-	github.com/davecgh/go-spew v1.1.2-0.20180830191138-d8f796af33cc // indirect
 	github.com/disintegration/imaging v1.6.2 // indirect
 	github.com/domodwyer/mailyak/v3 v3.6.2 // indirect
 	github.com/dustin/go-humanize v1.0.1 // indirect
 	github.com/eclipse/paho.golang v0.23.0 // indirect
 	github.com/fatih/color v1.19.0 // indirect
-	github.com/gabriel-vasile/mimetype v1.4.13 // indirect
-	github.com/ganigeorgiev/fexpr v0.5.0 // indirect
+	github.com/fsnotify/fsnotify v1.10.1 // indirect
+	github.com/gabriel-vasile/mimetype v1.4.15 // indirect
+	github.com/ganigeorgiev/fexpr v0.6.0 // indirect
 	github.com/go-ole/go-ole v1.3.0 // indirect
-	github.com/go-ozzo/ozzo-validation/v4 v4.3.0 // indirect
 	github.com/go-sql-driver/mysql v1.9.1 // indirect
 	github.com/godbus/dbus/v5 v5.2.2 // indirect
 	github.com/golang-jwt/jwt/v5 v5.3.1 // indirect
 	github.com/gorilla/websocket v1.5.3 // indirect
 	github.com/inconshreveable/mousetrap v1.1.0 // indirect
-	github.com/klauspost/compress v1.18.5 // indirect
-	github.com/lufia/plan9stats v0.0.0-20260324052639-156f7da3f749 // indirect
-	github.com/mattn/go-colorable v0.1.14 // indirect
-	github.com/mattn/go-isatty v0.0.20 // indirect
+	github.com/klauspost/compress v1.20.0 // indirect
+	github.com/lufia/plan9stats v0.0.0-20260802145828-341c2f0c90b5 // indirect
+	github.com/mattn/go-colorable v0.1.15 // indirect
+	github.com/mattn/go-isatty v0.0.24 // indirect
 	github.com/ncruces/go-strftime v1.0.0 // indirect
-	github.com/pmezard/go-difflib v1.0.1-0.20181226105442-5d4384ee4fb2 // indirect
-	github.com/power-devops/perfstat v0.0.0-20240221224432-82ca36839d55 // indirect
+	github.com/pocketbase/ozzo-validation/v4 v4.3.0 // indirect
+	github.com/power-devops/perfstat v0.0.0-20260805114148-88456608a4f6 // indirect
 	github.com/remyoudompheng/bigfft v0.0.0-20230129092748-24d4a6f8daec // indirect
-	github.com/tklauser/go-sysconf v0.3.16 // indirect
-	github.com/tklauser/numcpus v0.11.0 // indirect
+	github.com/tklauser/go-sysconf v0.4.0 // indirect
+	github.com/tklauser/numcpus v0.12.0 // indirect
 	github.com/x448/float16 v0.8.4 // indirect
 	github.com/yusufpapurcu/wmi v1.2.4 // indirect
-	golang.org/x/image v0.38.0 // indirect
-	golang.org/x/net v0.52.0 // indirect
+	go.yaml.in/yaml/v3 v3.0.5 // indirect
+	golang.org/x/image v0.45.0 // indirect
 	golang.org/x/oauth2 v0.36.0 // indirect
-	golang.org/x/sync v0.20.0 // indirect
-	golang.org/x/term v0.41.0 // indirect
-	golang.org/x/text v0.35.0 // indirect
-	howett.net/plist v1.0.1 // indirect
-	modernc.org/libc v1.70.0 // indirect
+	golang.org/x/sync v0.22.0 // indirect
+	golang.org/x/term v0.45.0 // indirect
+	golang.org/x/text v0.41.0 // indirect
+	modernc.org/libc v1.74.4 // indirect
 	modernc.org/mathutil v1.7.1 // indirect
-	modernc.org/memory v1.11.0 // indirect
-	modernc.org/sqlite v1.46.2 // indirect
+	modernc.org/memory v1.12.1 // indirect
+	modernc.org/sqlite v1.57.0 // indirect
 )

go.sum

@@ -1,7 +1,7 @@
 filippo.io/edwards25519 v1.1.0 h1:FNf4tywRC1HmFuKW5xopWpigGjJKiJSV0Cqo0cJWDaA=
 filippo.io/edwards25519 v1.1.0/go.mod h1:BxyFTGdWcka3PhytdK4V28tE5sGfRvvvRV7EaN4VDT4=
-github.com/Masterminds/semver/v3 v3.4.0 h1:Zog+i5UMtVoCU8oKka5P7i9q9HgrJeGzI9SA1Xbatp0=
-github.com/Masterminds/semver/v3 v3.4.0/go.mod h1:4V+yj/TJE1HU9XfppCwVMZq3I84lprf4nC11bSS5beM=
+github.com/Masterminds/semver/v3 v3.5.0 h1:kQceYJfbupGfZOKZQg0kou0DgAKhzDg2NZPAwZ/2OOE=
+github.com/Masterminds/semver/v3 v3.5.0/go.mod h1:4V+yj/TJE1HU9XfppCwVMZq3I84lprf4nC11bSS5beM=
 github.com/anmitsu/go-shlex v0.0.0-20200514113438-38f4b401e2be h1:9AeTilPcZAjCFIImctFaOjnTIavg87rW78vTPkQqLI8=
 github.com/anmitsu/go-shlex v0.0.0-20200514113438-38f4b401e2be/go.mod h1:ySMOLuWl6zY27l47sB3qLNK6tF2fkHG55UZxx8oIVo4=
 github.com/asaskevich/govalidator v0.0.0-20200108200545-475eaeb16496/go.mod h1:oGkLhpf+kjZl6xBf758TQhh5XrAeiJv/7FRz/2spLIg=
@@ -13,39 +13,35 @@ github.com/coreos/go-systemd/v22 v22.7.0 h1:LAEzFkke61DFROc7zNLX/WA2i5J8gYqe0rSj
 github.com/coreos/go-systemd/v22 v22.7.0/go.mod h1:xNUYtjHu2EDXbsxz1i41wouACIwT7Ybq9o0BQhMwD0w=
 github.com/cpuguy83/go-md2man/v2 v2.0.6/go.mod h1:oOW0eioCTA6cOiMLiUPZOpcVxMig6NIQQ7OS05n1F4g=
 github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
-github.com/davecgh/go-spew v1.1.2-0.20180830191138-d8f796af33cc h1:U9qPSI2PIWSS1VwoXQT9A3Wy9MM3WgvqSxFWenqJduM=
-github.com/davecgh/go-spew v1.1.2-0.20180830191138-d8f796af33cc/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
 github.com/disintegration/imaging v1.6.2 h1:w1LecBlG2Lnp8B3jk5zSuNqd7b4DXhcjwek1ei82L+c=
 github.com/disintegration/imaging v1.6.2/go.mod h1:44/5580QXChDfwIclfc/PCwrr44amcmDAg8hxG0Ewe4=
-github.com/distatus/battery v0.11.0 h1:KJk89gz90Iq/wJtbjjM9yUzBXV+ASV/EG2WOOL7N8lc=
-github.com/distatus/battery v0.11.0/go.mod h1:KmVkE8A8hpIX4T78QRdMktYpEp35QfOL8A8dwZBxq2k=
 github.com/domodwyer/mailyak/v3 v3.6.2 h1:x3tGMsyFhTCaxp6ycgR0FE/bu5QiNp+hetUuCOBXMn8=
 github.com/domodwyer/mailyak/v3 v3.6.2/go.mod h1:lOm/u9CyCVWHeaAmHIdF4RiKVxKUT/H5XX10lIKAL6c=
 github.com/dustin/go-humanize v1.0.1 h1:GzkhY7T5VNhEkwH0PVJgjz+fX1rhBrR7pRT3mDkpeCY=
 github.com/dustin/go-humanize v1.0.1/go.mod h1:Mu1zIs6XwVuF/gI1OepvI0qD18qycQx+mFykh5fBlto=
-github.com/ebitengine/purego v0.10.0 h1:QIw4xfpWT6GWTzaW5XEKy3HXoqrJGx1ijYHzTF0/ISU=
-github.com/ebitengine/purego v0.10.0/go.mod h1:iIjxzd6CiRiOG0UyXP+V1+jWqUXVjPKLAI0mRfJZTmQ=
+github.com/ebitengine/purego v0.11.0 h1:jhp/D+Nyv7UUW8HAcmcjt2N2rYrYi9m3SL21k0Ua/NI=
+github.com/ebitengine/purego v0.11.0/go.mod h1:DCHPP08djqhNSoTfImcnHYQRZmd0qhakvrozqaEYhGQ=
 github.com/eclipse/paho.golang v0.23.0 h1:KHgl2wz6EJo7cMBmkuhpt7C576vP+kpPv7jjvSyR6Mk=
 github.com/eclipse/paho.golang v0.23.0/go.mod h1:nQRhTkoZv8EAiNs5UU0/WdQIx2NrnWUpL9nsGJTQN04=
 github.com/fatih/color v1.19.0 h1:Zp3PiM21/9Ld6FzSKyL5c/BULoe/ONr9KlbYVOfG8+w=
 github.com/fatih/color v1.19.0/go.mod h1:zNk67I0ZUT1bEGsSGyCZYZNrHuTkJJB+r6Q9VuMi0LE=
 github.com/frankban/quicktest v1.14.6 h1:7Xjx+VpznH+oBnejlPUj8oUpdxnVs4f8XU8WnHkI4W8=
 github.com/frankban/quicktest v1.14.6/go.mod h1:4ptaffx2x8+WTWXmUCuVU6aPUX1/Mz7zb5vbUoiM6w0=
-github.com/fxamacker/cbor/v2 v2.9.0 h1:NpKPmjDBgUfBms6tr6JZkTHtfFGcMKsw3eGcmD/sapM=
-github.com/fxamacker/cbor/v2 v2.9.0/go.mod h1:vM4b+DJCtHn+zz7h3FFp/hDAI9WNWCsZj23V5ytsSxQ=
-github.com/gabriel-vasile/mimetype v1.4.13 h1:46nXokslUBsAJE/wMsp5gtO500a4F3Nkz9Ufpk2AcUM=
-github.com/gabriel-vasile/mimetype v1.4.13/go.mod h1:d+9Oxyo1wTzWdyVUPMmXFvp4F9tea18J8ufA774AB3s=
-github.com/ganigeorgiev/fexpr v0.5.0 h1:XA9JxtTE/Xm+g/JFI6RfZEHSiQlk+1glLvRK1Lpv/Tk=
-github.com/ganigeorgiev/fexpr v0.5.0/go.mod h1:RyGiGqmeXhEQ6+mlGdnUleLHgtzzu/VGO2WtJkF5drE=
+github.com/fsnotify/fsnotify v1.10.1 h1:b0/UzAf9yR5rhf3RPm9gf3ehBPpf0oZKIjtpKrx59Ho=
+github.com/fsnotify/fsnotify v1.10.1/go.mod h1:TLheqan6HD6GBK6PrDWyDPBaEV8LspOxvPSjC+bVfgo=
+github.com/fxamacker/cbor/v2 v2.9.3 h1:oQBnFATpNdY8gJHTndDDv5Xl4QqNaz51G5LLEPhng3Q=
+github.com/fxamacker/cbor/v2 v2.9.3/go.mod h1:vM4b+DJCtHn+zz7h3FFp/hDAI9WNWCsZj23V5ytsSxQ=
+github.com/gabriel-vasile/mimetype v1.4.15 h1:05iP/CYtZ/w455R/KZM6rZ5ieAdh99UPtd+d3YzLmaI=
+github.com/gabriel-vasile/mimetype v1.4.15/go.mod h1:azpTcoLcDZRNgFou5j+APrqQx9HqVPWa6ijYQIIVswQ=
+github.com/ganigeorgiev/fexpr v0.6.0 h1:Fza3O/QMBKEudUvxV862qe6GjxM60GJjjKytdp+VQus=
+github.com/ganigeorgiev/fexpr v0.6.0/go.mod h1:RyGiGqmeXhEQ6+mlGdnUleLHgtzzu/VGO2WtJkF5drE=
 github.com/gliderlabs/ssh v0.3.8 h1:a4YXD1V7xMF9g5nTkdfnja3Sxy1PVDCj1Zg4Wb8vY6c=
 github.com/gliderlabs/ssh v0.3.8/go.mod h1:xYoytBv1sV0aL3CavoDuJIQNURXkkfPA/wxQ1pL1fAU=
-github.com/go-logr/logr v1.4.3 h1:CjnDlHq8ikf6E492q6eKboGOC0T8CDaOvkHCIg8idEI=
-github.com/go-logr/logr v1.4.3/go.mod h1:9T104GzyrTigFIr8wt5mBrctHMim0Nb2HLGrmQ40KvY=
+github.com/go-logr/logr v1.4.4 h1:tG4xh9yMsRCAiodLVTxyrkzSZ9+o0L1Kg/+cPVcbP/8=
+github.com/go-logr/logr v1.4.4/go.mod h1:9T104GzyrTigFIr8wt5mBrctHMim0Nb2HLGrmQ40KvY=
 github.com/go-ole/go-ole v1.2.6/go.mod h1:pprOEPIfldk/42T2oK7lQ4v4JSDwmV0As9GaiUsvbm0=
 github.com/go-ole/go-ole v1.3.0 h1:Dt6ye7+vXGIKZ7Xtk4s6/xVdGDQynvom7xCFEdWr6uE=
 github.com/go-ole/go-ole v1.3.0/go.mod h1:5LS6F96DhAwUc7C+1HLexzMXY1xGRSryjyPPKW6zv78=
-github.com/go-ozzo/ozzo-validation/v4 v4.3.0 h1:byhDUpfEwjsVQb1vBunvIjh2BHQ9ead57VkAEY4V+Es=
-github.com/go-ozzo/ozzo-validation/v4 v4.3.0/go.mod h1:2NKgrcHl3z6cJs+3Oo940FPRiTzuqKbvfrL2RxCj6Ew=
 github.com/go-sql-driver/mysql v1.4.1/go.mod h1:zAC/RDZ24gD3HViQzih4MyKcchzm+sOG5ZlKdlhCg5w=
 github.com/go-sql-driver/mysql v1.9.1 h1:FrjNGn/BsJQjVRuSa8CBrM5BWA9BWoXXat3KrtSb/iI=
 github.com/go-sql-driver/mysql v1.9.1/go.mod h1:qn46aNg1333BRMNU69Lq93t8du/dwxI64Gl8i5p1WMU=
@@ -58,8 +54,8 @@ github.com/golang-jwt/jwt/v5 v5.3.1/go.mod h1:fxCRLWMO43lRc8nhHWY6LGqRcf+1gQWArs
 github.com/golang/protobuf v1.3.1/go.mod h1:6lQm79b+lXiMfvg/cZm0SGofjICqVBUtrP5yJMmIC1U=
 github.com/google/go-cmp v0.7.0 h1:wk8382ETsv4JYUZwIsn6YpYiWiBsYLSJiTsyBybVuN8=
 github.com/google/go-cmp v0.7.0/go.mod h1:pXiqmnSA92OHEEa9HXL2W4E7lf9JzCmGVUdgjX3N/iU=
-github.com/google/pprof v0.0.0-20260302011040-a15ffb7f9dcc h1:VBbFa1lDYWEeV5FZKUiYKYT0VxCp9twUmmaq9eb8sXw=
-github.com/google/pprof v0.0.0-20260302011040-a15ffb7f9dcc/go.mod h1:MxpfABSjhmINe3F1It9d+8exIHFvUqtLIRCdOGNXqiI=
+github.com/google/pprof v0.0.0-20260902005441-ca85771921e4 h1:/6mPXfWmhv8eKck12I0YNIcIjwHtxP3YRIMKiEgTjWg=
+github.com/google/pprof v0.0.0-20260902005441-ca85771921e4/go.mod h1:jl5iWTm0/hd5PjEYEOuwAJ57L/CibdZfrqZ5XA5GrCk=
 github.com/google/uuid v1.6.0 h1:NIvaJDMOsjHA8n1jAhLSgzrAzy1Hgr+hNrb57e+94F0=
 github.com/google/uuid v1.6.0/go.mod h1:TIyPZe4MgqvfeYDBFedMoGGpEw/LqOeaOT+nhxU+yHo=
 github.com/gorilla/websocket v1.5.3 h1:saDtZ6Pbx/0u+bgYQ3q96pZgCzfhKXGPqt7kZ72aNNg=
@@ -68,47 +64,47 @@ github.com/hashicorp/golang-lru/v2 v2.0.7 h1:a+bsQ5rvGLjzHuww6tVxozPZFVghXaHOwFs
 github.com/hashicorp/golang-lru/v2 v2.0.7/go.mod h1:QeFd9opnmA6QUJc5vARoKUSoFhyfM2/ZepoAG6RGpeM=
 github.com/inconshreveable/mousetrap v1.1.0 h1:wN+x4NVGpMsO7ErUn/mUI3vEoE6Jt13X2s0bqwp9tc8=
 github.com/inconshreveable/mousetrap v1.1.0/go.mod h1:vpF70FUmC8bwa3OWnCshd2FqLfsEA9PFc4w1p2J65bw=
-github.com/jarcoal/httpmock v1.4.1 h1:0Ju+VCFuARfFlhVXFc2HxlcQkfB+Xq12/EotHko+x2A=
-github.com/jarcoal/httpmock v1.4.1/go.mod h1:ftW1xULwo+j0R0JJkJIIi7UKigZUXCLLanykgjwBXL0=
+github.com/jarcoal/httpmock v1.4.2 h1:dKwiP/9zITCPfBLsDn3kchbSOu16JrnxtVEmL0fPRcI=
+github.com/jarcoal/httpmock v1.4.2/go.mod h1:ftW1xULwo+j0R0JJkJIIi7UKigZUXCLLanykgjwBXL0=
 github.com/jessevdk/go-flags v1.4.0/go.mod h1:4FA24M0QyGHXBuZZK/XkWh8h0e1EYbRYJSGM75WSRxI=
-github.com/klauspost/compress v1.18.5 h1:/h1gH5Ce+VWNLSWqPzOVn6XBO+vJbCNGvjoaGBFW2IE=
-github.com/klauspost/compress v1.18.5/go.mod h1:cwPg85FWrGar70rWktvGQj8/hthj3wpl0PGDogxkrSQ=
+github.com/klauspost/compress v1.20.0 h1:a3C1ke2ohxFymNlb2HWAHjDeKCI90scRskErZkR0ezA=
+github.com/klauspost/compress v1.20.0/go.mod h1:LUdAzn7YLVvxLpc7y3V1m40wESHTgc1422pwwBSKYuI=
 github.com/kr/pretty v0.3.1 h1:flRD4NNwYAUpkphVc1HcthR4KEIFJ65n8Mw5qdRn3LE=
 github.com/kr/pretty v0.3.1/go.mod h1:hoEshYVHaxMs3cyo3Yncou5ZscifuDolrwPKZanG3xk=
 github.com/kr/text v0.2.0 h1:5Nx0Ya0ZqY2ygV366QzturHI13Jq95ApcVaJBhpS+AY=
 github.com/kr/text v0.2.0/go.mod h1:eLer722TekiGuMkidMxC/pM04lWEeraHUUmBw8l2grE=
-github.com/lufia/plan9stats v0.0.0-20260324052639-156f7da3f749 h1:Qj3hTcdWH8uMZDI41HNuTuJN525C7NBrbtH5kSO6fPk=
-github.com/lufia/plan9stats v0.0.0-20260324052639-156f7da3f749/go.mod h1:autxFIvghDt3jPTLoqZ9OZ7s9qTGNAWmYCjVFWPX/zg=
-github.com/lxzan/gws v1.9.1 h1:4lbIp4cW0hOLP3ejFHR/uWRy741AURx7oKkNNi2OT9o=
-github.com/lxzan/gws v1.9.1/go.mod h1:gXHSCPmTGryWJ4icuqy8Yho32E4YIMHH0fkDRYJRbdc=
-github.com/mattn/go-colorable v0.1.14 h1:9A9LHSqF/7dyVVX6g0U9cwm9pG3kP9gSzcuIPHPsaIE=
-github.com/mattn/go-colorable v0.1.14/go.mod h1:6LmQG8QLFO4G5z1gPvYEzlUgJ2wF+stgPZH1UqBm1s8=
-github.com/mattn/go-isatty v0.0.20 h1:xfD0iDuEKnDkl03q4limB+vH+GxLEtL/jb4xVJSWWEY=
-github.com/mattn/go-isatty v0.0.20/go.mod h1:W+V8PltTTMOvKvAeJH7IuucS94S2C6jfK/D7dTCTo3Y=
+github.com/lufia/plan9stats v0.0.0-20260802145828-341c2f0c90b5 h1:eveIIGn4BGM3qknO74omf6HYr30/exH+eVUTuAgwjZ0=
+github.com/lufia/plan9stats v0.0.0-20260802145828-341c2f0c90b5/go.mod h1:autxFIvghDt3jPTLoqZ9OZ7s9qTGNAWmYCjVFWPX/zg=
+github.com/lxzan/gws v1.10.1 h1:1xG+tDOV0lgDeVPf0wNT74u3cn0K3LpcavRrTPTrMwQ=
+github.com/lxzan/gws v1.10.1/go.mod h1:gXHSCPmTGryWJ4icuqy8Yho32E4YIMHH0fkDRYJRbdc=
+github.com/mattn/go-colorable v0.1.15 h1:+u9SLTRGnXv73cEsnsmoZBom+dMU88B2M0aDcWy0/jY=
+github.com/mattn/go-colorable v0.1.15/go.mod h1:6LmQG8QLFO4G5z1gPvYEzlUgJ2wF+stgPZH1UqBm1s8=
+github.com/mattn/go-isatty v0.0.24 h1:tGZZoVgT/KiqK1c8ocVLeDS8BSWMRd47J3Lbz7vsReI=
+github.com/mattn/go-isatty v0.0.24/go.mod h1:nMCL3Zebbrt45jsMDgnfIwz6ydEQApk5oEI3HqDio6A=
 github.com/ncruces/go-strftime v1.0.0 h1:HMFp8mLCTPp341M/ZnA4qaf7ZlsbTc+miZjCLOFAw7w=
 github.com/ncruces/go-strftime v1.0.0/go.mod h1:Fwc5htZGVVkseilnfgOVb9mKy6w1naJmn9CehxcKcls=
-github.com/nicholas-fedor/shoutrrr v0.14.1 h1:6sx4cJNfNuUtD6ygGlB0dqcCQ+abfsUh+b+6jgujf6A=
-github.com/nicholas-fedor/shoutrrr v0.14.1/go.mod h1:U7IywBkLpBV7rgn8iLbQ9/LklJG1gm24bFv5cXXsDKs=
-github.com/onsi/ginkgo/v2 v2.28.1 h1:S4hj+HbZp40fNKuLUQOYLDgZLwNUVn19N3Atb98NCyI=
-github.com/onsi/ginkgo/v2 v2.28.1/go.mod h1:CLtbVInNckU3/+gC8LzkGUb9oF+e8W8TdUsxPwvdOgE=
-github.com/onsi/gomega v1.39.1 h1:1IJLAad4zjPn2PsnhH70V4DKRFlrCzGBNrNaru+Vf28=
-github.com/onsi/gomega v1.39.1/go.mod h1:hL6yVALoTOxeWudERyfppUcZXjMwIMLnuSfruD2lcfg=
+github.com/nicholas-fedor/shoutrrr v0.19.0 h1:Rl6bpK3DXuR2Trtx2JV8t+wjUwkHdRHrc8nBKoEpHr0=
+github.com/nicholas-fedor/shoutrrr v0.19.0/go.mod h1:Glfdi8AGTbnEn2k2+hW62n8oL0i9vqRVFtXaUIthNks=
+github.com/onsi/ginkgo/v2 v2.32.1 h1:6tlvcDm/3sE8lGJbZ4+d4mO3RLy24/tQWOFzVSQNIfw=
+github.com/onsi/ginkgo/v2 v2.32.1/go.mod h1:+aXOY+vzZ5mu2iI2HpTZUPmM//oQfsNFX6gU9kNcA44=
+github.com/onsi/gomega v1.43.0 h1:VlG/1FxqNxhSO+lq/OHBNaaqwiBK/mO8JbVkX9Y+FeU=
+github.com/onsi/gomega v1.43.0/go.mod h1:REff/hsDsodHoKlWsP2mAPhu1+5/6hVYNf9rIEBpeSg=
 github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
-github.com/pmezard/go-difflib v1.0.1-0.20181226105442-5d4384ee4fb2 h1:Jamvg5psRIccs7FGNTlIRMkT8wgtp5eCXdBlqhYGL6U=
-github.com/pmezard/go-difflib v1.0.1-0.20181226105442-5d4384ee4fb2/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
 github.com/pocketbase/dbx v1.12.0 h1:/oLErM+A0b4xI0PWTGPqSDVjzix48PqI/bng2l0PzoA=
 github.com/pocketbase/dbx v1.12.0/go.mod h1:xXRCIAKTHMgUCyCKZm55pUOdvFziJjQfXaWKhu2vhMs=
-github.com/pocketbase/pocketbase v0.36.7 h1:MrViB7BptPYrf2Nt25pJEYBqUdFjuhRKu1p5GTrkvPA=
-github.com/pocketbase/pocketbase v0.36.7/go.mod h1:qX4HuVjoKXtEg41fSJVM0JLfGWXbBmHxVv/FaE446r4=
-github.com/power-devops/perfstat v0.0.0-20240221224432-82ca36839d55 h1:o4JXh1EVt9k/+g42oCprj/FisM4qX9L3sZB3upGN2ZU=
-github.com/power-devops/perfstat v0.0.0-20240221224432-82ca36839d55/go.mod h1:OmDBASR4679mdNQnz2pUhc2G8CO2JrUAVFDRBDP/hJE=
+github.com/pocketbase/ozzo-validation/v4 v4.3.0 h1:uKBDVma7bZqgR2a6AwE+k9hkuDFfiZMpBHQdZ1z3iQs=
+github.com/pocketbase/ozzo-validation/v4 v4.3.0/go.mod h1:6XNjSTw/Jb2F8LOkKO3oyzIWExbrGiYoS4uVxVwz90g=
+github.com/pocketbase/pocketbase v0.40.2 h1:7gTqvt3bmilkphyZZ1QNhX19g3BXHqT7ynDyU81RVT4=
+github.com/pocketbase/pocketbase v0.40.2/go.mod h1:jc3YuyToy+ZXM4CeO7uSCN/htgR8yv+tjSE3eJZ8eh8=
+github.com/power-devops/perfstat v0.0.0-20260805114148-88456608a4f6 h1:jL3a8soXdzuTCcRnKhOmtcsVOObdDTFf4O2B403HPRU=
+github.com/power-devops/perfstat v0.0.0-20260805114148-88456608a4f6/go.mod h1:OmDBASR4679mdNQnz2pUhc2G8CO2JrUAVFDRBDP/hJE=
 github.com/remyoudompheng/bigfft v0.0.0-20230129092748-24d4a6f8daec h1:W09IVJc94icq4NjY3clb7Lk8O1qJ8BdBEF8z0ibU0rE=
 github.com/remyoudompheng/bigfft v0.0.0-20230129092748-24d4a6f8daec/go.mod h1:qqbHyh8v60DhA7CoWK5oRCqLrMHRGoxYCSS9EjAz6Eo=
 github.com/rogpeppe/go-internal v1.9.0 h1:73kH8U+JUqXU8lRuOHeVHaa/SZPifC7BkcraZVejAe8=
 github.com/rogpeppe/go-internal v1.9.0/go.mod h1:WtVeX8xhTBvf0smdhujwtBcq4Qrzq/fJaraNFVN+nFs=
 github.com/russross/blackfriday/v2 v2.1.0/go.mod h1:+Rmxgy9KzJVeS9/2gXHxylqXiyQDYRxCVz55jmeOWTM=
-github.com/shirou/gopsutil/v4 v4.26.2 h1:X8i6sicvUFih4BmYIGT1m2wwgw2VG9YgrDTi7cIRGUI=
-github.com/shirou/gopsutil/v4 v4.26.2/go.mod h1:LZ6ewCSkBqUpvSOf+LsTGnRinC6iaNUNMGBtDkJBaLQ=
+github.com/shirou/gopsutil/v4 v4.26.8 h1:YQMTF/1J50B5+Y0vlo1eDRf5DoR7Gk69hY+8wjYkQeo=
+github.com/shirou/gopsutil/v4 v4.26.8/go.mod h1:5O9FjBiXoTDFatIWjZZosqj4pV0DRtLx598xGbBehzM=
 github.com/spf13/cast v1.10.0 h1:h2x0u2shc1QuLHfxi+cTJvs30+ZAHOGRic8uyGTDWxY=
 github.com/spf13/cast v1.10.0/go.mod h1:jNfB8QC9IA6ZuY2ZjDp0KtFO2LZZlg4S/7bzP6qqeHo=
 github.com/spf13/cobra v1.10.2 h1:DMTTonx5m65Ic0GOoRY2c16WCbHxOOw6xxezuLaBpcU=
@@ -120,87 +116,86 @@ github.com/stretchr/objx v0.1.0/go.mod h1:HFkY916IF+rwdDfMAkV7OtwuqBVzrE8GR6GFx+
 github.com/stretchr/objx v0.5.3 h1:jmXUvGomnU1o3W/V5h2VEradbpJDwGrzugQQvL0POH4=
 github.com/stretchr/objx v0.5.3/go.mod h1:rDQraq+vQZU7Fde9LOZLr8Tax6zZvy4kuNKF+QYS+U0=
 github.com/stretchr/testify v1.4.0/go.mod h1:j7eGeouHqKxXV5pUuKE4zz7dFj8WfuZ+81PSLYec5m4=
-github.com/stretchr/testify v1.11.1 h1:7s2iGBzp5EwR7/aIZr8ao5+dra3wiQyKjjFuvgVKu7U=
-github.com/stretchr/testify v1.11.1/go.mod h1:wZwfW3scLgRK+23gO65QZefKpKQRnfz6sD981Nm4B6U=
-github.com/tklauser/go-sysconf v0.3.16 h1:frioLaCQSsF5Cy1jgRBrzr6t502KIIwQ0MArYICU0nA=
-github.com/tklauser/go-sysconf v0.3.16/go.mod h1:/qNL9xxDhc7tx3HSRsLWNnuzbVfh3e7gh/BmM179nYI=
-github.com/tklauser/numcpus v0.11.0 h1:nSTwhKH5e1dMNsCdVBukSZrURJRoHbSEQjdEbY+9RXw=
-github.com/tklauser/numcpus v0.11.0/go.mod h1:z+LwcLq54uWZTX0u/bGobaV34u6V7KNlTZejzM6/3MQ=
+github.com/stretchr/testify v1.12.1 h1:EuwCh5fleGS7H32xRwO3wRGT7DxrDhLAT6FF8MpWDWE=
+github.com/stretchr/testify v1.12.1/go.mod h1:MDEgiDPPsNp5cuIrHPPCyornHKgEVbtFUmoNlxoYthg=
+github.com/tklauser/go-sysconf v0.4.0 h1:7H0uAN+7RkwWRaxhYXDLqa5V3LPrJeV8wmD9dRUgPQU=
+github.com/tklauser/go-sysconf v0.4.0/go.mod h1:8mTNWyog7H+MpKijp4VmKJAd2bbYQ2zuUwkYRbUArPI=
+github.com/tklauser/numcpus v0.12.0 h1:NR85qdvHA9pFse3x3weVZ0r0ST8R6l5RHbZrlRaqob4=
+github.com/tklauser/numcpus v0.12.0/go.mod h1:ABHeXzJnr/qqwguhClkZKT1/8VABcYrsyUiUGobwWJg=
 github.com/x448/float16 v0.8.4 h1:qLwI1I70+NjRFUR3zs1JPUCgaCXSh3SW62uAKT1mSBM=
 github.com/x448/float16 v0.8.4/go.mod h1:14CWIYCyZA/cWjXOioeEpHeN/83MdbZDRQHoFcYsOfg=
 github.com/yusufpapurcu/wmi v1.2.4 h1:zFUKzehAFReQwLys1b/iSMl+JQGSCSjtVqQn9bBrPo0=
 github.com/yusufpapurcu/wmi v1.2.4/go.mod h1:SBZ9tNy3G9/m5Oi98Zks0QjeHVDvuK0qfxQmPyzfmi0=
 go.uber.org/goleak v1.3.0 h1:2K3zAYmnTNqV73imy9J1T3WC+gmCePx2hEGkimedGto=
 go.uber.org/goleak v1.3.0/go.mod h1:CoHD4mav9JJNrW/WLlf7HGZPjdw8EucARQHekz1X6bE=
-go.yaml.in/yaml/v3 v3.0.4 h1:tfq32ie2Jv2UxXFdLJdh3jXuOzWiL1fo0bu/FbuKpbc=
 go.yaml.in/yaml/v3 v3.0.4/go.mod h1:DhzuOOF2ATzADvBadXxruRBLzYTpT36CKvDb3+aBEFg=
+go.yaml.in/yaml/v3 v3.0.5 h1:N6y/pJk8buWs9NY5ERU2HSMfm+IuD/OtfdAnq6kESPw=
+go.yaml.in/yaml/v3 v3.0.5/go.mod h1:HVTZu1O7/Vkt2N+BFy8Zza+lnLsABggaTM2ZpNIGuKg=
 golang.org/x/crypto v0.0.0-20190308221718-c2843e01d9a2/go.mod h1:djNgcEr1/C05ACkg1iLfiJU5Ep61QUkGW8qpdssI0+w=
-golang.org/x/crypto v0.49.0 h1:+Ng2ULVvLHnJ/ZFEq4KdcDd/cfjrrjjNSXNzxg0Y4U4=
-golang.org/x/crypto v0.49.0/go.mod h1:ErX4dUh2UM+CFYiXZRTcMpEcN8b/1gxEuv3nODoYtCA=
-golang.org/x/exp v0.0.0-20260312153236-7ab1446f8b90 h1:jiDhWWeC7jfWqR9c/uplMOqJ0sbNlNWv0UkzE0vX1MA=
-golang.org/x/exp v0.0.0-20260312153236-7ab1446f8b90/go.mod h1:xE1HEv6b+1SCZ5/uscMRjUBKtIxworgEcEi+/n9NQDQ=
+golang.org/x/crypto v0.56.0 h1:GUh5Ii4J5jtcseSMiRqr1jXCNHoxjeV9Fmekc2oLy6Y=
+golang.org/x/crypto v0.56.0/go.mod h1:OMW5y6CY9l38uPLmxU6l6pwcXp1obtLo3e6gT7gQR2I=
+golang.org/x/exp v0.0.0-20260824195058-e88cd73687aa h1:QSyA8ishJCyT21kER9KwNt0b7BM3iRK4x9QXhjN5Fdk=
+golang.org/x/exp v0.0.0-20260824195058-e88cd73687aa/go.mod h1:zeBbvyFKDaLwa7CH/zI8KXt7gTl14SF7sO08Pl5jBCM=
 golang.org/x/image v0.0.0-20191009234506-e7c1f5e7dbb8/go.mod h1:FeLwcggjj3mMvU+oOTbSwawSJRM1uh48EjtB4UJZlP0=
-golang.org/x/image v0.38.0 h1:5l+q+Y9JDC7mBOMjo4/aPhMDcxEptsX+Tt3GgRQRPuE=
-golang.org/x/image v0.38.0/go.mod h1:/3f6vaXC+6CEanU4KJxbcUZyEePbyKbaLoDOe4ehFYY=
-golang.org/x/mod v0.34.0 h1:xIHgNUUnW6sYkcM5Jleh05DvLOtwc6RitGHbDk4akRI=
-golang.org/x/mod v0.34.0/go.mod h1:ykgH52iCZe79kzLLMhyCUzhMci+nQj+0XkbXpNYtVjY=
+golang.org/x/image v0.45.0 h1:FMb1nTbH5H9vF55SriQHgFw5GnNL9Jg6L25BwXKzhB0=
+golang.org/x/image v0.45.0/go.mod h1:n62x/7RqlwXDvGsSU4u6IUTUf6KghUZ9Bt7cG/T9Fx4=
+golang.org/x/mod v0.40.0 h1:hUv+3cXcdRHz08UmSiOob7sadHig73uo5bkXxQ/tvUs=
+golang.org/x/mod v0.40.0/go.mod h1:0/weTWkPWGBikyTWAX3dkjVztMmBA5hM0DH6BElSupE=
 golang.org/x/net v0.0.0-20190603091049-60506f45cf65/go.mod h1:HSz+uSET+XFnRR8LxR5pz3Of3rY3CfYBVs4xY44aLks=
-golang.org/x/net v0.52.0 h1:He/TN1l0e4mmR3QqHMT2Xab3Aj3L9qjbhRm78/6jrW0=
-golang.org/x/net v0.52.0/go.mod h1:R1MAz7uMZxVMualyPXb+VaqGSa3LIaUqk0eEt3w36Sw=
+golang.org/x/net v0.58.0 h1:ynWG7rqYi4ccpTEuPZ2QGWHktVEM9DMCj9yzDE0Q7To=
+golang.org/x/net v0.58.0/go.mod h1:YwCddHnFlT7eLQqVprV19OnhLGtc5xOKgE0RyqgfWAU=
 golang.org/x/oauth2 v0.36.0 h1:peZ/1z27fi9hUOFCAZaHyrpWG5lwe0RJEEEeH0ThlIs=
 golang.org/x/oauth2 v0.36.0/go.mod h1:YDBUJMTkDnJS+A4BP4eZBjCqtokkg1hODuPjwiGPO7Q=
-golang.org/x/sync v0.20.0 h1:e0PTpb7pjO8GAtTs2dQ6jYa5BWYlMuX047Dco/pItO4=
-golang.org/x/sync v0.20.0/go.mod h1:9xrNwdLfx4jkKbNva9FpL6vEN7evnE43NNNJQ2LF3+0=
+golang.org/x/sync v0.22.0 h1:SZjpbeLmrCk4xhRSZFNZW5gFUeCeFgjekvI/+gfScek=
+golang.org/x/sync v0.22.0/go.mod h1:9xrNwdLfx4jkKbNva9FpL6vEN7evnE43NNNJQ2LF3+0=
 golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
 golang.org/x/sys v0.0.0-20190916202348-b4ddaad3f8a3/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20201204225414-ed752295db88/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.1.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
-golang.org/x/sys v0.6.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
-golang.org/x/sys v0.42.0 h1:omrd2nAlyT5ESRdCLYdm3+fMfNFE/+Rf4bDIQImRJeo=
-golang.org/x/sys v0.42.0/go.mod h1:4GL1E5IUh+htKOUEOaiffhrAeqysfVGipDYzABqnCmw=
-golang.org/x/term v0.41.0 h1:QCgPso/Q3RTJx2Th4bDLqML4W6iJiaXFq2/ftQF13YU=
-golang.org/x/term v0.41.0/go.mod h1:3pfBgksrReYfZ5lvYM0kSO0LIkAl4Yl2bXOkKP7Ec2A=
+golang.org/x/sys v0.47.0 h1:o7XGOvZQCADBQQ4Y7VNq2dRWQR7JmOUW8Kxx4ZsNgWs=
+golang.org/x/sys v0.47.0/go.mod h1:4GL1E5IUh+htKOUEOaiffhrAeqysfVGipDYzABqnCmw=
+golang.org/x/term v0.45.0 h1:NwWyBmoJCbfTHpxrWoZ9C6/VxOf7ic219I8xZZFdrf0=
+golang.org/x/term v0.45.0/go.mod h1:9aqxs0blBcrm/n0L9QW0aRVD+ktan8ssZromtqJC43w=
 golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
 golang.org/x/text v0.3.2/go.mod h1:bEr9sfX3Q8Zfm5fL9x+3itogRgK3+ptLWKqgva+5dAk=
-golang.org/x/text v0.35.0 h1:JOVx6vVDFokkpaq1AEptVzLTpDe9KGpj5tR4/X+ybL8=
-golang.org/x/text v0.35.0/go.mod h1:khi/HExzZJ2pGnjenulevKNX1W67CUy0AsXcNubPGCA=
+golang.org/x/text v0.41.0 h1:vz/seA0lnX87Othu2f/0L24RcgrXD9/YFTSuGjj3rH8=
+golang.org/x/text v0.41.0/go.mod h1:jvf1O8ajNzZqhSrQBPbutR/EB83Cc0CFrezNQIwbb5M=
 golang.org/x/tools v0.0.0-20180917221912-90fa682c2a6e/go.mod h1:n7NCudcB/nEzxVGmLbDWY5pfWTLqBcC2KZ6jyYvM4mQ=
-golang.org/x/tools v0.43.0 h1:12BdW9CeB3Z+J/I/wj34VMl8X+fEXBxVR90JeMX5E7s=
-golang.org/x/tools v0.43.0/go.mod h1:uHkMso649BX2cZK6+RpuIPXS3ho2hZo4FVwfoy1vIk0=
+golang.org/x/tools v0.49.0 h1:3NI7VXzL9+1WZD52Dx2ttoPwD5DWrFGpl9mFZDlmisI=
+golang.org/x/tools v0.49.0/go.mod h1:SJNXV9DBKT0UbdttsQjbfJlAE/q+y36++zo3uL3N0Oo=
 google.golang.org/appengine v1.6.5/go.mod h1:8WjMMxjGQR8xUklV/ARdw2HLXBOI7O7uCIDZVag1xfc=
+gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405 h1:yhCVgyC4o1eVCa2tZl7eS0r+SDo693bJlVdllGtEeKM=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
-gopkg.in/check.v1 v1.0.0-20201130134442-10cb98267c6c h1:Hei/4ADfdWqJk1ZMxUNpqntNwaWcugrBjAiHlqqRiVk=
-gopkg.in/check.v1 v1.0.0-20201130134442-10cb98267c6c/go.mod h1:JHkPIbrfpd72SG/EVd6muEfDQjcINNoR0C8j2r3qZ4Q=
 gopkg.in/yaml.v1 v1.0.0-20140924161607-9f9df34309c0/go.mod h1:WDnlLJ4WF5VGsH/HVa3CI79GS0ol3YnhVnKP89i0kNg=
 gopkg.in/yaml.v2 v2.2.2/go.mod h1:hI93XBmqTisBFMUTm0b8Fm+jr3Dg1NNxqwp+5A1VGuI=
 gopkg.in/yaml.v3 v3.0.1 h1:fxVm/GzAzEWqLHuvctI91KS9hhNmmWOoWu0XTYJS7CA=
 gopkg.in/yaml.v3 v3.0.1/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
 howett.net/plist v1.0.1 h1:37GdZ8tP09Q35o9ych3ehygcsL+HqKSwzctveSlarvM=
 howett.net/plist v1.0.1/go.mod h1:lqaXoTrLY4hg8tnEzNru53gicrbv7rrk+2xJA/7hw9g=
-modernc.org/cc/v4 v4.27.1 h1:9W30zRlYrefrDV2JE2O8VDtJ1yPGownxciz5rrbQZis=
-modernc.org/cc/v4 v4.27.1/go.mod h1:uVtb5OGqUKpoLWhqwNQo/8LwvoiEBLvZXIQ/SmO6mL0=
-modernc.org/ccgo/v4 v4.32.0 h1:hjG66bI/kqIPX1b2yT6fr/jt+QedtP2fqojG2VrFuVw=
-modernc.org/ccgo/v4 v4.32.0/go.mod h1:6F08EBCx5uQc38kMGl+0Nm0oWczoo1c7cgpzEry7Uc0=
+modernc.org/cc/v4 v4.29.1 h1:MKgdCV3WykTSPqpVrnxdEDS0HEd2FHpKZDzxzU5LyeI=
+modernc.org/cc/v4 v4.29.1/go.mod h1:OnovgIhbbMXMu1aISnJ0wvVD1KnW+cAUJkIrAWh+kVI=
+modernc.org/ccgo/v4 v4.34.6 h1:sBgfIwyN0TQ9C5hwIeuqyeAKyMWnbvj2fvpF4L11uzU=
+modernc.org/ccgo/v4 v4.34.6/go.mod h1:SZ8YcN9NG7XVsQYdm6jYBvi8PQP1qi+kqB6OhjqI3Fk=
 modernc.org/fileutil v1.4.0 h1:j6ZzNTftVS054gi281TyLjHPp6CPHr2KCxEXjEbD6SM=
 modernc.org/fileutil v1.4.0/go.mod h1:EqdKFDxiByqxLk8ozOxObDSfcVOv/54xDs/DUHdvCUU=
 modernc.org/gc/v2 v2.6.5 h1:nyqdV8q46KvTpZlsw66kWqwXRHdjIlJOhG6kxiV/9xI=
 modernc.org/gc/v2 v2.6.5/go.mod h1:YgIahr1ypgfe7chRuJi2gD7DBQiKSLMPgBQe9oIiito=
-modernc.org/gc/v3 v3.1.2 h1:ZtDCnhonXSZexk/AYsegNRV1lJGgaNZJuKjJSWKyEqo=
-modernc.org/gc/v3 v3.1.2/go.mod h1:HFK/6AGESC7Ex+EZJhJ2Gni6cTaYpSMmU/cT9RmlfYY=
+modernc.org/gc/v3 v3.1.4 h1:2g65LGVSmFQrXeITAw97x7hCRvZFcyE1uDP+7Vng7JI=
+modernc.org/gc/v3 v3.1.4/go.mod h1:HFK/6AGESC7Ex+EZJhJ2Gni6cTaYpSMmU/cT9RmlfYY=
 modernc.org/goabi0 v0.2.0 h1:HvEowk7LxcPd0eq6mVOAEMai46V+i7Jrj13t4AzuNks=
 modernc.org/goabi0 v0.2.0/go.mod h1:CEFRnnJhKvWT1c1JTI3Avm+tgOWbkOu5oPA8eH8LnMI=
-modernc.org/libc v1.70.0 h1:U58NawXqXbgpZ/dcdS9kMshu08aiA6b7gusEusqzNkw=
-modernc.org/libc v1.70.0/go.mod h1:OVmxFGP1CI/Z4L3E0Q3Mf1PDE0BucwMkcXjjLntvHJo=
+modernc.org/libc v1.74.4 h1:fX1Omw4o2/1C2iRkkIsrQTasJQldLhRmuPreXLoWs9k=
+modernc.org/libc v1.74.4/go.mod h1:eeQAS9W3sZeKYMFubydxJpII9ybHWshk+7or7bLG9co=
 modernc.org/mathutil v1.7.1 h1:GCZVGXdaN8gTqB1Mf/usp1Y/hSqgI2vAGGP4jZMCxOU=
 modernc.org/mathutil v1.7.1/go.mod h1:4p5IwJITfppl0G4sUEDtCr4DthTaT47/N3aT6MhfgJg=
-modernc.org/memory v1.11.0 h1:o4QC8aMQzmcwCK3t3Ux/ZHmwFPzE6hf2Y5LbkRs+hbI=
-modernc.org/memory v1.11.0/go.mod h1:/JP4VbVC+K5sU2wZi9bHoq2MAkCnrt2r98UGeSK7Mjw=
-modernc.org/opt v0.1.4 h1:2kNGMRiUjrp4LcaPuLY2PzUfqM/w9N23quVwhKt5Qm8=
-modernc.org/opt v0.1.4/go.mod h1:03fq9lsNfvkYSfxrfUhZCWPk1lm4cq4N+Bh//bEtgns=
+modernc.org/memory v1.12.1 h1:nFMiWrpStgZczNl6XI9GnIk/rWhYIyHGUaR04pGbp9g=
+modernc.org/memory v1.12.1/go.mod h1:/JP4VbVC+K5sU2wZi9bHoq2MAkCnrt2r98UGeSK7Mjw=
+modernc.org/opt v0.2.0 h1:tGyef5ApycA7FSEOMraay9SaTk5zmbx7Tu+cJs4QKZg=
+modernc.org/opt v0.2.0/go.mod h1:03fq9lsNfvkYSfxrfUhZCWPk1lm4cq4N+Bh//bEtgns=
 modernc.org/sortutil v1.2.1 h1:+xyoGf15mM3NMlPDnFqrteY07klSFxLElE2PVuWIJ7w=
 modernc.org/sortutil v1.2.1/go.mod h1:7ZI3a3REbai7gzCLcotuw9AC4VZVpYMjDzETGsSMqJE=
-modernc.org/sqlite v1.46.2 h1:gkXQ6R0+AjxFC/fTDaeIVLbNLNrRoOK7YYVz5BKhTcE=
-modernc.org/sqlite v1.46.2/go.mod h1:hWjRO6Tj/5Ik8ieqxQybiEOUXy0NJFNp2tpvVpKlvig=
+modernc.org/sqlite v1.57.0 h1:qNQP6xnx5M0ISNtlnxoOX0+cD5bJ0/gr9aMmndFczzg=
+modernc.org/sqlite v1.57.0/go.mod h1:yCJ2cmAaIkHQ25oXWrF8H4O1lIfPYPR26yCEDj2P3pQ=
 modernc.org/strutil v1.2.1 h1:UneZBkQA+DX2Rp35KcM69cSsNES9ly8mQWD71HKlOA0=
 modernc.org/strutil v1.2.1/go.mod h1:EHkiggD70koQxjVdSBM3JKM7k6L0FbGE5eymy9i3B9A=
 modernc.org/token v1.1.0 h1:Xl7Ap9dKaEs5kLoOQeQmPWevfnk/DM5qcLcYlA8ys6Y=

internal/alerts/alerts.go

@@ -20,10 +20,10 @@ type hubLike interface {
 }
 
 type AlertManager struct {
-	hub           hubLike
-	stopOnce      sync.Once
-	pendingAlerts sync.Map
-	alertsCache   *AlertsCache
+	hub                    hubLike
+	stopOnce               sync.Once
+	pendingAlerts          sync.Map
+	alertsCache            *AlertsCache
 }
 
 type AlertMessageData struct {
@@ -48,6 +48,7 @@ type SystemAlertFsStats struct {
 // Values pulled from system_stats.stats that are relevant to alerts.
 type SystemAlertStats struct {
 	Cpu          float64                       `json:"cpu"`
+	CpuBreakdown []float64                     `json:"cpub"`
 	Mem          float64                       `json:"mp"`
 	Disk         float64                       `json:"dp"`
 	Bandwidth    [2]uint64                     `json:"b"`
@@ -55,13 +56,20 @@ type SystemAlertStats struct {
 	Temperatures map[string]float32            `json:"t"`
 	LoadAvg      [3]float64                    `json:"la"`
 	Battery      [2]uint8                      `json:"bat"`
+	Batteries    map[string]uint8              `json:"bats"`
 	ExtraFs      map[string]SystemAlertFsStats `json:"efs"`
+	ZfsPools     map[string]SystemAlertZfsPool `json:"z"`
 }
 
 type SystemAlertGPUData struct {
 	Usage float64 `json:"u"`
 }
 
+type SystemAlertZfsPool struct {
+	Total float64 `json:"d"`
+	Used  float64 `json:"du"`
+}
+
 type SystemAlertData struct {
 	systemRecord *core.Record
 	alertData    CachedAlertData
@@ -110,6 +118,9 @@ func (am *AlertManager) bindEvents() {
 	am.hub.OnRecordAfterUpdateSuccess("alerts").BindFunc(updateHistoryOnAlertUpdate)
 	am.hub.OnRecordAfterDeleteSuccess("alerts").BindFunc(resolveHistoryOnAlertDelete)
 	am.hub.OnRecordAfterUpdateSuccess("smart_devices").BindFunc(am.handleSmartDeviceAlert)
+	am.hub.OnRecordAfterCreateSuccess("zfs_pools").BindFunc(am.handleZfsPoolCreateAlert)
+	am.hub.OnRecordAfterUpdateSuccess("zfs_pools").BindFunc(am.handleZfsPoolAlert)
+	am.hub.OnRecordAfterDeleteSuccess("zfs_pools").BindFunc(resolveZfsPoolHistoryOnDelete)
 
 	am.hub.OnServe().BindFunc(func(e *core.ServeEvent) error {
 		// Populate all alerts into cache on startup
@@ -118,6 +129,9 @@ func (am *AlertManager) bindEvents() {
 		if err := resolveStatusAlerts(e.App); err != nil {
 			e.App.Logger().Error("Failed to resolve stale status alerts", "err", err)
 		}
+		if err := resolveSystemdAlerts(e.App); err != nil {
+			e.App.Logger().Error("Failed to resolve stale systemd alerts", "err", err)
+		}
 		if err := am.restorePendingStatusAlerts(); err != nil {
 			e.App.Logger().Error("Failed to restore pending status alerts", "err", err)
 		}
@@ -302,21 +316,6 @@ func (am *AlertManager) SendShoutrrrAlert(notificationUrl, title, message, link,
 	return nil
 }
 
-func (am *AlertManager) SendTestNotification(e *core.RequestEvent) error {
-	var data struct {
-		URL string `json:"url"`
-	}
-	err := e.BindBody(&data)
-	if err != nil || data.URL == "" {
-		return e.BadRequestError("URL is required", err)
-	}
-	err = am.SendShoutrrrAlert(data.URL, "Test Alert", "This is a notification from Beszel.", am.hub.Settings().Meta.AppURL, "View Beszel")
-	if err != nil {
-		return e.JSON(200, map[string]string{"err": err.Error()})
-	}
-	return e.JSON(200, map[string]bool{"err": false})
-}
-
 // setAlertTriggered updates the "triggered" status of an alert record in the database
 func (am *AlertManager) setAlertTriggered(alert CachedAlertData, triggered bool) error {
 	alertRecord, err := am.hub.FindRecordById("alerts", alert.Id)

internal/alerts/alerts_api.go

@@ -3,8 +3,13 @@ package alerts
 import (
 	"database/sql"
 	"errors"
+	"net"
 	"net/http"
+	"net/url"
+	"slices"
+	"strings"
 
+	"github.com/henrygd/beszel/internal/hub/utils"
 	"github.com/pocketbase/dbx"
 	"github.com/pocketbase/pocketbase/core"
 )
@@ -33,6 +38,9 @@ func UpsertUserAlerts(e *core.RequestEvent) error {
 
 	err = e.App.RunInTransaction(func(txApp core.App) error {
 		for _, systemId := range reqData.Systems {
+			if !userHasSystem(txApp, userID, systemId) {
+				continue
+			}
 			// find existing matching alert
 			alertRecord, err := txApp.FindFirstRecordByFilter(alertsCollection,
 				"system={:system} && name={:name} && user={:user}",
@@ -90,6 +98,9 @@ func DeleteUserAlerts(e *core.RequestEvent) error {
 
 	err = e.App.RunInTransaction(func(txApp core.App) error {
 		for _, systemId := range reqData.Systems {
+			if !userHasSystem(txApp, userID, systemId) {
+				continue
+			}
 			// Find existing alert to delete
 			alertRecord, err := txApp.FindFirstRecordByFilter("alerts",
 				"system={:system} && name={:name} && user={:user}",
@@ -117,3 +128,91 @@ func DeleteUserAlerts(e *core.RequestEvent) error {
 
 	return e.JSON(http.StatusOK, map[string]any{"success": true, "count": numDeleted})
 }
+
+func userHasSystem(app core.App, userID, systemID string) bool {
+	system, err := app.FindRecordById("systems", systemID)
+	if err != nil {
+		return false
+	}
+	shareAll, _ := utils.GetEnv("SHARE_ALL_SYSTEMS")
+	return shareAll == "true" || slices.Contains(system.GetStringSlice("users"), userID)
+}
+
+// SendTestNotification handles API request to send a test notification to a specified Shoutrrr URL
+func (am *AlertManager) SendTestNotification(e *core.RequestEvent) error {
+	var data struct {
+		URL string `json:"url"`
+	}
+	err := e.BindBody(&data)
+	if err != nil || data.URL == "" {
+		return e.BadRequestError("URL is required", err)
+	}
+	// Only allow admins to send test notifications to internal URLs
+	if !e.Auth.IsSuperuser() && e.Auth.GetString("role") != "admin" {
+		internalURL, err := isInternalURL(data.URL)
+		if err != nil {
+			return e.BadRequestError(err.Error(), nil)
+		}
+		if internalURL {
+			return e.ForbiddenError("Only admins can send to internal destinations", nil)
+		}
+	}
+	err = am.SendShoutrrrAlert(data.URL, "Test Alert", "This is a notification from Beszel.", am.hub.Settings().Meta.AppURL, "View Beszel")
+	if err != nil {
+		return e.JSON(200, map[string]string{"err": err.Error()})
+	}
+	return e.JSON(200, map[string]bool{"err": false})
+}
+
+// isInternalURL checks if the given shoutrrr URL points to an internal destination (localhost or private IP)
+func isInternalURL(rawURL string) (bool, error) {
+	parsedURL, err := url.Parse(rawURL)
+	if err != nil {
+		return false, err
+	}
+
+	host := parsedURL.Hostname()
+	if host == "" {
+		return false, nil
+	}
+
+	if strings.EqualFold(host, "localhost") {
+		return true, nil
+	}
+
+	if ip := net.ParseIP(host); ip != nil {
+		return isInternalIP(ip), nil
+	}
+
+	// Some Shoutrrr URLs use the host position for service identifiers rather than a
+	// network hostname (for example, discord://token@webhookid). Restrict DNS lookups
+	// to names that look like actual hostnames so valid service URLs keep working.
+	if !strings.Contains(host, ".") {
+		return false, nil
+	}
+
+	ips, err := net.LookupIP(host)
+	if err != nil {
+		return false, nil
+	}
+
+	if slices.ContainsFunc(ips, isInternalIP) {
+		return true, nil
+	}
+
+	return false, nil
+}
+
+var cgnatNetwork = &net.IPNet{
+	IP:   net.IPv4(100, 64, 0, 0),
+	Mask: net.CIDRMask(10, 32),
+}
+
+func isInternalIP(ip net.IP) bool {
+	return ip.IsPrivate() ||
+		ip.IsLoopback() ||
+		ip.IsUnspecified() ||
+		ip.IsLinkLocalUnicast() ||
+		ip.IsMulticast() ||
+		cgnatNetwork.Contains(ip)
+}

internal/alerts/alerts_api_test.go

@@ -0,0 +1,562 @@
+//go:build testing
+
+package alerts_test
+
+import (
+	"bytes"
+	"encoding/json"
+	"io"
+	"net/http"
+	"strings"
+	"testing"
+
+	"github.com/henrygd/beszel/internal/alerts"
+	beszelTests "github.com/henrygd/beszel/internal/tests"
+	pbTests "github.com/pocketbase/pocketbase/tests"
+
+	"github.com/pocketbase/dbx"
+	"github.com/pocketbase/pocketbase/core"
+
+	"github.com/stretchr/testify/assert"
+)
+
+// marshal to json and return an io.Reader (for use in ApiScenario.Body)
+func jsonReader(v any) io.Reader {
+	data, err := json.Marshal(v)
+	if err != nil {
+		panic(err)
+	}
+	return bytes.NewReader(data)
+}
+
+func TestIsInternalURL(t *testing.T) {
+	testCases := []struct {
+		name     string
+		url      string
+		internal bool
+	}{
+		{name: "loopback ipv4", url: "generic://127.0.0.1", internal: true},
+		{name: "private ipv4", url: "generic://10.0.0.1", internal: true},
+		{name: "localhost hostname", url: "generic://localhost", internal: true},
+		{name: "localhost with path", url: "generic+http://localhost/api/v1/postStuff", internal: true},
+		{name: "loopback with port and path", url: "generic+http://127.0.0.1:8080/api/v1/postStuff", internal: true},
+		{name: "public hostname", url: "generic+https://beszel.dev/api/v1/postStuff", internal: false},
+		{name: "cloud metadata ipv4", url: "generic://169.254.169.254", internal: true},
+		{name: "link-local ipv4", url: "generic://169.254.1.1", internal: true},
+		{name: "link-local ipv6", url: "generic://[fe80::1]", internal: true},
+		{name: "mapped link-local ipv4", url: "generic://[::ffff:169.254.169.254]", internal: true},
+		{name: "cgnat lower boundary", url: "generic://100.64.0.0", internal: true},
+		{name: "cgnat upper boundary", url: "generic://100.127.255.255", internal: true},
+		{name: "below cgnat", url: "generic://100.63.255.255", internal: false},
+		{name: "above cgnat", url: "generic://100.128.0.0", internal: false},
+		{name: "multicast ipv4", url: "generic://224.0.0.1", internal: true},
+		{name: "multicast ipv6", url: "generic://[ff02::1]", internal: true},
+		{name: "public ipv4", url: "generic://8.8.8.8", internal: false},
+		{name: "public ipv6", url: "generic://[2001:4860:4860::8888]", internal: false},
+		{name: "token style service url", url: "discord://abc123@123456789", internal: false},
+		{name: "single label service url", url: "slack://token@team/channel", internal: false},
+	}
+
+	for _, testCase := range testCases {
+		t.Run(testCase.name, func(t *testing.T) {
+			internal, err := alerts.IsInternalURL(testCase.url)
+			assert.NoError(t, err)
+			assert.Equal(t, testCase.internal, internal)
+		})
+	}
+}
+
+func TestUserAlertsApi(t *testing.T) {
+	hub, _ := beszelTests.NewTestHub(t.TempDir())
+	defer hub.Cleanup()
+
+	hub.StartHub()
+
+	user1, _ := beszelTests.CreateUser(hub, "alertstest@example.com", "password")
+	user1Token, _ := user1.NewAuthToken()
+
+	user2, _ := beszelTests.CreateUser(hub, "alertstest2@example.com", "password")
+	user2Token, _ := user2.NewAuthToken()
+
+	system1, _ := beszelTests.CreateRecord(hub, "systems", map[string]any{
+		"name":  "system1",
+		"users": []string{user1.Id},
+		"host":  "127.0.0.1",
+	})
+
+	system2, _ := beszelTests.CreateRecord(hub, "systems", map[string]any{
+		"name":  "system2",
+		"users": []string{user1.Id, user2.Id},
+		"host":  "127.0.0.2",
+	})
+
+	userRecords, _ := hub.CountRecords("users")
+	assert.EqualValues(t, 2, userRecords, "all users should be created")
+
+	systemRecords, _ := hub.CountRecords("systems")
+	assert.EqualValues(t, 2, systemRecords, "all systems should be created")
+
+	testAppFactory := func(t testing.TB) *pbTests.TestApp {
+		return hub.TestApp
+	}
+
+	scenarios := []beszelTests.ApiScenario{
+		// {
+		// 	Name:            "GET not implemented - returns index",
+		// 	Method:          http.MethodGet,
+		// 	URL:             "/api/beszel/user-alerts",
+		// 	ExpectedStatus:  200,
+		// 	ExpectedContent: []string{"<html ", "globalThis.BESZEL"},
+		// 	TestAppFactory:  testAppFactory,
+		// },
+		{
+			Name:            "POST no auth",
+			Method:          http.MethodPost,
+			URL:             "/api/beszel/user-alerts",
+			ExpectedStatus:  401,
+			ExpectedContent: []string{"requires valid"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "POST no body",
+			Method: http.MethodPost,
+			URL:    "/api/beszel/user-alerts",
+			Headers: map[string]string{
+				"Authorization": user1Token,
+			},
+			ExpectedStatus:  400,
+			ExpectedContent: []string{"Bad data"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "POST bad data",
+			Method: http.MethodPost,
+			URL:    "/api/beszel/user-alerts",
+			Headers: map[string]string{
+				"Authorization": user1Token,
+			},
+			ExpectedStatus:  400,
+			ExpectedContent: []string{"Bad data"},
+			TestAppFactory:  testAppFactory,
+			Body: jsonReader(map[string]any{
+				"invalidField": "this should cause validation error",
+				"threshold":    "not a number",
+			}),
+		},
+		{
+			Name:   "POST malformed JSON",
+			Method: http.MethodPost,
+			URL:    "/api/beszel/user-alerts",
+			Headers: map[string]string{
+				"Authorization": user1Token,
+			},
+			ExpectedStatus:  400,
+			ExpectedContent: []string{"Bad data"},
+			TestAppFactory:  testAppFactory,
+			Body:            strings.NewReader(`{"alertType": "cpu", "threshold": 80, "enabled": true,}`),
+		},
+		{
+			Name:   "POST valid alert data multiple systems",
+			Method: http.MethodPost,
+			URL:    "/api/beszel/user-alerts",
+			Headers: map[string]string{
+				"Authorization": user1Token,
+			},
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"success\":true"},
+			TestAppFactory:  testAppFactory,
+			Body: jsonReader(map[string]any{
+				"name":      "CPU",
+				"value":     69,
+				"min":       9,
+				"systems":   []string{system1.Id, system2.Id},
+				"overwrite": false,
+			}),
+			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
+				// check total alerts
+				alerts, _ := app.CountRecords("alerts")
+				assert.EqualValues(t, 2, alerts, "should have 2 alerts")
+				// check alert has correct values
+				matchingAlerts, _ := app.CountRecords("alerts", dbx.HashExp{"name": "CPU", "user": user1.Id, "system": system1.Id, "value": 69, "min": 9})
+				assert.EqualValues(t, 1, matchingAlerts, "should have 1 alert")
+			},
+		},
+		{
+			Name:   "POST valid alert data single system",
+			Method: http.MethodPost,
+			URL:    "/api/beszel/user-alerts",
+			Headers: map[string]string{
+				"Authorization": user1Token,
+			},
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"success\":true"},
+			TestAppFactory:  testAppFactory,
+			Body: jsonReader(map[string]any{
+				"name":    "Memory",
+				"systems": []string{system1.Id},
+				"value":   90,
+				"min":     10,
+			}),
+			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
+				user1Alerts, _ := app.CountRecords("alerts", dbx.HashExp{"user": user1.Id})
+				assert.EqualValues(t, 3, user1Alerts, "should have 3 alerts")
+			},
+		},
+		{
+			Name:   "POST ignores systems the user cannot access",
+			Method: http.MethodPost,
+			URL:    "/api/beszel/user-alerts",
+			Headers: map[string]string{
+				"Authorization": user2Token,
+			},
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"success\":true"},
+			TestAppFactory:  testAppFactory,
+			Body: jsonReader(map[string]any{
+				"name":    "CPU",
+				"systems": []string{system1.Id},
+				"value":   90,
+				"min":     10,
+			}),
+			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
+				beszelTests.ClearCollection(t, app, "alerts")
+			},
+			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
+				alerts, _ := app.CountRecords("alerts")
+				assert.Zero(t, alerts)
+			},
+		},
+		{
+			Name:   "Overwrite: false, should not overwrite existing alert",
+			Method: http.MethodPost,
+			URL:    "/api/beszel/user-alerts",
+			Headers: map[string]string{
+				"Authorization": user1Token,
+			},
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"success\":true"},
+			TestAppFactory:  testAppFactory,
+			Body: jsonReader(map[string]any{
+				"name":      "CPU",
+				"value":     45,
+				"min":       5,
+				"systems":   []string{system1.Id},
+				"overwrite": false,
+			}),
+			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
+				beszelTests.ClearCollection(t, app, "alerts")
+				beszelTests.CreateRecord(app, "alerts", map[string]any{
+					"name":   "CPU",
+					"system": system1.Id,
+					"user":   user1.Id,
+					"value":  80,
+					"min":    10,
+				})
+			},
+			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
+				alerts, _ := app.CountRecords("alerts")
+				assert.EqualValues(t, 1, alerts, "should have 1 alert")
+				alert, _ := app.FindFirstRecordByFilter("alerts", "name = 'CPU' && user = {:user}", dbx.Params{"user": user1.Id})
+				assert.EqualValues(t, 80, alert.Get("value"), "should have 80 as value")
+			},
+		},
+		{
+			Name:   "Overwrite: true, should overwrite existing alert",
+			Method: http.MethodPost,
+			URL:    "/api/beszel/user-alerts",
+			Headers: map[string]string{
+				"Authorization": user2Token,
+			},
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"success\":true"},
+			TestAppFactory:  testAppFactory,
+			Body: jsonReader(map[string]any{
+				"name":      "CPU",
+				"value":     45,
+				"min":       5,
+				"systems":   []string{system2.Id},
+				"overwrite": true,
+			}),
+			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
+				beszelTests.ClearCollection(t, app, "alerts")
+				beszelTests.CreateRecord(app, "alerts", map[string]any{
+					"name":   "CPU",
+					"system": system2.Id,
+					"user":   user2.Id,
+					"value":  80,
+					"min":    10,
+				})
+			},
+			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
+				alerts, _ := app.CountRecords("alerts")
+				assert.EqualValues(t, 1, alerts, "should have 1 alert")
+				alert, _ := app.FindFirstRecordByFilter("alerts", "name = 'CPU' && user = {:user}", dbx.Params{"user": user2.Id})
+				assert.EqualValues(t, 45, alert.Get("value"), "should have 45 as value")
+			},
+		},
+		{
+			Name:            "DELETE no auth",
+			Method:          http.MethodDelete,
+			URL:             "/api/beszel/user-alerts",
+			ExpectedStatus:  401,
+			ExpectedContent: []string{"requires valid"},
+			TestAppFactory:  testAppFactory,
+			Body: jsonReader(map[string]any{
+				"name":    "CPU",
+				"systems": []string{system1.Id},
+			}),
+			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
+				beszelTests.ClearCollection(t, app, "alerts")
+				beszelTests.CreateRecord(app, "alerts", map[string]any{
+					"name":   "CPU",
+					"system": system1.Id,
+					"user":   user1.Id,
+					"value":  80,
+					"min":    10,
+				})
+			},
+			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
+				alerts, _ := app.CountRecords("alerts")
+				assert.EqualValues(t, 1, alerts, "should have 1 alert")
+			},
+		},
+		{
+			Name:   "DELETE alert",
+			Method: http.MethodDelete,
+			URL:    "/api/beszel/user-alerts",
+			Headers: map[string]string{
+				"Authorization": user1Token,
+			},
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"count\":1", "\"success\":true"},
+			TestAppFactory:  testAppFactory,
+			Body: jsonReader(map[string]any{
+				"name":    "CPU",
+				"systems": []string{system1.Id},
+			}),
+			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
+				beszelTests.ClearCollection(t, app, "alerts")
+				beszelTests.CreateRecord(app, "alerts", map[string]any{
+					"name":   "CPU",
+					"system": system1.Id,
+					"user":   user1.Id,
+					"value":  80,
+					"min":    10,
+				})
+			},
+			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
+				alerts, _ := app.CountRecords("alerts")
+				assert.Zero(t, alerts, "should have 0 alerts")
+			},
+		},
+		{
+			Name:   "DELETE alert multiple systems",
+			Method: http.MethodDelete,
+			URL:    "/api/beszel/user-alerts",
+			Headers: map[string]string{
+				"Authorization": user1Token,
+			},
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"count\":2", "\"success\":true"},
+			TestAppFactory:  testAppFactory,
+			Body: jsonReader(map[string]any{
+				"name":    "Memory",
+				"systems": []string{system1.Id, system2.Id},
+			}),
+			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
+				beszelTests.ClearCollection(t, app, "alerts")
+				for _, systemId := range []string{system1.Id, system2.Id} {
+					_, err := beszelTests.CreateRecord(app, "alerts", map[string]any{
+						"name":   "Memory",
+						"system": systemId,
+						"user":   user1.Id,
+						"value":  90,
+						"min":    10,
+					})
+					assert.NoError(t, err, "should create alert")
+				}
+				alerts, _ := app.CountRecords("alerts")
+				assert.EqualValues(t, 2, alerts, "should have 2 alerts")
+			},
+			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
+				alerts, _ := app.CountRecords("alerts")
+				assert.Zero(t, alerts, "should have 0 alerts")
+			},
+		},
+		{
+			Name:   "DELETE ignores systems the user cannot access",
+			Method: http.MethodDelete,
+			URL:    "/api/beszel/user-alerts",
+			Headers: map[string]string{
+				"Authorization": user2Token,
+			},
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"count\":0", "\"success\":true"},
+			TestAppFactory:  testAppFactory,
+			Body: jsonReader(map[string]any{
+				"name":    "CPU",
+				"systems": []string{system1.Id},
+			}),
+			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
+				beszelTests.ClearCollection(t, app, "alerts")
+				beszelTests.CreateRecord(app, "alerts", map[string]any{
+					"name": "CPU", "system": system1.Id, "user": user2.Id, "value": 80,
+				})
+			},
+			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
+				alerts, _ := app.CountRecords("alerts")
+				assert.EqualValues(t, 1, alerts)
+			},
+		},
+		{
+			Name:   "User 2 should not be able to delete alert of user 1",
+			Method: http.MethodDelete,
+			URL:    "/api/beszel/user-alerts",
+			Headers: map[string]string{
+				"Authorization": user2Token,
+			},
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"count\":1", "\"success\":true"},
+			TestAppFactory:  testAppFactory,
+			Body: jsonReader(map[string]any{
+				"name":    "CPU",
+				"systems": []string{system2.Id},
+			}),
+			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
+				beszelTests.ClearCollection(t, app, "alerts")
+				for _, user := range []string{user1.Id, user2.Id} {
+					beszelTests.CreateRecord(app, "alerts", map[string]any{
+						"name":   "CPU",
+						"system": system2.Id,
+						"user":   user,
+						"value":  80,
+						"min":    10,
+					})
+				}
+				alerts, _ := app.CountRecords("alerts")
+				assert.EqualValues(t, 2, alerts, "should have 2 alerts")
+				user1AlertCount, _ := app.CountRecords("alerts", dbx.HashExp{"user": user1.Id})
+				assert.EqualValues(t, 1, user1AlertCount, "should have 1 alert")
+				user2AlertCount, _ := app.CountRecords("alerts", dbx.HashExp{"user": user2.Id})
+				assert.EqualValues(t, 1, user2AlertCount, "should have 1 alert")
+			},
+			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
+				user1AlertCount, _ := app.CountRecords("alerts", dbx.HashExp{"user": user1.Id})
+				assert.EqualValues(t, 1, user1AlertCount, "should have 1 alert")
+				user2AlertCount, _ := app.CountRecords("alerts", dbx.HashExp{"user": user2.Id})
+				assert.Zero(t, user2AlertCount, "should have 0 alerts")
+			},
+		},
+	}
+
+	for _, scenario := range scenarios {
+		scenario.Test(t)
+	}
+}
+func TestSendTestNotification(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	userToken, err := user.NewAuthToken()
+
+	adminUser, err := beszelTests.CreateUserWithRole(hub, "admin@example.com", "password123", "admin")
+	assert.NoError(t, err, "Failed to create admin user")
+	adminUserToken, err := adminUser.NewAuthToken()
+
+	superuser, err := beszelTests.CreateSuperuser(hub, "superuser@example.com", "password123")
+	assert.NoError(t, err, "Failed to create superuser")
+	superuserToken, err := superuser.NewAuthToken()
+	assert.NoError(t, err, "Failed to create superuser auth token")
+
+	testAppFactory := func(t testing.TB) *pbTests.TestApp {
+		return hub.TestApp
+	}
+
+	scenarios := []beszelTests.ApiScenario{
+		{
+			Name:            "POST /test-notification - no auth should fail",
+			Method:          http.MethodPost,
+			URL:             "/api/beszel/test-notification",
+			ExpectedStatus:  401,
+			ExpectedContent: []string{"requires valid"},
+			TestAppFactory:  testAppFactory,
+			Body: jsonReader(map[string]any{
+				"url": "generic://127.0.0.1",
+			}),
+		},
+		{
+			Name:           "POST /test-notification - with external auth should succeed",
+			Method:         http.MethodPost,
+			URL:            "/api/beszel/test-notification",
+			TestAppFactory: testAppFactory,
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			Body: jsonReader(map[string]any{
+				"url": "generic://8.8.8.8",
+			}),
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"err\":"},
+		},
+		{
+			Name:           "POST /test-notification - local url with user auth should fail",
+			Method:         http.MethodPost,
+			URL:            "/api/beszel/test-notification",
+			TestAppFactory: testAppFactory,
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			Body: jsonReader(map[string]any{
+				"url": "generic://localhost:8010",
+			}),
+			ExpectedStatus:  403,
+			ExpectedContent: []string{"Only admins"},
+		},
+		{
+			Name:           "POST /test-notification - internal url with user auth should fail",
+			Method:         http.MethodPost,
+			URL:            "/api/beszel/test-notification",
+			TestAppFactory: testAppFactory,
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			Body: jsonReader(map[string]any{
+				"url": "generic+http://192.168.0.5",
+			}),
+			ExpectedStatus:  403,
+			ExpectedContent: []string{"Only admins"},
+		},
+		{
+			Name:           "POST /test-notification - internal url with admin auth should succeed",
+			Method:         http.MethodPost,
+			URL:            "/api/beszel/test-notification",
+			TestAppFactory: testAppFactory,
+			Headers: map[string]string{
+				"Authorization": adminUserToken,
+			},
+			Body: jsonReader(map[string]any{
+				"url": "generic://127.0.0.1",
+			}),
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"err\":"},
+		},
+		{
+			Name:           "POST /test-notification - internal url with superuser auth should succeed",
+			Method:         http.MethodPost,
+			URL:            "/api/beszel/test-notification",
+			TestAppFactory: testAppFactory,
+			Headers: map[string]string{
+				"Authorization": superuserToken,
+			},
+			Body: jsonReader(map[string]any{
+				"url": "generic://127.0.0.1",
+			}),
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"err\":"},
+		},
+	}
+
+	for _, scenario := range scenarios {
+		scenario.Test(t)
+	}
+}

internal/alerts/alerts_cache.go

@@ -1,6 +1,8 @@
 package alerts
 
 import (
+	"time"
+
 	"github.com/pocketbase/dbx"
 	"github.com/pocketbase/pocketbase/core"
 	"github.com/pocketbase/pocketbase/tools/store"
@@ -8,13 +10,14 @@ import (
 
 // CachedAlertData represents the relevant fields of an alert record for status checking and updates.
 type CachedAlertData struct {
-	Id        string
-	SystemID  string
-	UserID    string
-	Name      string
-	Value     float64
-	Triggered bool
-	Min       uint8
+	Id           string
+	SystemID     string
+	UserID       string
+	Name         string
+	Value        float64
+	Triggered    bool
+	Min          uint8
+	PendingSince time.Time
 	// Created   types.DateTime
 }
 
@@ -26,6 +29,7 @@ func (a *CachedAlertData) PopulateFromRecord(record *core.Record) {
 	a.Value = record.GetFloat("value")
 	a.Triggered = record.GetBool("triggered")
 	a.Min = uint8(record.GetInt("min"))
+	a.PendingSince = record.GetDateTime("pending_since").Time()
 	// a.Created = record.GetDateTime("created")
 }
 

internal/alerts/alerts_container.go

@@ -0,0 +1,318 @@
+package alerts
+
+import (
+	"errors"
+	"fmt"
+	"strings"
+	"time"
+
+	"github.com/henrygd/beszel/internal/entities/container"
+	"github.com/henrygd/beszel/internal/entities/system"
+
+	"github.com/pocketbase/pocketbase/core"
+)
+
+const (
+	// containerAlertName is the value stored in the alerts.name field for this alert type.
+	containerAlertName = "ContainerHealth"
+
+	// containerLogMaxLines caps how many matched (error/fatal) log lines are kept.
+	containerLogMaxLines = 12
+	// containerLogFallbackLines is how many trailing raw log lines are used when no
+	// line matches "error" or "fatal", so the notification still carries some context.
+	containerLogFallbackLines = 6
+	// containerLogExcerptMaxChars bounds a single container's log excerpt so a
+	// handful of containers can't blow past Discord's message size limit.
+	containerLogExcerptMaxChars = 500
+	// containerAlertMaxLogged is the max number of unhealthy containers we fetch
+	// and embed logs for in a single alert message.
+	containerAlertMaxLogged = 2
+	// containerAlertMessageMaxChars is a final safety cap on the whole message body.
+	containerAlertMessageMaxChars = 1800
+)
+
+// FetchContainerLogsFunc retrieves recent logs for a container ID from its
+// connected agent. Implementations should apply their own timeout. This is a
+// type alias (not a defined type) so it satisfies the hubLike interface in
+// internal/hub/systems, which declares the same func signature without
+// importing this package.
+type FetchContainerLogsFunc = func(containerID string) (string, error)
+
+// containerAlertTarget is an immutable snapshot of the fields needed after the
+// alert fires. Keeping agent-owned container records out of notification work
+// avoids retaining and concurrently reading data that is refreshed in place.
+type containerAlertTarget struct {
+	id   string
+	name string
+}
+
+// HandleContainerAlerts checks configured "ContainerHealth" alerts for a system
+// against the Docker container health data included in the latest agent update.
+// It persists when containers first become unhealthy, fires from a fresh poll
+// once the configured delay has elapsed, and resolves once containers recover.
+// fetchLogs is used when an alert actually fires so the notification can include
+// a log excerpt (prioritizing lines containing "error"/"fatal") for context.
+func (am *AlertManager) HandleContainerAlerts(systemRecord *core.Record, data *system.CombinedData, fetchLogs FetchContainerLogsFunc) error {
+	alerts := am.alertsCache.GetAlertsByName(systemRecord.Id, containerAlertName)
+	if len(alerts) == 0 {
+		return nil
+	}
+	if data.Containers == nil {
+		// An unknown Docker state must not resolve a triggered alert or count
+		// toward the minimum unhealthy duration.
+		var result error
+		for _, alertData := range alerts {
+			if err := am.clearPendingContainerAlert(alertData); err != nil {
+				result = errors.Join(result, err)
+			}
+		}
+		return result
+	}
+
+	var unhealthy []*container.Stats
+	for _, c := range data.Containers {
+		if c.Health == container.DockerHealthUnhealthy {
+			unhealthy = append(unhealthy, c)
+		}
+	}
+
+	systemName := systemRecord.GetString("name")
+	now := time.Now().UTC()
+	var result error
+	for _, alertData := range alerts {
+		if len(unhealthy) > 0 {
+			if alertData.Triggered {
+				continue
+			}
+			min := max(1, int(alertData.Min))
+			if alertData.PendingSince.IsZero() {
+				pendingSince, err := am.setPendingContainerAlert(alertData, now)
+				if err != nil {
+					result = errors.Join(result, err)
+					continue
+				}
+				if pendingSince.IsZero() {
+					continue
+				}
+				alertData.PendingSince = pendingSince
+				if min > 1 {
+					continue
+				}
+			}
+			if min > 1 && now.Before(alertData.PendingSince.Add(time.Duration(min)*time.Minute)) {
+				continue
+			}
+			if err := am.sendContainerHealthAlert(true, systemName, alertData, snapshotContainerAlertTargets(unhealthy), fetchLogs); err != nil {
+				result = errors.Join(result, err)
+			}
+			continue
+		}
+
+		// no unhealthy containers right now
+		if err := am.clearPendingContainerAlert(alertData); err != nil {
+			result = errors.Join(result, err)
+		}
+		if !alertData.Triggered {
+			continue
+		}
+		if err := am.sendContainerHealthAlert(false, systemName, alertData, nil, fetchLogs); err != nil {
+			result = errors.Join(result, err)
+		}
+	}
+	return result
+}
+
+func snapshotContainerAlertTargets(containers []*container.Stats) []containerAlertTarget {
+	targets := make([]containerAlertTarget, len(containers))
+	for i, c := range containers {
+		targets[i] = containerAlertTarget{id: c.Id, name: c.Name}
+	}
+	return targets
+}
+
+// setPendingContainerAlert durably records the first unhealthy observation and
+// returns the persisted generation used to claim delivery.
+func (am *AlertManager) setPendingContainerAlert(alertData CachedAlertData, since time.Time) (time.Time, error) {
+	record, err := am.hub.FindRecordById("alerts", alertData.Id)
+	if err != nil {
+		return time.Time{}, err
+	}
+	if record.GetBool("triggered") {
+		return time.Time{}, nil
+	}
+	if pendingSince := record.GetDateTime("pending_since").Time(); !pendingSince.IsZero() {
+		return pendingSince, nil
+	}
+	// PocketBase date fields are persisted with millisecond precision. Normalize
+	// before saving so the update-hook cache and a subsequent database read agree.
+	since = since.Truncate(time.Millisecond)
+	record.Set("pending_since", since)
+	return since, am.hub.Save(record)
+}
+
+func (am *AlertManager) clearPendingContainerAlert(alertData CachedAlertData) error {
+	if alertData.PendingSince.IsZero() {
+		return nil
+	}
+	record, err := am.hub.FindRecordById("alerts", alertData.Id)
+	if err != nil {
+		return err
+	}
+	if record.GetDateTime("pending_since").Time().IsZero() {
+		return nil
+	}
+	record.Set("pending_since", nil)
+	return am.hub.Save(record)
+}
+
+// claimPendingContainerAlert marks an alert triggered only if the pending
+// generation is still current. A healthy/unknown update can clear the timestamp
+// while logs are being fetched, causing this claim to become a no-op.
+func (am *AlertManager) claimPendingContainerAlert(alertData CachedAlertData) (bool, error) {
+	record, err := am.hub.FindRecordById("alerts", alertData.Id)
+	if err != nil {
+		return false, err
+	}
+	pendingSince := record.GetDateTime("pending_since").Time()
+	if record.GetBool("triggered") || pendingSince.IsZero() || pendingSince.UnixMilli() != alertData.PendingSince.UnixMilli() {
+		return false, nil
+	}
+	record.Set("pending_since", nil)
+	record.Set("triggered", true)
+	return true, am.hub.Save(record)
+}
+
+// CancelPendingContainerAlerts clears pending container-health durations for a
+// system. Called when monitoring pauses or the system goes down.
+func (am *AlertManager) CancelPendingContainerAlerts(systemID string) {
+	for _, alertData := range am.alertsCache.GetAlertsByName(systemID, containerAlertName) {
+		if err := am.clearPendingContainerAlert(alertData); err != nil {
+			am.hub.Logger().Error("Failed to clear pending container alert", "err", err)
+		}
+	}
+}
+
+// sendContainerHealthAlert updates the alert's triggered state and sends the
+// notification. When unhealthy is true, it embeds a log excerpt (prioritizing
+// error/fatal lines) for up to containerAlertMaxLogged of the affected containers.
+func (am *AlertManager) sendContainerHealthAlert(unhealthy bool, systemName string, alertData CachedAlertData, containers []containerAlertTarget, fetchLogs FetchContainerLogsFunc) error {
+	link := am.hub.MakeLink("system", alertData.SystemID)
+	linkText := "View " + systemName
+
+	if !unhealthy {
+		if err := am.setAlertTriggered(alertData, false); err != nil {
+			return err
+		}
+		title := fmt.Sprintf("%s containers are healthy ✅", systemName)
+		return am.SendAlert(AlertMessageData{
+			UserID:   alertData.UserID,
+			SystemID: alertData.SystemID,
+			Title:    title,
+			Message:  strings.TrimSuffix(title, " ✅"),
+			Link:     link,
+			LinkText: linkText,
+		})
+	}
+
+	names := make([]string, len(containers))
+	for i, c := range containers {
+		names[i] = c.name
+	}
+
+	var title string
+	if len(names) == 1 {
+		title = fmt.Sprintf("Unhealthy container %s on %s \U0001F534", names[0], systemName)
+	} else {
+		title = fmt.Sprintf("%d unhealthy containers on %s \U0001F534", len(names), systemName)
+	}
+
+	var body strings.Builder
+	fmt.Fprintf(&body, "Unhealthy: %s", strings.Join(names, ", "))
+	body.WriteString(am.buildContainerLogsSection(containers, fetchLogs))
+
+	message := body.String()
+	if len(message) > containerAlertMessageMaxChars {
+		message = message[:containerAlertMessageMaxChars] + "\n…(truncated)"
+	}
+
+	claimed, err := am.claimPendingContainerAlert(alertData)
+	if err != nil || !claimed {
+		return err
+	}
+
+	return am.SendAlert(AlertMessageData{
+		UserID:   alertData.UserID,
+		SystemID: alertData.SystemID,
+		Title:    title,
+		Message:  message,
+		Link:     link,
+		LinkText: linkText,
+	})
+}
+
+// buildContainerLogsSection attempts to fetch and format log excerpts for up to
+// containerAlertMaxLogged unhealthy containers, to append to an alert message.
+func (am *AlertManager) buildContainerLogsSection(containers []containerAlertTarget, fetchLogs FetchContainerLogsFunc) string {
+	if fetchLogs == nil {
+		return ""
+	}
+
+	var section strings.Builder
+	attempts := min(len(containers), containerAlertMaxLogged)
+	for _, c := range containers[:attempts] {
+		rawLogs, err := fetchLogs(c.id)
+		if err != nil {
+			am.hub.Logger().Warn("Failed to fetch container logs for alert", "container", c.name, "err", err)
+			continue
+		}
+		excerpt := buildContainerLogExcerpt(rawLogs)
+		if excerpt == "" {
+			continue
+		}
+		fmt.Fprintf(&section, "\n\n%s logs:\n```\n%s\n```", c.name, excerpt)
+	}
+
+	if len(containers) > containerAlertMaxLogged {
+		fmt.Fprintf(&section, "\n\n(+%d more unhealthy container(s), logs omitted)", len(containers)-containerAlertMaxLogged)
+	}
+
+	return section.String()
+}
+
+// buildContainerLogExcerpt filters raw container log output down to the lines
+// most likely to explain why the container is unhealthy: lines containing
+// "error" or "fatal" (case-insensitive) are preferred. If none match, the tail
+// of the raw output is used instead so the notification still carries context.
+func buildContainerLogExcerpt(raw string) string {
+	raw = strings.TrimSpace(raw)
+	if raw == "" {
+		return ""
+	}
+	lines := strings.Split(raw, "\n")
+
+	var matched []string
+	for _, line := range lines {
+		line = strings.TrimRight(line, "\r")
+		if line == "" {
+			continue
+		}
+		lower := strings.ToLower(line)
+		if strings.Contains(lower, "error") || strings.Contains(lower, "fatal") {
+			matched = append(matched, line)
+		}
+	}
+
+	selected := matched
+	if len(selected) == 0 {
+		start := max(0, len(lines)-containerLogFallbackLines)
+		selected = lines[start:]
+	} else if len(selected) > containerLogMaxLines {
+		selected = selected[len(selected)-containerLogMaxLines:]
+	}
+
+	excerpt := strings.TrimSpace(strings.Join(selected, "\n"))
+	if len(excerpt) > containerLogExcerptMaxChars {
+		excerpt = "…" + excerpt[len(excerpt)-containerLogExcerptMaxChars:]
+	}
+	return excerpt
+}

internal/alerts/alerts_container_test.go

@@ -0,0 +1,349 @@
+//go:build testing
+
+package alerts_test
+
+import (
+	"fmt"
+	"strings"
+	"testing"
+	"testing/synctest"
+	"time"
+
+	"github.com/henrygd/beszel/internal/alerts"
+	"github.com/henrygd/beszel/internal/entities/container"
+	"github.com/henrygd/beszel/internal/entities/system"
+	beszelTests "github.com/henrygd/beszel/internal/tests"
+	"github.com/pocketbase/pocketbase/core"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+type containerAlertTestFixture struct {
+	hub          *beszelTests.TestHub
+	am           *alerts.AlertManager
+	alertID      string
+	systemRecord *core.Record
+}
+
+func newContainerAlertTestFixture(t *testing.T, min int) *containerAlertTestFixture {
+	t.Helper()
+
+	hub, user := beszelTests.GetHubWithUser(t)
+
+	systems, err := beszelTests.CreateSystems(hub, 1, user.Id, "up")
+	require.NoError(t, err)
+	systemRecord := systems[0]
+
+	userSettings, err := hub.FindFirstRecordByFilter("user_settings", "user={:user}", map[string]any{"user": user.Id})
+	require.NoError(t, err)
+	userSettings.Set("settings", `{"emails":["test@example.com"],"webhooks":[]}`)
+	require.NoError(t, hub.Save(userSettings))
+
+	alertRecord, err := beszelTests.CreateRecord(hub, "alerts", map[string]any{
+		"name":   "ContainerHealth",
+		"system": systemRecord.Id,
+		"user":   user.Id,
+		"min":    min,
+	})
+	require.NoError(t, err)
+	assert.False(t, alertRecord.GetBool("triggered"), "Alert should not be triggered initially")
+
+	return &containerAlertTestFixture{
+		hub:          hub,
+		am:           alerts.NewTestAlertManagerWithoutWorker(hub),
+		alertID:      alertRecord.Id,
+		systemRecord: systemRecord,
+	}
+}
+
+func (f *containerAlertTestFixture) cleanup() {
+	f.hub.Cleanup()
+}
+
+func (f *containerAlertTestFixture) submit(t *testing.T, containers []*container.Stats, fetchLogs alerts.FetchContainerLogsFunc) {
+	t.Helper()
+	data := &system.CombinedData{Containers: containers}
+	require.NoError(t, f.am.HandleContainerAlerts(f.systemRecord, data, fetchLogs))
+}
+
+func (f *containerAlertTestFixture) submitInvalid(t *testing.T) {
+	t.Helper()
+	require.NoError(t, f.am.HandleContainerAlerts(f.systemRecord, &system.CombinedData{}, nil))
+}
+
+func (f *containerAlertTestFixture) assertTriggered(t *testing.T, triggered bool, message string) {
+	t.Helper()
+	alertRecord, err := f.hub.FindRecordById("alerts", f.alertID)
+	require.NoError(t, err)
+	assert.Equal(t, triggered, alertRecord.GetBool("triggered"), message)
+}
+
+func (f *containerAlertTestFixture) assertPending(t *testing.T, pending bool) {
+	t.Helper()
+	alertRecord, err := f.hub.FindRecordById("alerts", f.alertID)
+	require.NoError(t, err)
+	assert.Equal(t, pending, !alertRecord.GetDateTime("pending_since").Time().IsZero())
+}
+
+func waitForContainerAlert(d time.Duration) {
+	time.Sleep(d)
+	synctest.Wait()
+}
+
+func healthyContainer(name string) *container.Stats {
+	return &container.Stats{Name: name, Id: "abc123def456", Health: container.DockerHealthHealthy}
+}
+
+func unhealthyContainer(name string) *container.Stats {
+	return &container.Stats{Name: name, Id: "abc123def456", Health: container.DockerHealthUnhealthy}
+}
+
+func TestContainerHealthAlertTriggersAndResolves(t *testing.T) {
+	fixture := newContainerAlertTestFixture(t, 1)
+	defer fixture.cleanup()
+
+	synctest.Test(t, func(t *testing.T) {
+		fixture.submit(t, []*container.Stats{unhealthyContainer("web")}, nil)
+
+		fixture.assertTriggered(t, true, "A one-minute alert should trigger on the first unhealthy update")
+		require.Equal(t, 1, fixture.hub.TestMailer.TotalSend(), "An email should have been sent")
+
+		msg := fixture.hub.TestMailer.LastMessage()
+		assert.Contains(t, msg.Subject, "web", "Subject should name the unhealthy container")
+		assert.Contains(t, strings.ToLower(msg.Subject), "unhealthy")
+
+		fixture.submit(t, []*container.Stats{unhealthyContainer("web")}, nil)
+		fixture.assertPending(t, false)
+
+		fixture.submitInvalid(t)
+		fixture.assertTriggered(t, true, "An invalid container snapshot should not resolve the alert")
+		assert.Equal(t, 1, fixture.hub.TestMailer.TotalSend(), "An invalid snapshot should not send a recovery")
+
+		fixture.submit(t, []*container.Stats{}, nil)
+		waitForContainerAlert(time.Second)
+
+		fixture.assertTriggered(t, false, "Alert should resolve once the container is healthy again")
+		assert.Equal(t, 2, fixture.hub.TestMailer.TotalSend(), "A second email should have been sent for the recovery")
+		assert.Contains(t, fixture.hub.TestMailer.LastMessage().Subject, " healthy")
+	})
+}
+
+func TestContainerHealthAlertInvalidSnapshotCancelsPending(t *testing.T) {
+	fixture := newContainerAlertTestFixture(t, 5)
+	defer fixture.cleanup()
+
+	synctest.Test(t, func(t *testing.T) {
+		fixture.submit(t, []*container.Stats{unhealthyContainer("db")}, nil)
+		fixture.assertPending(t, true)
+		waitForContainerAlert(time.Minute)
+		fixture.submitInvalid(t)
+		fixture.assertPending(t, false)
+		waitForContainerAlert(10 * time.Minute)
+		fixture.submit(t, []*container.Stats{unhealthyContainer("db")}, nil)
+
+		fixture.assertTriggered(t, false, "Stale unhealthy data should not trigger an alert")
+		fixture.assertPending(t, true)
+		assert.Equal(t, 0, fixture.hub.TestMailer.TotalSend())
+	})
+}
+
+func TestContainerHealthAlertSystemDownCancelsPending(t *testing.T) {
+	fixture := newContainerAlertTestFixture(t, 5)
+	defer fixture.cleanup()
+
+	// Use the hub's alert manager because the system-manager status hook invokes
+	// cancellation on that instance.
+	am := fixture.hub.GetAlertManager()
+	require.NoError(t, am.HandleContainerAlerts(
+		fixture.systemRecord,
+		&system.CombinedData{Containers: []*container.Stats{unhealthyContainer("db")}},
+		nil,
+	))
+	fixture.assertPending(t, true)
+
+	fixture.systemRecord.Set("status", "down")
+	require.NoError(t, fixture.hub.Save(fixture.systemRecord))
+
+	fixture.assertPending(t, false)
+}
+
+func TestContainerHealthAlertResolvesBeforeMinDelayCancelsPending(t *testing.T) {
+	fixture := newContainerAlertTestFixture(t, 5)
+	defer fixture.cleanup()
+
+	synctest.Test(t, func(t *testing.T) {
+		fixture.submit(t, []*container.Stats{unhealthyContainer("db")}, nil)
+		waitForContainerAlert(time.Minute)
+
+		fixture.assertTriggered(t, false, "Alert should not fire until the min delay elapses")
+		fixture.assertPending(t, true)
+		assert.Equal(t, 0, fixture.hub.TestMailer.TotalSend())
+
+		// container recovers before the 5 minute delay elapses
+		fixture.submit(t, []*container.Stats{healthyContainer("db")}, nil)
+		waitForContainerAlert(10 * time.Minute)
+		fixture.submit(t, []*container.Stats{healthyContainer("db")}, nil)
+
+		fixture.assertTriggered(t, false, "Alert should remain untriggered")
+		fixture.assertPending(t, false)
+		assert.Equal(t, 0, fixture.hub.TestMailer.TotalSend(), "No email should be sent for a container that recovered before the delay")
+	})
+}
+
+func TestContainerHealthAlertPreservesPendingDurationAcrossManagerRestart(t *testing.T) {
+	fixture := newContainerAlertTestFixture(t, 2)
+	defer fixture.cleanup()
+
+	synctest.Test(t, func(t *testing.T) {
+		fixture.submit(t, []*container.Stats{unhealthyContainer("db")}, nil)
+		waitForContainerAlert(30 * time.Second)
+
+		restarted := alerts.NewTestAlertManagerWithoutWorker(fixture.hub)
+		waitForContainerAlert(91 * time.Second)
+		require.NoError(t, restarted.HandleContainerAlerts(
+			fixture.systemRecord,
+			&system.CombinedData{Containers: []*container.Stats{unhealthyContainer("db")}},
+			nil,
+		))
+
+		fixture.assertTriggered(t, true, "Restart should preserve the original unhealthy start time")
+		fixture.assertPending(t, false)
+		assert.Equal(t, 1, fixture.hub.TestMailer.TotalSend())
+	})
+}
+
+func TestContainerHealthAlertClaimsPendingTimestampAtDatabasePrecision(t *testing.T) {
+	fixture := newContainerAlertTestFixture(t, 1)
+	defer fixture.cleanup()
+
+	alertRecord, err := fixture.hub.FindRecordById("alerts", fixture.alertID)
+	require.NoError(t, err)
+	// PocketBase persists dates to milliseconds, while record update hooks can
+	// retain the original sub-millisecond value in the in-memory alert cache.
+	alertRecord.Set("pending_since", time.Now().UTC().Add(-2*time.Minute).Truncate(time.Millisecond).Add(123*time.Nanosecond))
+	require.NoError(t, fixture.hub.Save(alertRecord))
+
+	fixture.submit(t, []*container.Stats{unhealthyContainer("db")}, nil)
+
+	fixture.assertTriggered(t, true, "Equivalent persisted and cached timestamps should claim the alert")
+	fixture.assertPending(t, false)
+	assert.Equal(t, 1, fixture.hub.TestMailer.TotalSend())
+}
+
+func TestContainerHealthAlertRecoveryWhileFetchingLogsCancelsDelivery(t *testing.T) {
+	fixture := newContainerAlertTestFixture(t, 1)
+	defer fixture.cleanup()
+
+	synctest.Test(t, func(t *testing.T) {
+		fetchLogs := func(containerID string) (string, error) {
+			fixture.submit(t, []*container.Stats{healthyContainer("api")}, nil)
+			return "FATAL stale failure", nil
+		}
+		fixture.submit(t, []*container.Stats{unhealthyContainer("api")}, fetchLogs)
+
+		fixture.assertTriggered(t, false, "Recovery should cancel delivery while logs are fetched")
+		fixture.assertPending(t, false)
+		assert.Equal(t, 0, fixture.hub.TestMailer.TotalSend())
+	})
+}
+
+func TestContainerHealthAlertIncludesLogExcerpt(t *testing.T) {
+	fixture := newContainerAlertTestFixture(t, 1)
+	defer fixture.cleanup()
+
+	rawLogs := strings.Join([]string{
+		"2026-08-16T10:00:00Z booting",
+		"2026-08-16T10:00:01Z ERROR could not reach upstream",
+		"2026-08-16T10:00:02Z FATAL giving up after 3 retries",
+	}, "\n")
+	fetchLogs := func(containerID string) (string, error) {
+		assert.Equal(t, "abc123def456", containerID)
+		return rawLogs, nil
+	}
+
+	synctest.Test(t, func(t *testing.T) {
+		fixture.submit(t, []*container.Stats{unhealthyContainer("api")}, fetchLogs)
+
+		fixture.assertTriggered(t, true, "Alert should be triggered")
+		require.Equal(t, 1, fixture.hub.TestMailer.TotalSend())
+
+		body := fixture.hub.TestMailer.LastMessage().Text
+		assert.Contains(t, body, "could not reach upstream")
+		assert.Contains(t, body, "giving up after 3 retries")
+		assert.NotContains(t, body, "booting", "non error/fatal lines should be dropped when matches exist")
+	})
+}
+
+func TestContainerHealthAlertSkipsLogsOnFetchError(t *testing.T) {
+	fixture := newContainerAlertTestFixture(t, 1)
+	defer fixture.cleanup()
+
+	fetchLogs := func(containerID string) (string, error) {
+		return "", fmt.Errorf("agent unreachable")
+	}
+
+	synctest.Test(t, func(t *testing.T) {
+		fixture.submit(t, []*container.Stats{unhealthyContainer("api")}, fetchLogs)
+
+		fixture.assertTriggered(t, true, "Alert should still be triggered even if logs can't be fetched")
+		require.Equal(t, 1, fixture.hub.TestMailer.TotalSend())
+	})
+}
+
+func TestContainerHealthAlertCapsLogFetchAttempts(t *testing.T) {
+	fixture := newContainerAlertTestFixture(t, 1)
+	defer fixture.cleanup()
+
+	containers := make([]*container.Stats, 100)
+	for i := range containers {
+		containers[i] = &container.Stats{
+			Name:   fmt.Sprintf("container-%d", i),
+			Id:     fmt.Sprintf("id-%d", i),
+			Health: container.DockerHealthUnhealthy,
+		}
+	}
+	attempts := 0
+	fetchLogs := func(containerID string) (string, error) {
+		attempts++
+		return "", fmt.Errorf("agent unreachable")
+	}
+
+	synctest.Test(t, func(t *testing.T) {
+		fixture.submit(t, containers, fetchLogs)
+
+		fixture.assertTriggered(t, true, "Alert should still fire when log retrieval fails")
+		assert.Equal(t, 2, attempts, "Log retrieval should attempt at most two containers")
+		require.Equal(t, 1, fixture.hub.TestMailer.TotalSend())
+	})
+}
+
+func TestBuildContainerLogExcerptPrefersErrorAndFatalLines(t *testing.T) {
+	raw := strings.Join([]string{
+		"2026-08-16T10:00:00Z starting up",
+		"2026-08-16T10:00:01Z listening on :8080",
+		"2026-08-16T10:00:02Z ERROR failed to connect to db",
+		"2026-08-16T10:00:03Z retrying connection",
+		"2026-08-16T10:00:04Z FATAL could not recover, exiting",
+	}, "\n")
+
+	excerpt := alerts.BuildContainerLogExcerpt(raw)
+	assert.Contains(t, excerpt, "failed to connect to db")
+	assert.Contains(t, excerpt, "could not recover, exiting")
+	assert.NotContains(t, excerpt, "starting up", "non-matching lines should be dropped when error/fatal lines exist")
+}
+
+func TestBuildContainerLogExcerptFallsBackToTailWhenNoMatches(t *testing.T) {
+	var lines []string
+	for i := range 20 {
+		lines = append(lines, fmt.Sprintf("line %d: all good here", i))
+	}
+	raw := strings.Join(lines, "\n")
+
+	excerpt := alerts.BuildContainerLogExcerpt(raw)
+	assert.Contains(t, excerpt, "line 19", "should keep the tail of the output")
+	assert.NotContains(t, excerpt, "line 0:", "should not keep the very start when falling back to a short tail")
+}
+
+func TestBuildContainerLogExcerptEmpty(t *testing.T) {
+	assert.Equal(t, "", alerts.BuildContainerLogExcerpt("   \n  \n"))
+}

internal/alerts/alerts_quiet_hours_test.go

@@ -322,8 +322,9 @@ func TestAlertSilencedMultiUser(t *testing.T) {
 }
 
 func TestAlertSilencedWithActualAlert(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+
 	synctest.Test(t, func(t *testing.T) {
-		hub, user := beszelTests.GetHubWithUser(t)
 		defer hub.Cleanup()
 
 		// Create a system

internal/alerts/alerts_status.go

@@ -109,6 +109,18 @@ func (am *AlertManager) cancelPendingAlert(alertID string) bool {
 	return true
 }
 
+// CancelPendingStatusAlerts cancels all pending status alert timers for a given system.
+// This is called when a system is paused to prevent delayed alerts from firing.
+func (am *AlertManager) CancelPendingStatusAlerts(systemID string) {
+	am.pendingAlerts.Range(func(key, value any) bool {
+		info := value.(*alertInfo)
+		if info.alertData.SystemID == systemID {
+			am.cancelPendingAlert(key.(string))
+		}
+		return true
+	})
+}
+
 // processPendingAlert sends a "down" alert if the pending alert has expired and the system is still down.
 func (am *AlertManager) processPendingAlert(alertID string) {
 	value, loaded := am.pendingAlerts.LoadAndDelete(alertID)

internal/alerts/alerts_status_test.go

@@ -29,8 +29,9 @@ func setStatusAlertEmail(t *testing.T, hub core.App, userID, email string) {
 }
 
 func TestStatusAlerts(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+
 	synctest.Test(t, func(t *testing.T) {
-		hub, user := beszelTests.GetHubWithUser(t)
 		defer hub.Cleanup()
 
 		systems, err := beszelTests.CreateSystems(hub, 4, user.Id, "paused")
@@ -234,8 +235,9 @@ func TestHandleStatusAlertsDoesNotSendRecoveryWhileDownIsOnlyPending(t *testing.
 }
 
 func TestStatusAlertTimerCancellationPreventsBoundaryDelivery(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+
 	synctest.Test(t, func(t *testing.T) {
-		hub, user := beszelTests.GetHubWithUser(t)
 		defer hub.Cleanup()
 
 		userSettings, err := hub.FindFirstRecordByFilter("user_settings", "user={:user}", map[string]any{"user": user.Id})
@@ -336,8 +338,9 @@ func TestStatusAlertDownFiresAfterDelayExpires(t *testing.T) {
 }
 
 func TestStatusAlertMultipleUsersRespectDifferentMinutes(t *testing.T) {
+	hub, user1 := beszelTests.GetHubWithUser(t)
+
 	synctest.Test(t, func(t *testing.T) {
-		hub, user1 := beszelTests.GetHubWithUser(t)
 		defer hub.Cleanup()
 
 		setStatusAlertEmail(t, hub, user1.Id, "user1@example.com")
@@ -423,8 +426,9 @@ func TestStatusAlertMultipleUsersRespectDifferentMinutes(t *testing.T) {
 }
 
 func TestStatusAlertMultipleUsersRecoveryBetweenMinutesOnlyAlertsEarlierUser(t *testing.T) {
+	hub, user1 := beszelTests.GetHubWithUser(t)
+
 	synctest.Test(t, func(t *testing.T) {
-		hub, user1 := beszelTests.GetHubWithUser(t)
 		defer hub.Cleanup()
 
 		setStatusAlertEmail(t, hub, user1.Id, "user1@example.com")
@@ -816,8 +820,9 @@ func TestResolveStatusAlerts(t *testing.T) {
 }
 
 func TestAlertsHistoryStatus(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+
 	synctest.Test(t, func(t *testing.T) {
-		hub, user := beszelTests.GetHubWithUser(t)
 		defer hub.Cleanup()
 
 		// Create a system
@@ -882,8 +887,9 @@ func TestAlertsHistoryStatus(t *testing.T) {
 }
 
 func TestStatusAlertClearedBeforeSend(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+
 	synctest.Test(t, func(t *testing.T) {
-		hub, user := beszelTests.GetHubWithUser(t)
 		defer hub.Cleanup()
 
 		// Create a system
@@ -941,3 +947,68 @@ func TestStatusAlertClearedBeforeSend(t *testing.T) {
 		assert.EqualValues(t, 0, alertHistoryCount, "Should have no unresolved alert history records since alert never triggered")
 	})
 }
+
+func TestCancelPendingStatusAlertsClearsAllAlertsForSystem(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	userSettings, err := hub.FindFirstRecordByFilter("user_settings", "user={:user}", map[string]any{"user": user.Id})
+	require.NoError(t, err)
+	userSettings.Set("settings", `{"emails":["test@example.com"],"webhooks":[]}`)
+	require.NoError(t, hub.Save(userSettings))
+
+	systemCollection, err := hub.FindCollectionByNameOrId("systems")
+	require.NoError(t, err)
+
+	system1 := core.NewRecord(systemCollection)
+	system1.Set("name", "system-1")
+	system1.Set("status", "up")
+	system1.Set("host", "127.0.0.1")
+	system1.Set("users", []string{user.Id})
+	require.NoError(t, hub.Save(system1))
+
+	system2 := core.NewRecord(systemCollection)
+	system2.Set("name", "system-2")
+	system2.Set("status", "up")
+	system2.Set("host", "127.0.0.2")
+	system2.Set("users", []string{user.Id})
+	require.NoError(t, hub.Save(system2))
+
+	alertCollection, err := hub.FindCollectionByNameOrId("alerts")
+	require.NoError(t, err)
+
+	alert1 := core.NewRecord(alertCollection)
+	alert1.Set("user", user.Id)
+	alert1.Set("system", system1.Id)
+	alert1.Set("name", "Status")
+	alert1.Set("triggered", false)
+	alert1.Set("min", 5)
+	require.NoError(t, hub.Save(alert1))
+
+	alert2 := core.NewRecord(alertCollection)
+	alert2.Set("user", user.Id)
+	alert2.Set("system", system2.Id)
+	alert2.Set("name", "Status")
+	alert2.Set("triggered", false)
+	alert2.Set("min", 5)
+	require.NoError(t, hub.Save(alert2))
+
+	am := alerts.NewTestAlertManagerWithoutWorker(hub)
+	initialEmailCount := hub.TestMailer.TotalSend()
+
+	// Both systems go down
+	require.NoError(t, am.HandleStatusAlerts("down", system1))
+	require.NoError(t, am.HandleStatusAlerts("down", system2))
+	assert.Equal(t, 2, am.GetPendingAlertsCount(), "both systems should have pending alerts")
+
+	// System 1 is paused — cancel its pending alerts
+	am.CancelPendingStatusAlerts(system1.Id)
+	assert.Equal(t, 1, am.GetPendingAlertsCount(), "only system2 alert should remain pending after pausing system1")
+
+	// Expire and process remaining alerts — only system2 should fire
+	am.ForceExpirePendingAlerts()
+	processed, err := am.ProcessPendingAlerts()
+	require.NoError(t, err)
+	assert.Len(t, processed, 1, "only the non-paused system's alert should be processed")
+	assert.Equal(t, initialEmailCount+1, hub.TestMailer.TotalSend(), "only system2 should send a down notification")
+}

internal/alerts/alerts_system.go

@@ -13,8 +13,41 @@ import (
 	"github.com/pocketbase/pocketbase/tools/types"
 )
 
+var cpuStateAlerts = map[string]struct {
+	index int
+	label string
+}{
+	"CPUIOWait": {2, "CPU I/O Wait"},
+	"CPUSteal":  {3, "CPU Steal Time"},
+}
+
+func cpuStateAlertValue(name string, breakdown []float64) (float64, bool) {
+	state, ok := cpuStateAlerts[name]
+	if !ok || len(breakdown) < 5 {
+		return 0, false
+	}
+	var total float64
+	for _, value := range breakdown {
+		total += value
+	}
+	if total <= 0 {
+		return 0, false
+	}
+	return breakdown[state.index], true
+}
+
 func (am *AlertManager) HandleSystemAlerts(systemRecord *core.Record, data *system.CombinedData) error {
-	alerts := am.alertsCache.GetAlertsExcludingNames(systemRecord.Id, "Status")
+	// Systemd alerts are binary state, not numeric thresholds, so they're handled
+	// separately. They read their own state from the database and don't use data.
+	if err := am.HandleSystemdAlerts(systemRecord); err != nil {
+		am.hub.Logger().Error("Error handling systemd alerts", "err", err)
+	}
+
+	if data == nil {
+		return nil
+	}
+
+	alerts := am.alertsCache.GetAlertsExcludingNames(systemRecord.Id, "Status", alertNameSystemdFailed, containerAlertName)
 	if len(alerts) == 0 {
 		return nil
 	}
@@ -44,6 +77,14 @@ func (am *AlertManager) HandleSystemAlerts(systemRecord *core.Record, data *syst
 					maxUsedPct = usedPct
 				}
 			}
+			for _, pool := range data.Stats.ZfsPools {
+				if pool != nil && pool.Total > 0 {
+					usedPct := pool.Used / pool.Total * 100
+					if usedPct > maxUsedPct {
+						maxUsedPct = usedPct
+					}
+				}
+			}
 			val = maxUsedPct
 		case "Temperature":
 			if data.Info.DashboardTemp < 1 {
@@ -63,10 +104,15 @@ func (am *AlertManager) HandleSystemAlerts(systemRecord *core.Record, data *syst
 		case "GPU":
 			val = data.Info.GpuPct
 		case "Battery":
-			if data.Stats.Battery[0] == 0 {
+			if !hasRepresentativeBattery(data.Stats.Battery, data.Stats.Batteries) {
 				continue
 			}
 			val = float64(data.Stats.Battery[0])
+		default:
+			var ok bool
+			if val, ok = cpuStateAlertValue(name, data.Stats.CpuBreakdown); !ok {
+				continue
+			}
 		}
 
 		triggered := alertData.Triggered
@@ -167,6 +213,7 @@ func (am *AlertManager) HandleSystemAlerts(systemRecord *core.Record, data *syst
 		stat := systemStats[i]
 		// subtract 10 seconds to give a small time buffer
 		systemStatsCreation := stat.Created.Time().Add(-time.Second * 10)
+		stats = SystemAlertStats{}
 		if err := json.Unmarshal(stat.Stats, &stats); err != nil {
 			return err
 		}
@@ -207,6 +254,16 @@ func (am *AlertManager) HandleSystemAlerts(systemRecord *core.Record, data *syst
 						alert.mapSums[key] += float32(fs.DiskUsed / fs.DiskTotal * 100)
 					}
 				}
+				// add zfs pool usage from historical record
+				for key, pool := range stats.ZfsPools {
+					if pool.Total > 0 {
+						zfsKey := zfsDiskAlertKey(key)
+						if _, ok := alert.mapSums[zfsKey]; !ok {
+							alert.mapSums[zfsKey] = 0.0
+						}
+						alert.mapSums[zfsKey] += float32(pool.Used / pool.Total * 100)
+					}
+				}
 			case "Temperature":
 				if alert.mapSums == nil {
 					alert.mapSums = make(map[string]float32, len(stats.Temperatures))
@@ -235,15 +292,25 @@ func (am *AlertManager) HandleSystemAlerts(systemRecord *core.Record, data *syst
 				}
 				alert.val += maxUsage
 			case "Battery":
+				if !hasRepresentativeBattery(stats.Battery, stats.Batteries) {
+					continue
+				}
 				alert.val += float64(stats.Battery[0])
 			default:
-				continue
+				value, ok := cpuStateAlertValue(alert.name, stats.CpuBreakdown)
+				if !ok {
+					continue
+				}
+				alert.val += value
 			}
 			alert.count++
 		}
 	}
 	// sum up vals for each alert
 	for _, alert := range validAlerts {
+		if alert.count == 0 {
+			continue
+		}
 		switch alert.name {
 		case "Disk":
 			maxPct := float32(0)
@@ -251,7 +318,7 @@ func (am *AlertManager) HandleSystemAlerts(systemRecord *core.Record, data *syst
 				sumPct := float32(value)
 				if sumPct > maxPct {
 					maxPct = sumPct
-					alert.descriptor = fmt.Sprintf("Usage of %s", key)
+					alert.descriptor = diskAlertDescriptor(key)
 				}
 			}
 			alert.val = float64(maxPct / float32(alert.count))
@@ -297,10 +364,28 @@ func (am *AlertManager) HandleSystemAlerts(systemRecord *core.Record, data *syst
 	return nil
 }
 
+func zfsDiskAlertKey(poolName string) string {
+	return "zfs:" + poolName
+}
+
+func diskAlertDescriptor(key string) string {
+	if poolName, ok := strings.CutPrefix(key, "zfs:"); ok {
+		return fmt.Sprintf("Usage of ZFS pool %s", poolName)
+	}
+	return fmt.Sprintf("Usage of %s", key)
+}
+
+func hasRepresentativeBattery(legacy [2]uint8, batteries map[string]uint8) bool {
+	return legacy != [2]uint8{} || len(batteries) > 0
+}
+
 func (am *AlertManager) sendSystemAlert(alert SystemAlertData) {
 	// log.Printf("Sending alert %s: val %f | count %d | threshold %f\n", alert.name, alert.val, alert.count, alert.threshold)
 	systemName := alert.systemRecord.GetString("name")
 
+	if state, ok := cpuStateAlerts[alert.name]; ok {
+		alert.name = state.label
+	}
 	// change Disk to Disk usage
 	if alert.name == "Disk" {
 		alert.name += " usage"
@@ -312,7 +397,7 @@ func (am *AlertManager) sendSystemAlert(alert SystemAlertData) {
 
 	// make title alert name lowercase if not CPU or GPU
 	titleAlertName := alert.name
-	if titleAlertName != "CPU" && titleAlertName != "GPU" {
+	if titleAlertName != "CPU" && titleAlertName != "GPU" && !strings.HasPrefix(titleAlertName, "CPU") {
 		titleAlertName = strings.ToLower(titleAlertName)
 	}
 

internal/alerts/alerts_system_test.go

@@ -95,11 +95,10 @@ func waitForSystemAlert(d time.Duration) {
 
 func testOneMinuteSystemAlert[T any](t *testing.T, alertName string, threshold float64, setValue systemAlertValueSetter[T], triggerValue, resolveValue T) {
 	t.Helper()
+	fixture := newSystemAlertTestFixture(t, alertName, 1, threshold)
+	defer fixture.cleanup()
 
 	synctest.Test(t, func(t *testing.T) {
-		fixture := newSystemAlertTestFixture(t, alertName, 1, threshold)
-		defer fixture.cleanup()
-
 		submitValue(fixture, t, triggerValue, setValue)
 		waitForSystemAlert(time.Second)
 
@@ -118,11 +117,10 @@ func testOneMinuteSystemAlert[T any](t *testing.T, alertName string, threshold f
 
 func testMultiMinuteSystemAlert[T any](t *testing.T, alertName string, threshold float64, min int, setValue systemAlertValueSetter[T], baselineValue, triggerValue, resolveValue T) {
 	t.Helper()
+	fixture := newSystemAlertTestFixture(t, alertName, min, threshold)
+	defer fixture.cleanup()
 
 	synctest.Test(t, func(t *testing.T) {
-		fixture := newSystemAlertTestFixture(t, alertName, min, threshold)
-		defer fixture.cleanup()
-
 		submitValue(fixture, t, baselineValue, setValue)
 		waitForSystemAlert(time.Minute + time.Second)
 		fixture.assertTriggered(t, false, "Alert should not be triggered yet")
@@ -148,6 +146,20 @@ func setCPUAlertValue(info *system.Info, stats *system.Stats, value float64) {
 	stats.Cpu = value
 }
 
+func setCPUStateAlertValue(_ *system.Info, stats *system.Stats, value []float64) {
+	stats.CpuBreakdown = value
+}
+
+var cpuStateAlertTests = []struct {
+	name     string
+	trigger  []float64
+	resolve  []float64
+	baseline []float64
+}{
+	{"CPUIOWait", []float64{0, 0, 51, 0, 49}, []float64{0, 0, 48, 0, 52}, []float64{0, 0, 10, 0, 90}},
+	{"CPUSteal", []float64{0, 0, 0, 51, 49}, []float64{0, 0, 0, 48, 52}, []float64{0, 0, 0, 10, 90}},
+}
+
 func setMemoryAlertValue(info *system.Info, stats *system.Stats, value float64) {
 	info.MemPct = value
 	stats.MemPct = value
@@ -193,6 +205,11 @@ func setBatteryAlertValue(info *system.Info, stats *system.Stats, value [2]uint8
 
 func TestSystemAlertsOneMin(t *testing.T) {
 	testOneMinuteSystemAlert(t, "CPU", 50, setCPUAlertValue, 51, 49)
+	for _, test := range cpuStateAlertTests {
+		t.Run(test.name, func(t *testing.T) {
+			testOneMinuteSystemAlert(t, test.name, 50, setCPUStateAlertValue, test.trigger, test.resolve)
+		})
+	}
 	testOneMinuteSystemAlert(t, "Memory", 50, setMemoryAlertValue, 51, 49)
 	testOneMinuteSystemAlert(t, "Disk", 50, setDiskAlertValue, 51, 49)
 	testOneMinuteSystemAlert(t, "Bandwidth", 50, setBandwidthAlertValue, [2]uint64{megabytesToBytes(26), megabytesToBytes(25)}, [2]uint64{megabytesToBytes(25), megabytesToBytes(24)})
@@ -201,11 +218,16 @@ func TestSystemAlertsOneMin(t *testing.T) {
 	testOneMinuteSystemAlert(t, "LoadAvg1", 4, setLoadAvgAlertValue, [3]float64{4.1, 0, 0}, [3]float64{3.9, 0, 0})
 	testOneMinuteSystemAlert(t, "LoadAvg5", 4, setLoadAvgAlertValue, [3]float64{0, 4.1, 0}, [3]float64{0, 3.9, 0})
 	testOneMinuteSystemAlert(t, "LoadAvg15", 4, setLoadAvgAlertValue, [3]float64{0, 0, 4.1}, [3]float64{0, 0, 3.9})
-	testOneMinuteSystemAlert(t, "Battery", 20, setBatteryAlertValue, [2]uint8{19, 0}, [2]uint8{21, 0})
+	testOneMinuteSystemAlert(t, "Battery", 20, setBatteryAlertValue, [2]uint8{0, 1}, [2]uint8{21, 0})
 }
 
 func TestSystemAlertsTwoMin(t *testing.T) {
 	testMultiMinuteSystemAlert(t, "CPU", 50, 2, setCPUAlertValue, 10, 51, 48)
+	for _, test := range cpuStateAlertTests {
+		t.Run(test.name, func(t *testing.T) {
+			testMultiMinuteSystemAlert(t, test.name, 50, 2, setCPUStateAlertValue, test.baseline, test.trigger, test.resolve)
+		})
+	}
 	testMultiMinuteSystemAlert(t, "Memory", 50, 2, setMemoryAlertValue, 10, 51, 48)
 	testMultiMinuteSystemAlert(t, "Disk", 50, 2, setDiskAlertValue, 10, 51, 48)
 	testMultiMinuteSystemAlert(t, "Bandwidth", 50, 2, setBandwidthAlertValue, [2]uint64{megabytesToBytes(10), megabytesToBytes(10)}, [2]uint64{megabytesToBytes(26), megabytesToBytes(25)}, [2]uint64{megabytesToBytes(10), megabytesToBytes(10)})
@@ -216,3 +238,17 @@ func TestSystemAlertsTwoMin(t *testing.T) {
 	testMultiMinuteSystemAlert(t, "LoadAvg15", 4, 2, setLoadAvgAlertValue, [3]float64{0, 0, 2}, [3]float64{0, 0, 4.1}, [3]float64{0, 0, 3.5})
 	testMultiMinuteSystemAlert(t, "Battery", 20, 2, setBatteryAlertValue, [2]uint8{21, 0}, [2]uint8{19, 0}, [2]uint8{25, 1})
 }
+
+func TestCPUStateAlertWithoutBreakdown(t *testing.T) {
+	fixture := newSystemAlertTestFixture(t, "CPUSteal", 1, 1)
+	defer fixture.cleanup()
+
+	synctest.Test(t, func(t *testing.T) {
+		submitValue(fixture, t, []float64(nil), setCPUStateAlertValue)
+		submitValue(fixture, t, []float64{0, 0, 0, 0, 0}, setCPUStateAlertValue)
+		waitForSystemAlert(time.Second)
+
+		fixture.assertTriggered(t, false, "Alert should ignore missing CPU breakdown data")
+		assert.Zero(t, fixture.hub.TestMailer.TotalSend(), "No email should be sent without CPU breakdown data")
+	})
+}

internal/alerts/alerts_systemd.go

@@ -0,0 +1,190 @@
+package alerts
+
+import (
+	"fmt"
+	"strings"
+
+	"github.com/henrygd/beszel/internal/entities/system"
+	"github.com/henrygd/beszel/internal/entities/systemd"
+
+	"github.com/pocketbase/dbx"
+	"github.com/pocketbase/pocketbase/core"
+)
+
+// alertNameSystemdFailed is the alerts.name value for the failed systemd services alert.
+const alertNameSystemdFailed = "SystemdFailed"
+
+// maxListedServices caps how many service names are listed in a notification body.
+const maxListedServices = 10
+
+// HandleSystemdAlerts manages alerts for systemd services in the failed state.
+//
+// This is a binary state alert and fires on the first observation of a failed
+// service rather than using a delay. The agent only refreshes systemd state every
+// 10 minutes, so a shorter delay could never observe new data before expiring, and
+// that poll interval already hides services that fail and restart quickly.
+func (am *AlertManager) HandleSystemdAlerts(systemRecord *core.Record) error {
+	alerts := am.alertsCache.GetAlertsByName(systemRecord.Id, alertNameSystemdFailed)
+	if len(alerts) == 0 {
+		return nil
+	}
+
+	// State is read from the systemd_services snapshot rather than the update payload.
+	// The payload is not a reliable source here: realtime dashboard subscriptions fetch
+	// from the agent with a shorter cache time, and the agent omits systemd services from
+	// those responses, overwriting the cached payload roughly once a second while a system
+	// is being viewed. The snapshot table is only written by the full update cycle.
+	total, failed, err := am.queryServiceStates(systemRecord.Id)
+	if err != nil {
+		return err
+	}
+	if total == 0 {
+		// No rows normally means no systemd data for this system (agent without
+		// systemd, or not yet reported), which must not be treated as a recovery.
+		// Read info only in this ambiguous case. The record being saved is used
+		// instead of data because dashboard polling can replace the system's
+		// in-memory payload concurrently.
+		var currentInfo system.Info
+		if err := systemRecord.UnmarshalJSONField("info", &currentInfo); err != nil ||
+			len(currentInfo.Services) == 0 || currentInfo.Services[0] != 0 {
+			return nil
+		}
+	}
+
+	systemName := systemRecord.GetString("name")
+
+	for _, alertData := range alerts {
+		triggered := len(failed) > 0
+		// Only notify on a change of state, so a service that stays failed across
+		// cycles doesn't re-notify every update.
+		if triggered == alertData.Triggered {
+			continue
+		}
+		if err := am.sendSystemdAlert(triggered, systemName, alertData, failed); err != nil {
+			am.hub.Logger().Error("Failed to send alert", "err", err)
+		}
+	}
+	return nil
+}
+
+// queryServiceStates returns the number of services reported in the most recent update
+// for a system, and the names of those in the failed state.
+//
+// Rows are restricted to the latest update because systemd_services is upserted, never
+// pruned on change: a service that no longer exists on the host stops being reported and
+// its row keeps its last known state until the retention sweep removes it. Every row
+// written in one cycle shares a single updated timestamp, so the newest timestamp
+// identifies exactly the services the agent last reported.
+func (am *AlertManager) queryServiceStates(systemID string) (total int, failed []string, err error) {
+	var rows []struct {
+		Name  string               `db:"name"`
+		State systemd.ServiceState `db:"state"`
+	}
+	err = am.hub.DB().
+		Select("name", "state").
+		From("systemd_services").
+		Where(dbx.NewExp(
+			"system={:system} AND updated=(SELECT MAX(updated) FROM systemd_services WHERE system={:system})",
+			dbx.Params{"system": systemID},
+		)).
+		OrderBy("name").
+		All(&rows)
+	if err != nil {
+		return 0, nil, err
+	}
+	for _, row := range rows {
+		if row.State == systemd.StatusFailed {
+			failed = append(failed, row.Name)
+		}
+	}
+	return len(rows), failed, nil
+}
+
+// sendSystemdAlert sends a failed or recovered systemd services alert to the alert's user.
+func (am *AlertManager) sendSystemdAlert(triggered bool, systemName string, alertData CachedAlertData, failed []string) error {
+	// Update trigger state for alert record before sending alert
+	if err := am.setAlertTriggered(alertData, triggered); err != nil {
+		return err
+	}
+
+	var title, message string
+	if triggered {
+		title = fmt.Sprintf("Failed services on %s %v", systemName, "\U0001F534") // Red alert emoji
+		message = fmt.Sprintf("%s on %s: %s", pluralizeServices(len(failed)), systemName, formatServiceList(failed))
+	} else {
+		title = fmt.Sprintf("Services recovered on %s %v", systemName, "✅") // Green checkmark emoji
+		message = fmt.Sprintf("No services are in the failed state on %s.", systemName)
+	}
+
+	systemID := alertData.SystemID
+
+	return am.SendAlert(AlertMessageData{
+		UserID:   alertData.UserID,
+		SystemID: systemID,
+		Title:    title,
+		Message:  message,
+		Link:     am.hub.MakeLink("system", systemID),
+		LinkText: "View " + systemName,
+	})
+}
+
+// pluralizeServices returns a count label like "1 failed service" or "3 failed services".
+func pluralizeServices(count int) string {
+	if count == 1 {
+		return "1 failed service"
+	}
+	return fmt.Sprintf("%d failed services", count)
+}
+
+// formatServiceList joins service names, truncating long lists.
+func formatServiceList(names []string) string {
+	if len(names) <= maxListedServices {
+		return strings.Join(names, ", ")
+	}
+	remaining := len(names) - maxListedServices
+	return fmt.Sprintf("%s and %d more", strings.Join(names[:maxListedServices], ", "), remaining)
+}
+
+// resolveSystemdAlerts resolves triggered systemd alerts for systems that no longer
+// have any failed services. This clears stale state left by a hub restart.
+func resolveSystemdAlerts(app core.App) error {
+	db := app.DB()
+	var alertIds []string
+	err := db.NewQuery(`
+		SELECT a.id
+		FROM alerts a
+		JOIN systems sys ON sys.id = a.system
+		WHERE a.name = {:name}
+		AND a.triggered = true
+		AND (
+			EXISTS (
+				SELECT 1 FROM systemd_services cur
+				WHERE cur.system = a.system
+				AND cur.updated = (SELECT MAX(updated) FROM systemd_services WHERE system = a.system)
+			)
+			OR json_extract(sys.info, '$.sv[0]') = 0
+		)
+		AND NOT EXISTS (
+			SELECT 1 FROM systemd_services s
+			WHERE s.system = a.system AND s.state = {:state}
+			AND s.updated = (SELECT MAX(updated) FROM systemd_services WHERE system = a.system)
+		)
+	`).Bind(dbx.Params{
+		"name":  alertNameSystemdFailed,
+		"state": systemd.StatusFailed,
+	}).Column(&alertIds)
+	if err != nil {
+		return err
+	}
+	for _, alertId := range alertIds {
+		alert, err := app.FindRecordById("alerts", alertId)
+		if err != nil {
+			return err
+		}
+		alert.Set("triggered", false)
+		if err := app.Save(alert); err != nil {
+			return err
+		}
+	}
+	return nil
+}

internal/alerts/alerts_systemd_test.go

@@ -0,0 +1,383 @@
+//go:build testing
+
+package alerts_test
+
+import (
+	"testing"
+	"time"
+
+	"github.com/henrygd/beszel/internal/alerts"
+	systemEntity "github.com/henrygd/beszel/internal/entities/system"
+	"github.com/henrygd/beszel/internal/entities/systemd"
+	beszelTests "github.com/henrygd/beszel/internal/tests"
+	"github.com/pocketbase/dbx"
+	"github.com/pocketbase/pocketbase/core"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// setSystemdServiceState upserts a systemd_services row mirroring the raw SQL write
+// path used by the hub (createSystemdStatsRecords), which bypasses record hooks.
+func setSystemdServiceState(t *testing.T, hub core.App, systemID, name string, state systemd.ServiceState, updated int64) {
+	t.Helper()
+
+	_, err := hub.DB().NewQuery(
+		"INSERT INTO systemd_services (id, system, name, state, sub, cpu, cpuPeak, memory, memPeak, updated) " +
+			"VALUES ({:id}, {:system}, {:name}, {:state}, 0, 0, 0, 0, 0, {:updated}) " +
+			"ON CONFLICT(id) DO UPDATE SET state = excluded.state, updated = excluded.updated",
+	).Bind(dbx.Params{
+		"id":      systemID + "-" + name,
+		"system":  systemID,
+		"name":    name,
+		"state":   state,
+		"updated": updated,
+	}).Execute()
+	require.NoError(t, err)
+}
+
+// seedServices writes a set of services into the systemd_services snapshot, which is the
+// source HandleSystemdAlerts reads from. All rows share one updated timestamp, matching
+// how the hub writes a batch in createSystemdStatsRecords.
+func seedServices(t *testing.T, hub core.App, systemID string, states ...systemd.ServiceState) {
+	t.Helper()
+	seedServicesAt(t, hub, systemID, time.Now().UTC().UnixMilli(), states...)
+}
+
+// seedServicesAt writes services with an explicit batch timestamp.
+func seedServicesAt(t *testing.T, hub core.App, systemID string, updated int64, states ...systemd.ServiceState) {
+	t.Helper()
+	for i, state := range states {
+		setSystemdServiceState(t, hub, systemID, serviceName(i), state, updated)
+	}
+}
+
+func serviceName(i int) string {
+	return string(rune('a'+i)) + ".service"
+}
+
+// systemdTestSetup creates a user with an email, a system, and a SystemdFailed alert.
+func systemdTestSetup(t *testing.T, triggered bool) (*beszelTests.TestHub, *core.Record, *core.Record) {
+	t.Helper()
+
+	hub, user := beszelTests.GetHubWithUser(t)
+
+	userSettings, err := hub.FindFirstRecordByFilter("user_settings", "user={:user}", map[string]any{"user": user.Id})
+	require.NoError(t, err)
+	userSettings.Set("settings", `{"emails":["test@example.com"],"webhooks":[]}`)
+	require.NoError(t, hub.Save(userSettings))
+
+	// "paused" avoids spawning a background updater goroutine that would outlive
+	// the test hub; these tests drive HandleSystemdAlerts directly.
+	systems, err := beszelTests.CreateSystems(hub, 1, user.Id, "paused")
+	require.NoError(t, err)
+	system := systems[0]
+
+	alert, err := beszelTests.CreateRecord(hub, "alerts", map[string]any{
+		"name":      "SystemdFailed",
+		"system":    system.Id,
+		"user":      user.Id,
+		"triggered": triggered,
+	})
+	require.NoError(t, err)
+
+	return hub, system, alert
+}
+
+func TestSystemdAlertFiresImmediately(t *testing.T) {
+	hub, system, alert := systemdTestSetup(t, false)
+	defer hub.Cleanup()
+
+	initialEmailCount := hub.TestMailer.TotalSend()
+	am := alerts.NewTestAlertManagerWithoutWorker(hub)
+
+	seedServices(t, hub, system.Id, systemd.StatusFailed, systemd.StatusActive)
+	require.NoError(t, am.HandleSystemdAlerts(system))
+
+	assert.Equal(t, initialEmailCount+1, hub.TestMailer.TotalSend(), "failed service should notify on first observation")
+
+	messages := hub.TestMailer.Messages()
+	require.NotEmpty(t, messages)
+	last := messages[len(messages)-1]
+	assert.Contains(t, last.Subject, "Failed services")
+	assert.Contains(t, last.Text, "a.service", "notification should name the failed service")
+
+	alertRecord, err := hub.FindRecordById("alerts", alert.Id)
+	require.NoError(t, err)
+	assert.True(t, alertRecord.GetBool("triggered"), "alert should be marked triggered")
+
+	// history record should be created via the alerts update hook
+	historyCount, err := hub.CountRecords("alerts_history", dbx.HashExp{"resolved": ""})
+	require.NoError(t, err)
+	assert.EqualValues(t, 1, historyCount, "should have one unresolved alert history record")
+}
+
+func TestSystemdAlertFullCycle(t *testing.T) {
+	hub, system, alert := systemdTestSetup(t, false)
+	defer hub.Cleanup()
+
+	initialEmailCount := hub.TestMailer.TotalSend()
+	am := alerts.NewTestAlertManagerWithoutWorker(hub)
+
+	// Fail, then recover.
+	seedServices(t, hub, system.Id, systemd.StatusFailed)
+	require.NoError(t, am.HandleSystemdAlerts(system))
+	seedServices(t, hub, system.Id, systemd.StatusActive)
+	require.NoError(t, am.HandleSystemdAlerts(system))
+
+	assert.Equal(t, initialEmailCount+2, hub.TestMailer.TotalSend(), "should send a failure and a recovery notification")
+
+	messages := hub.TestMailer.Messages()
+	require.Len(t, messages, 2)
+	assert.Contains(t, messages[0].Subject, "Failed services")
+	assert.Contains(t, messages[1].Subject, "Services recovered")
+
+	alertRecord, err := hub.FindRecordById("alerts", alert.Id)
+	require.NoError(t, err)
+	assert.False(t, alertRecord.GetBool("triggered"), "alert should be cleared after recovery")
+
+	// history record should be resolved
+	historyCount, err := hub.CountRecords("alerts_history", dbx.HashExp{"resolved": ""})
+	require.NoError(t, err)
+	assert.Zero(t, historyCount, "alert history record should be resolved")
+}
+
+func TestSystemdAlertSendsRecoveryWhenTriggered(t *testing.T) {
+	hub, system, alert := systemdTestSetup(t, true)
+	defer hub.Cleanup()
+
+	initialEmailCount := hub.TestMailer.TotalSend()
+	am := alerts.NewTestAlertManagerWithoutWorker(hub)
+
+	seedServices(t, hub, system.Id, systemd.StatusActive, systemd.StatusInactive)
+	require.NoError(t, am.HandleSystemdAlerts(system))
+
+	assert.Equal(t, initialEmailCount+1, hub.TestMailer.TotalSend(), "recovery notification should be sent")
+	messages := hub.TestMailer.Messages()
+	require.NotEmpty(t, messages)
+	assert.Contains(t, messages[len(messages)-1].Subject, "Services recovered")
+
+	alertRecord, err := hub.FindRecordById("alerts", alert.Id)
+	require.NoError(t, err)
+	assert.False(t, alertRecord.GetBool("triggered"), "alert should be cleared after recovery")
+}
+
+func TestSystemdAlertDoesNotResendWhileTriggered(t *testing.T) {
+	hub, system, _ := systemdTestSetup(t, true)
+	defer hub.Cleanup()
+
+	initialEmailCount := hub.TestMailer.TotalSend()
+	am := alerts.NewTestAlertManagerWithoutWorker(hub)
+
+	// Still failing across several cycles — should not re-notify.
+	for range 3 {
+		seedServices(t, hub, system.Id, systemd.StatusFailed)
+		require.NoError(t, am.HandleSystemdAlerts(system))
+	}
+
+	assert.Equal(t, initialEmailCount, hub.TestMailer.TotalSend(), "should not re-notify while still triggered")
+}
+
+func TestSystemdAlertRepeatedFailureNotifiesOnce(t *testing.T) {
+	hub, system, _ := systemdTestSetup(t, false)
+	defer hub.Cleanup()
+
+	initialEmailCount := hub.TestMailer.TotalSend()
+	am := alerts.NewTestAlertManagerWithoutWorker(hub)
+
+	for range 3 {
+		seedServices(t, hub, system.Id, systemd.StatusFailed)
+		require.NoError(t, am.HandleSystemdAlerts(system))
+	}
+
+	assert.Equal(t, initialEmailCount+1, hub.TestMailer.TotalSend(), "repeated failures should only notify once")
+}
+
+// A service that no longer exists on the host stops being reported, but its row stays
+// in systemd_services with its last known state until the retention sweep. That stale
+// row must not keep the alert triggered.
+func TestSystemdAlertIgnoresServicesNoLongerReported(t *testing.T) {
+	hub, system, alert := systemdTestSetup(t, true)
+	defer hub.Cleanup()
+
+	initialEmailCount := hub.TestMailer.TotalSend()
+	am := alerts.NewTestAlertManagerWithoutWorker(hub)
+
+	now := time.Now().UTC().UnixMilli()
+	// Older batch still holding a failed service that has since been removed.
+	setSystemdServiceState(t, hub, system.Id, "gone.service", systemd.StatusFailed, now-60_000)
+	// Current batch reports only healthy services.
+	seedServicesAt(t, hub, system.Id, now, systemd.StatusActive, systemd.StatusActive)
+
+	require.NoError(t, am.HandleSystemdAlerts(system))
+
+	assert.Equal(t, initialEmailCount+1, hub.TestMailer.TotalSend(), "stale failed row should not block recovery")
+	alertRecord, err := hub.FindRecordById("alerts", alert.Id)
+	require.NoError(t, err)
+	assert.False(t, alertRecord.GetBool("triggered"), "alert should resolve once the service stops being reported")
+}
+
+func TestResolveSystemdAlertsIgnoresStaleFailedRows(t *testing.T) {
+	hub, system, alert := systemdTestSetup(t, true)
+	defer hub.Cleanup()
+
+	now := time.Now().UTC().UnixMilli()
+	setSystemdServiceState(t, hub, system.Id, "gone.service", systemd.StatusFailed, now-60_000)
+	seedServicesAt(t, hub, system.Id, now, systemd.StatusActive)
+
+	require.NoError(t, alerts.ResolveSystemdAlerts(hub))
+
+	alertRecord, err := hub.FindRecordById("alerts", alert.Id)
+	require.NoError(t, err)
+	assert.False(t, alertRecord.GetBool("triggered"), "stale failed row should not keep the alert triggered")
+}
+
+func TestSystemdAlertNoSystemdDataIsIgnored(t *testing.T) {
+	hub, system, alert := systemdTestSetup(t, true)
+	defer hub.Cleanup()
+
+	initialEmailCount := hub.TestMailer.TotalSend()
+	am := alerts.NewTestAlertManagerWithoutWorker(hub)
+
+	// A system with no systemd_services rows (agent without systemd, or nothing
+	// reported yet) must not be treated as a recovery.
+	require.NoError(t, am.HandleSystemdAlerts(system))
+	require.NoError(t, am.HandleSystemdAlerts(system))
+
+	assert.Equal(t, initialEmailCount, hub.TestMailer.TotalSend(), "missing systemd data should not send a recovery")
+	alertRecord, err := hub.FindRecordById("alerts", alert.Id)
+	require.NoError(t, err)
+	assert.True(t, alertRecord.GetBool("triggered"), "triggered state should be preserved when data is absent")
+}
+
+func TestSystemdAlertFreshEmptySnapshotResolves(t *testing.T) {
+	hub, system, alert := systemdTestSetup(t, true)
+	defer hub.Cleanup()
+
+	initialEmailCount := hub.TestMailer.TotalSend()
+	am := alerts.NewTestAlertManagerWithoutWorker(hub)
+
+	// An explicit zero service count on the saved system record distinguishes a
+	// confirmed empty snapshot from an agent response that omitted systemd data.
+	system.Set("info", systemEntity.Info{Services: []uint16{0, 0}})
+	require.NoError(t, am.HandleSystemAlerts(system, nil))
+
+	assert.Equal(t, initialEmailCount+1, hub.TestMailer.TotalSend(), "fresh empty snapshot should send a recovery")
+	alertRecord, err := hub.FindRecordById("alerts", alert.Id)
+	require.NoError(t, err)
+	assert.False(t, alertRecord.GetBool("triggered"), "fresh empty snapshot should resolve the alert")
+}
+
+func TestSystemdAlertNoAlertRecord(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	systems, err := beszelTests.CreateSystems(hub, 1, user.Id, "paused")
+	require.NoError(t, err)
+	system := systems[0]
+
+	initialEmailCount := hub.TestMailer.TotalSend()
+	am := alerts.NewTestAlertManagerWithoutWorker(hub)
+
+	seedServices(t, hub, system.Id, systemd.StatusFailed)
+	require.NoError(t, am.HandleSystemdAlerts(system))
+	assert.Equal(t, initialEmailCount, hub.TestMailer.TotalSend(), "no email when no alert record exists")
+}
+
+func TestResolveSystemdAlertsClearsStaleTriggered(t *testing.T) {
+	hub, system, alert := systemdTestSetup(t, true)
+	defer hub.Cleanup()
+
+	// No failed services in the snapshot, but the alert is still marked triggered
+	// (e.g. the hub restarted while the alert was active).
+	setSystemdServiceState(t, hub, system.Id, "a.service", systemd.StatusActive, time.Now().UTC().UnixMilli())
+
+	require.NoError(t, alerts.ResolveSystemdAlerts(hub))
+
+	alertRecord, err := hub.FindRecordById("alerts", alert.Id)
+	require.NoError(t, err)
+	assert.False(t, alertRecord.GetBool("triggered"), "stale triggered flag should be cleared")
+}
+
+func TestResolveSystemdAlertsKeepsTriggeredWithoutSystemdData(t *testing.T) {
+	hub, _, alert := systemdTestSetup(t, true)
+	defer hub.Cleanup()
+
+	// Missing rows do not prove recovery. This can happen when a system is offline
+	// and its last service snapshot has been removed by retention.
+	require.NoError(t, alerts.ResolveSystemdAlerts(hub))
+
+	alertRecord, err := hub.FindRecordById("alerts", alert.Id)
+	require.NoError(t, err)
+	assert.True(t, alertRecord.GetBool("triggered"), "missing systemd data should preserve triggered state")
+}
+
+func TestResolveSystemdAlertsClearsConfirmedEmptySnapshot(t *testing.T) {
+	hub, system, alert := systemdTestSetup(t, true)
+	defer hub.Cleanup()
+
+	// Update the persisted snapshot directly so record hooks don't alter alert state
+	// before the startup resolver is exercised.
+	_, err := hub.DB().NewQuery(
+		"UPDATE systems SET info = {:info} WHERE id = {:id}",
+	).Bind(dbx.Params{"info": `{"sv":[0,0]}`, "id": system.Id}).Execute()
+	require.NoError(t, err)
+	require.NoError(t, alerts.ResolveSystemdAlerts(hub))
+
+	alertRecord, err := hub.FindRecordById("alerts", alert.Id)
+	require.NoError(t, err)
+	assert.False(t, alertRecord.GetBool("triggered"), "confirmed empty snapshot should clear triggered state")
+}
+
+func TestResolveSystemdAlertsKeepsStillFailing(t *testing.T) {
+	hub, system, alert := systemdTestSetup(t, true)
+	defer hub.Cleanup()
+
+	setSystemdServiceState(t, hub, system.Id, "a.service", systemd.StatusFailed, time.Now().UTC().UnixMilli())
+
+	require.NoError(t, alerts.ResolveSystemdAlerts(hub))
+
+	alertRecord, err := hub.FindRecordById("alerts", alert.Id)
+	require.NoError(t, err)
+	assert.True(t, alertRecord.GetBool("triggered"), "alert should stay triggered while a service is still failed")
+}
+
+func TestSystemdAlertMultipleUsersRespectOwnAlerts(t *testing.T) {
+	hub, user1 := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	setStatusAlertEmail(t, hub, user1.Id, "user1@example.com")
+
+	user2, err := beszelTests.CreateUser(hub, "user2@example.com", "password")
+	require.NoError(t, err)
+	_, err = beszelTests.CreateRecord(hub, "user_settings", map[string]any{
+		"user": user2.Id,
+		"settings": map[string]any{
+			"emails":   []string{"user2@example.com"},
+			"webhooks": []string{},
+		},
+	})
+	require.NoError(t, err)
+
+	system, err := beszelTests.CreateRecord(hub, "systems", map[string]any{
+		"name":  "shared-system",
+		"users": []string{user1.Id, user2.Id},
+		"host":  "127.0.0.1",
+	})
+	require.NoError(t, err)
+
+	for _, user := range []*core.Record{user1, user2} {
+		_, err = beszelTests.CreateRecord(hub, "alerts", map[string]any{
+			"name":   "SystemdFailed",
+			"system": system.Id,
+			"user":   user.Id,
+		})
+		require.NoError(t, err)
+	}
+
+	am := alerts.NewTestAlertManagerWithoutWorker(hub)
+	seedServices(t, hub, system.Id, systemd.StatusFailed)
+	require.NoError(t, am.HandleSystemdAlerts(system))
+
+	messages := hub.TestMailer.Messages()
+	require.Len(t, messages, 2, "each user should receive their own alert")
+}

internal/alerts/alerts_test.go

@@ -3,11 +3,6 @@
 package alerts_test
 
 import (
-	"bytes"
-	"encoding/json"
-	"io"
-	"net/http"
-	"strings"
 	"testing"
 	"testing/synctest"
 	"time"
@@ -16,362 +11,13 @@ import (
 
 	"github.com/henrygd/beszel/internal/alerts"
 	"github.com/pocketbase/dbx"
-	"github.com/pocketbase/pocketbase/core"
-	pbTests "github.com/pocketbase/pocketbase/tests"
 	"github.com/stretchr/testify/assert"
 )
 
-// marshal to json and return an io.Reader (for use in ApiScenario.Body)
-func jsonReader(v any) io.Reader {
-	data, err := json.Marshal(v)
-	if err != nil {
-		panic(err)
-	}
-	return bytes.NewReader(data)
-}
-
-func TestUserAlertsApi(t *testing.T) {
-	hub, _ := beszelTests.NewTestHub(t.TempDir())
-	defer hub.Cleanup()
-
-	hub.StartHub()
-
-	user1, _ := beszelTests.CreateUser(hub, "alertstest@example.com", "password")
-	user1Token, _ := user1.NewAuthToken()
-
-	user2, _ := beszelTests.CreateUser(hub, "alertstest2@example.com", "password")
-	user2Token, _ := user2.NewAuthToken()
-
-	system1, _ := beszelTests.CreateRecord(hub, "systems", map[string]any{
-		"name":  "system1",
-		"users": []string{user1.Id},
-		"host":  "127.0.0.1",
-	})
-
-	system2, _ := beszelTests.CreateRecord(hub, "systems", map[string]any{
-		"name":  "system2",
-		"users": []string{user1.Id, user2.Id},
-		"host":  "127.0.0.2",
-	})
-
-	userRecords, _ := hub.CountRecords("users")
-	assert.EqualValues(t, 2, userRecords, "all users should be created")
-
-	systemRecords, _ := hub.CountRecords("systems")
-	assert.EqualValues(t, 2, systemRecords, "all systems should be created")
-
-	testAppFactory := func(t testing.TB) *pbTests.TestApp {
-		return hub.TestApp
-	}
-
-	scenarios := []beszelTests.ApiScenario{
-		// {
-		// 	Name:            "GET not implemented - returns index",
-		// 	Method:          http.MethodGet,
-		// 	URL:             "/api/beszel/user-alerts",
-		// 	ExpectedStatus:  200,
-		// 	ExpectedContent: []string{"<html ", "globalThis.BESZEL"},
-		// 	TestAppFactory:  testAppFactory,
-		// },
-		{
-			Name:            "POST no auth",
-			Method:          http.MethodPost,
-			URL:             "/api/beszel/user-alerts",
-			ExpectedStatus:  401,
-			ExpectedContent: []string{"requires valid"},
-			TestAppFactory:  testAppFactory,
-		},
-		{
-			Name:   "POST no body",
-			Method: http.MethodPost,
-			URL:    "/api/beszel/user-alerts",
-			Headers: map[string]string{
-				"Authorization": user1Token,
-			},
-			ExpectedStatus:  400,
-			ExpectedContent: []string{"Bad data"},
-			TestAppFactory:  testAppFactory,
-		},
-		{
-			Name:   "POST bad data",
-			Method: http.MethodPost,
-			URL:    "/api/beszel/user-alerts",
-			Headers: map[string]string{
-				"Authorization": user1Token,
-			},
-			ExpectedStatus:  400,
-			ExpectedContent: []string{"Bad data"},
-			TestAppFactory:  testAppFactory,
-			Body: jsonReader(map[string]any{
-				"invalidField": "this should cause validation error",
-				"threshold":    "not a number",
-			}),
-		},
-		{
-			Name:   "POST malformed JSON",
-			Method: http.MethodPost,
-			URL:    "/api/beszel/user-alerts",
-			Headers: map[string]string{
-				"Authorization": user1Token,
-			},
-			ExpectedStatus:  400,
-			ExpectedContent: []string{"Bad data"},
-			TestAppFactory:  testAppFactory,
-			Body:            strings.NewReader(`{"alertType": "cpu", "threshold": 80, "enabled": true,}`),
-		},
-		{
-			Name:   "POST valid alert data multiple systems",
-			Method: http.MethodPost,
-			URL:    "/api/beszel/user-alerts",
-			Headers: map[string]string{
-				"Authorization": user1Token,
-			},
-			ExpectedStatus:  200,
-			ExpectedContent: []string{"\"success\":true"},
-			TestAppFactory:  testAppFactory,
-			Body: jsonReader(map[string]any{
-				"name":      "CPU",
-				"value":     69,
-				"min":       9,
-				"systems":   []string{system1.Id, system2.Id},
-				"overwrite": false,
-			}),
-			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
-				// check total alerts
-				alerts, _ := app.CountRecords("alerts")
-				assert.EqualValues(t, 2, alerts, "should have 2 alerts")
-				// check alert has correct values
-				matchingAlerts, _ := app.CountRecords("alerts", dbx.HashExp{"name": "CPU", "user": user1.Id, "system": system1.Id, "value": 69, "min": 9})
-				assert.EqualValues(t, 1, matchingAlerts, "should have 1 alert")
-			},
-		},
-		{
-			Name:   "POST valid alert data single system",
-			Method: http.MethodPost,
-			URL:    "/api/beszel/user-alerts",
-			Headers: map[string]string{
-				"Authorization": user1Token,
-			},
-			ExpectedStatus:  200,
-			ExpectedContent: []string{"\"success\":true"},
-			TestAppFactory:  testAppFactory,
-			Body: jsonReader(map[string]any{
-				"name":    "Memory",
-				"systems": []string{system1.Id},
-				"value":   90,
-				"min":     10,
-			}),
-			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
-				user1Alerts, _ := app.CountRecords("alerts", dbx.HashExp{"user": user1.Id})
-				assert.EqualValues(t, 3, user1Alerts, "should have 3 alerts")
-			},
-		},
-		{
-			Name:   "Overwrite: false, should not overwrite existing alert",
-			Method: http.MethodPost,
-			URL:    "/api/beszel/user-alerts",
-			Headers: map[string]string{
-				"Authorization": user1Token,
-			},
-			ExpectedStatus:  200,
-			ExpectedContent: []string{"\"success\":true"},
-			TestAppFactory:  testAppFactory,
-			Body: jsonReader(map[string]any{
-				"name":      "CPU",
-				"value":     45,
-				"min":       5,
-				"systems":   []string{system1.Id},
-				"overwrite": false,
-			}),
-			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
-				beszelTests.ClearCollection(t, app, "alerts")
-				beszelTests.CreateRecord(app, "alerts", map[string]any{
-					"name":   "CPU",
-					"system": system1.Id,
-					"user":   user1.Id,
-					"value":  80,
-					"min":    10,
-				})
-			},
-			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
-				alerts, _ := app.CountRecords("alerts")
-				assert.EqualValues(t, 1, alerts, "should have 1 alert")
-				alert, _ := app.FindFirstRecordByFilter("alerts", "name = 'CPU' && user = {:user}", dbx.Params{"user": user1.Id})
-				assert.EqualValues(t, 80, alert.Get("value"), "should have 80 as value")
-			},
-		},
-		{
-			Name:   "Overwrite: true, should overwrite existing alert",
-			Method: http.MethodPost,
-			URL:    "/api/beszel/user-alerts",
-			Headers: map[string]string{
-				"Authorization": user2Token,
-			},
-			ExpectedStatus:  200,
-			ExpectedContent: []string{"\"success\":true"},
-			TestAppFactory:  testAppFactory,
-			Body: jsonReader(map[string]any{
-				"name":      "CPU",
-				"value":     45,
-				"min":       5,
-				"systems":   []string{system2.Id},
-				"overwrite": true,
-			}),
-			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
-				beszelTests.ClearCollection(t, app, "alerts")
-				beszelTests.CreateRecord(app, "alerts", map[string]any{
-					"name":   "CPU",
-					"system": system2.Id,
-					"user":   user2.Id,
-					"value":  80,
-					"min":    10,
-				})
-			},
-			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
-				alerts, _ := app.CountRecords("alerts")
-				assert.EqualValues(t, 1, alerts, "should have 1 alert")
-				alert, _ := app.FindFirstRecordByFilter("alerts", "name = 'CPU' && user = {:user}", dbx.Params{"user": user2.Id})
-				assert.EqualValues(t, 45, alert.Get("value"), "should have 45 as value")
-			},
-		},
-		{
-			Name:            "DELETE no auth",
-			Method:          http.MethodDelete,
-			URL:             "/api/beszel/user-alerts",
-			ExpectedStatus:  401,
-			ExpectedContent: []string{"requires valid"},
-			TestAppFactory:  testAppFactory,
-			Body: jsonReader(map[string]any{
-				"name":    "CPU",
-				"systems": []string{system1.Id},
-			}),
-			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
-				beszelTests.ClearCollection(t, app, "alerts")
-				beszelTests.CreateRecord(app, "alerts", map[string]any{
-					"name":   "CPU",
-					"system": system1.Id,
-					"user":   user1.Id,
-					"value":  80,
-					"min":    10,
-				})
-			},
-			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
-				alerts, _ := app.CountRecords("alerts")
-				assert.EqualValues(t, 1, alerts, "should have 1 alert")
-			},
-		},
-		{
-			Name:   "DELETE alert",
-			Method: http.MethodDelete,
-			URL:    "/api/beszel/user-alerts",
-			Headers: map[string]string{
-				"Authorization": user1Token,
-			},
-			ExpectedStatus:  200,
-			ExpectedContent: []string{"\"count\":1", "\"success\":true"},
-			TestAppFactory:  testAppFactory,
-			Body: jsonReader(map[string]any{
-				"name":    "CPU",
-				"systems": []string{system1.Id},
-			}),
-			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
-				beszelTests.ClearCollection(t, app, "alerts")
-				beszelTests.CreateRecord(app, "alerts", map[string]any{
-					"name":   "CPU",
-					"system": system1.Id,
-					"user":   user1.Id,
-					"value":  80,
-					"min":    10,
-				})
-			},
-			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
-				alerts, _ := app.CountRecords("alerts")
-				assert.Zero(t, alerts, "should have 0 alerts")
-			},
-		},
-		{
-			Name:   "DELETE alert multiple systems",
-			Method: http.MethodDelete,
-			URL:    "/api/beszel/user-alerts",
-			Headers: map[string]string{
-				"Authorization": user1Token,
-			},
-			ExpectedStatus:  200,
-			ExpectedContent: []string{"\"count\":2", "\"success\":true"},
-			TestAppFactory:  testAppFactory,
-			Body: jsonReader(map[string]any{
-				"name":    "Memory",
-				"systems": []string{system1.Id, system2.Id},
-			}),
-			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
-				beszelTests.ClearCollection(t, app, "alerts")
-				for _, systemId := range []string{system1.Id, system2.Id} {
-					_, err := beszelTests.CreateRecord(app, "alerts", map[string]any{
-						"name":   "Memory",
-						"system": systemId,
-						"user":   user1.Id,
-						"value":  90,
-						"min":    10,
-					})
-					assert.NoError(t, err, "should create alert")
-				}
-				alerts, _ := app.CountRecords("alerts")
-				assert.EqualValues(t, 2, alerts, "should have 2 alerts")
-			},
-			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
-				alerts, _ := app.CountRecords("alerts")
-				assert.Zero(t, alerts, "should have 0 alerts")
-			},
-		},
-		{
-			Name:   "User 2 should not be able to delete alert of user 1",
-			Method: http.MethodDelete,
-			URL:    "/api/beszel/user-alerts",
-			Headers: map[string]string{
-				"Authorization": user2Token,
-			},
-			ExpectedStatus:  200,
-			ExpectedContent: []string{"\"count\":1", "\"success\":true"},
-			TestAppFactory:  testAppFactory,
-			Body: jsonReader(map[string]any{
-				"name":    "CPU",
-				"systems": []string{system2.Id},
-			}),
-			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
-				beszelTests.ClearCollection(t, app, "alerts")
-				for _, user := range []string{user1.Id, user2.Id} {
-					beszelTests.CreateRecord(app, "alerts", map[string]any{
-						"name":   "CPU",
-						"system": system2.Id,
-						"user":   user,
-						"value":  80,
-						"min":    10,
-					})
-				}
-				alerts, _ := app.CountRecords("alerts")
-				assert.EqualValues(t, 2, alerts, "should have 2 alerts")
-				user1AlertCount, _ := app.CountRecords("alerts", dbx.HashExp{"user": user1.Id})
-				assert.EqualValues(t, 1, user1AlertCount, "should have 1 alert")
-				user2AlertCount, _ := app.CountRecords("alerts", dbx.HashExp{"user": user2.Id})
-				assert.EqualValues(t, 1, user2AlertCount, "should have 1 alert")
-			},
-			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
-				user1AlertCount, _ := app.CountRecords("alerts", dbx.HashExp{"user": user1.Id})
-				assert.EqualValues(t, 1, user1AlertCount, "should have 1 alert")
-				user2AlertCount, _ := app.CountRecords("alerts", dbx.HashExp{"user": user2.Id})
-				assert.Zero(t, user2AlertCount, "should have 0 alerts")
-			},
-		},
-	}
-
-	for _, scenario := range scenarios {
-		scenario.Test(t)
-	}
-}
-
 func TestAlertsHistory(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+
 	synctest.Test(t, func(t *testing.T) {
-		hub, user := beszelTests.GetHubWithUser(t)
 		defer hub.Cleanup()
 
 		// Create systems and alerts

internal/alerts/alerts_test_helpers.go

@@ -88,6 +88,10 @@ func ResolveStatusAlerts(app core.App) error {
 	return resolveStatusAlerts(app)
 }
 
+func ResolveSystemdAlerts(app core.App) error {
+	return resolveSystemdAlerts(app)
+}
+
 func (am *AlertManager) RestorePendingStatusAlerts() error {
 	return am.restorePendingStatusAlerts()
 }
@@ -95,3 +99,12 @@ func (am *AlertManager) RestorePendingStatusAlerts() error {
 func (am *AlertManager) SetAlertTriggered(alert CachedAlertData, triggered bool) error {
 	return am.setAlertTriggered(alert, triggered)
 }
+
+func IsInternalURL(rawURL string) (bool, error) {
+	return isInternalURL(rawURL)
+}
+
+// BuildContainerLogExcerpt exposes buildContainerLogExcerpt for testing.
+func BuildContainerLogExcerpt(raw string) string {
+	return buildContainerLogExcerpt(raw)
+}

internal/alerts/alerts_zfs.go

@@ -0,0 +1,142 @@
+package alerts
+
+import (
+	"fmt"
+	"time"
+
+	"github.com/pocketbase/dbx"
+	"github.com/pocketbase/pocketbase/core"
+)
+
+// handleZfsPoolAlert sends alerts when a ZFS pool health state worsens and
+// resolves the alert history entry when the pool recovers. Like the SMART
+// hook, this is automatic and does not require user opt-in.
+func (am *AlertManager) handleZfsPoolAlert(e *core.RecordEvent) error {
+	return am.handleZfsPoolHealthAlert(e, e.Record.Original().GetString("health"))
+}
+
+func (am *AlertManager) handleZfsPoolCreateAlert(e *core.RecordEvent) error {
+	return am.handleZfsPoolHealthAlert(e, "")
+}
+
+func (am *AlertManager) handleZfsPoolHealthAlert(e *core.RecordEvent, oldHealth string) error {
+	newHealth := e.Record.GetString("health")
+	oldSeverity := zfsPoolSeverity(oldHealth)
+	newSeverity := zfsPoolSeverity(newHealth)
+
+	systemID := e.Record.GetString("system")
+	if systemID == "" {
+		return e.Next()
+	}
+
+	systemRecord, err := e.App.FindRecordById("systems", systemID)
+	if err != nil {
+		e.App.Logger().Error("Failed to find system for ZFS alert", "err", err, "systemID", systemID)
+		return e.Next()
+	}
+
+	// Pool recovered to a healthy state: resolve any open history entries.
+	if newSeverity == 1 && oldSeverity > 1 {
+		resolveAllAlertHistoryRecords(e.App, e.Record.Id)
+		return e.Next()
+	}
+
+	if !shouldSendZfsPoolAlert(oldSeverity, newSeverity) {
+		return e.Next()
+	}
+
+	systemName := systemRecord.GetString("name")
+	poolName := e.Record.GetString("name")
+
+	title := fmt.Sprintf("ZFS pool %s on %s: %s", newHealth, systemName, poolName)
+	message := fmt.Sprintf("ZFS pool %s (%s) was first observed as %s", poolName, systemName, newHealth)
+	if oldSeverity > 0 {
+		message = fmt.Sprintf("ZFS pool %s (%s) health changed from %s to %s", poolName, systemName, oldHealth, newHealth)
+	}
+
+	userIDs := systemRecord.GetStringSlice("users")
+	if len(userIDs) == 0 {
+		return e.Next()
+	}
+
+	for _, userID := range userIDs {
+		if err := am.SendAlert(AlertMessageData{
+			UserID:   userID,
+			SystemID: systemID,
+			Title:    title,
+			Message:  message,
+			Link:     am.hub.MakeLink("system", systemID),
+			LinkText: "View " + systemName,
+		}); err != nil {
+			e.App.Logger().Error("Failed to send ZFS alert", "err", err, "userID", userID)
+		}
+		_ = createZfsPoolHistoryRecord(e.App, userID, systemID, e.Record.Id, poolName)
+	}
+
+	return e.Next()
+}
+
+// resolveZfsPoolHistoryOnDelete resolves open alert history entries when a
+// pool record is deleted (manually or because the pool disappeared), so the
+// UI does not keep showing an ongoing alert for a pool that no longer exists.
+func resolveZfsPoolHistoryOnDelete(e *core.RecordEvent) error {
+	resolveAllAlertHistoryRecords(e.App, e.Record.Id)
+	return e.Next()
+}
+
+// shouldSendZfsPoolAlert reports whether a health transition warrants an alert.
+// First observations of unhealthy pools and worsening transitions are reported.
+func shouldSendZfsPoolAlert(oldSeverity, newSeverity int) bool {
+	return newSeverity > 1 && (oldSeverity == 0 || newSeverity > oldSeverity)
+}
+
+// zfsPoolSeverity ranks pool health states: healthy (1), degraded (2),
+// failed/unavailable (3), unknown (0).
+func zfsPoolSeverity(health string) int {
+	switch health {
+	case "ONLINE":
+		return 1
+	case "DEGRADED":
+		return 2
+	case "FAULTED", "OFFLINE", "UNAVAIL", "REMOVED", "SUSPENDED":
+		return 3
+	default:
+		return 0
+	}
+}
+
+// createZfsPoolHistoryRecord logs a pool health alert in the alerts history so
+// it is visible in the UI without creating an editable alert configuration.
+func createZfsPoolHistoryRecord(app core.App, userID, systemID, alertID, poolName string) error {
+	collection, err := app.FindCachedCollectionByNameOrId("alerts_history")
+	if err != nil {
+		return err
+	}
+	record := core.NewRecord(collection)
+	record.Set("user", userID)
+	record.Set("system", systemID)
+	record.Set("alert_id", alertID)
+	record.Set("name", "ZFS Pool: "+poolName)
+	return app.Save(record)
+}
+
+// resolveAllAlertHistoryRecords resolves every open history entry for an alert
+// record id (one per system user).
+func resolveAllAlertHistoryRecords(app core.App, alertID string) {
+	records, err := app.FindRecordsByFilter(
+		"alerts_history",
+		"alert_id={:alert_id} && resolved=null",
+		"", 0, 0,
+		dbx.Params{"alert_id": alertID},
+	)
+	if err != nil || len(records) == 0 {
+		return
+	}
+	now := time.Now().UTC()
+	for _, record := range records {
+		record.Set("resolved", now)
+		if err := app.Save(record); err != nil {
+			app.Logger().Error("Failed to resolve ZFS alert history", "err", err, "recordId", record.Id)
+		}
+	}
+}

internal/alerts/alerts_zfs_disk_test.go

@@ -0,0 +1,145 @@
+//go:build testing
+
+package alerts_test
+
+import (
+	"encoding/json"
+	"testing"
+	"time"
+
+	"github.com/henrygd/beszel/internal/entities/system"
+	beszelTests "github.com/henrygd/beszel/internal/tests"
+
+	"github.com/pocketbase/dbx"
+	"github.com/pocketbase/pocketbase/tools/types"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// TestDiskAlertZfsPoolMultiMinute verifies that ZFS pool usage participates in
+// the Disk threshold alert using historical per-minute values, mirroring the
+// extra-filesystem behavior.
+func TestDiskAlertZfsPoolMultiMinute(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	systems, err := beszelTests.CreateSystems(hub, 1, user.Id, "up")
+	require.NoError(t, err)
+	systemRecord := systems[0]
+
+	diskAlert, err := beszelTests.CreateRecord(hub, "alerts", map[string]any{
+		"name":   "Disk",
+		"system": systemRecord.Id,
+		"user":   user.Id,
+		"value":  80, // threshold: 80%
+		"min":    2,  // requires historical averaging
+	})
+	require.NoError(t, err)
+
+	am := hub.GetAlertManager()
+	now := time.Now().UTC()
+
+	poolHigh := map[string]*system.ZfsPool{
+		"tank": {Total: 1000, Used: 920}, // 92% - above threshold
+	}
+
+	recordTimes := []time.Duration{
+		-180 * time.Second,
+		-90 * time.Second,
+		-60 * time.Second,
+		-30 * time.Second,
+	}
+
+	for _, offset := range recordTimes {
+		stats := system.Stats{
+			DiskPct:  30, // root disk at 30% - below threshold
+			ZfsPools: poolHigh,
+		}
+		statsJSON, _ := json.Marshal(stats)
+
+		recordTime := now.Add(offset)
+		record, err := beszelTests.CreateRecord(hub, "system_stats", map[string]any{
+			"system": systemRecord.Id,
+			"type":   "1m",
+			"stats":  string(statsJSON),
+		})
+		require.NoError(t, err)
+		record.SetRaw("created", recordTime.Format(types.DefaultDateLayout))
+		err = hub.SaveNoValidate(record)
+		require.NoError(t, err)
+	}
+
+	combinedDataHigh := &system.CombinedData{
+		Stats: system.Stats{
+			DiskPct:  30,
+			ZfsPools: poolHigh,
+		},
+		Info: system.Info{
+			DiskPct: 30,
+		},
+	}
+
+	systemRecord.Set("updated", now)
+	err = hub.SaveNoValidate(systemRecord)
+	require.NoError(t, err)
+
+	err = am.HandleSystemAlerts(systemRecord, combinedDataHigh)
+	require.NoError(t, err)
+
+	time.Sleep(20 * time.Millisecond)
+
+	diskAlert, err = hub.FindFirstRecordByFilter("alerts", "id={:id}", dbx.Params{"id": diskAlert.Id})
+	require.NoError(t, err)
+	assert.True(t, diskAlert.GetBool("triggered"),
+		"Alert should be triggered when ZFS pool average (92%%) exceeds threshold (80%%)")
+
+	// --- Resolution: pool drops to 50%, alert should resolve ---
+
+	poolLow := map[string]*system.ZfsPool{
+		"tank": {Total: 1000, Used: 500}, // 50% - below threshold
+	}
+
+	newNow := now.Add(2 * time.Minute)
+	for _, offset := range recordTimes {
+		stats := system.Stats{
+			DiskPct:  30,
+			ZfsPools: poolLow,
+		}
+		statsJSON, _ := json.Marshal(stats)
+
+		recordTime := newNow.Add(offset)
+		record, err := beszelTests.CreateRecord(hub, "system_stats", map[string]any{
+			"system": systemRecord.Id,
+			"type":   "1m",
+			"stats":  string(statsJSON),
+		})
+		require.NoError(t, err)
+		record.SetRaw("created", recordTime.Format(types.DefaultDateLayout))
+		err = hub.SaveNoValidate(record)
+		require.NoError(t, err)
+	}
+
+	combinedDataLow := &system.CombinedData{
+		Stats: system.Stats{
+			DiskPct:  30,
+			ZfsPools: poolLow,
+		},
+		Info: system.Info{
+			DiskPct: 30,
+		},
+	}
+
+	systemRecord.Set("updated", newNow)
+	err = hub.SaveNoValidate(systemRecord)
+	require.NoError(t, err)
+
+	err = am.HandleSystemAlerts(systemRecord, combinedDataLow)
+	require.NoError(t, err)
+
+	time.Sleep(20 * time.Millisecond)
+
+	diskAlert, err = hub.FindFirstRecordByFilter("alerts", "id={:id}", dbx.Params{"id": diskAlert.Id})
+	require.NoError(t, err)
+	assert.False(t, diskAlert.GetBool("triggered"),
+		"Alert should be resolved when ZFS pool average (50%%) drops below threshold (80%%)")
+}

internal/alerts/alerts_zfs_key_test.go

@@ -0,0 +1,15 @@
+//go:build testing
+
+package alerts
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+func TestZfsDiskAlertKeyIsNamespaced(t *testing.T) {
+	assert.Equal(t, "zfs:tank", zfsDiskAlertKey("tank"))
+	assert.Equal(t, "Usage of ZFS pool tank", diskAlertDescriptor(zfsDiskAlertKey("tank")))
+	assert.Equal(t, "Usage of tank", diskAlertDescriptor("tank"))
+}

internal/alerts/alerts_zfs_test.go

@@ -0,0 +1,292 @@
+//go:build testing
+
+package alerts_test
+
+import (
+	"testing"
+	"time"
+
+	beszelTests "github.com/henrygd/beszel/internal/tests"
+	"github.com/pocketbase/pocketbase/core"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestZfsPoolAlertOnlineToDegraded(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	system, err := beszelTests.CreateRecord(hub, "systems", map[string]any{
+		"name":  "test-system",
+		"users": []string{user.Id},
+		"host":  "127.0.0.1",
+	})
+	assert.NoError(t, err)
+
+	pool, err := beszelTests.CreateRecord(hub, "zfs_pools", map[string]any{
+		"system": system.Id,
+		"name":   "tank",
+		"health": "ONLINE",
+	})
+	assert.NoError(t, err)
+
+	// Re-fetch so PocketBase tracks original values
+	pool, err = hub.FindRecordById("zfs_pools", pool.Id)
+	assert.NoError(t, err)
+
+	pool.Set("health", "DEGRADED")
+	err = hub.Save(pool)
+	assert.NoError(t, err)
+
+	time.Sleep(50 * time.Millisecond)
+
+	assert.EqualValues(t, 1, hub.TestMailer.TotalSend(), "should have 1 email sent after pool became DEGRADED")
+	lastMessage := hub.TestMailer.LastMessage()
+	assert.Contains(t, lastMessage.Subject, "ZFS pool DEGRADED on test-system")
+	assert.Contains(t, lastMessage.Subject, "tank")
+	assert.Contains(t, lastMessage.Text, "ONLINE to DEGRADED")
+}
+
+func TestZfsPoolAlertDegradedToFaulted(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	system, err := beszelTests.CreateRecord(hub, "systems", map[string]any{
+		"name":  "test-system",
+		"users": []string{user.Id},
+		"host":  "127.0.0.1",
+	})
+	assert.NoError(t, err)
+
+	pool, err := beszelTests.CreateRecord(hub, "zfs_pools", map[string]any{
+		"system": system.Id,
+		"name":   "rpool",
+		"health": "DEGRADED",
+	})
+	assert.NoError(t, err)
+
+	pool, err = hub.FindRecordById("zfs_pools", pool.Id)
+	assert.NoError(t, err)
+
+	pool.Set("health", "FAULTED")
+	err = hub.Save(pool)
+	assert.NoError(t, err)
+
+	time.Sleep(50 * time.Millisecond)
+
+	assert.EqualValues(t, 2, hub.TestMailer.TotalSend(), "should alert on initial DEGRADED state and later FAULTED transition")
+	lastMessage := hub.TestMailer.LastMessage()
+	assert.Contains(t, lastMessage.Subject, "ZFS pool FAULTED on test-system")
+}
+
+func TestZfsPoolAlertNoAlertOnRecovery(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	system, err := beszelTests.CreateRecord(hub, "systems", map[string]any{
+		"name":  "test-system",
+		"users": []string{user.Id},
+		"host":  "127.0.0.1",
+	})
+	assert.NoError(t, err)
+
+	pool, err := beszelTests.CreateRecord(hub, "zfs_pools", map[string]any{
+		"system": system.Id,
+		"name":   "tank",
+		"health": "DEGRADED",
+	})
+	assert.NoError(t, err)
+
+	// Trigger a worsening alert first
+	pool, err = hub.FindRecordById("zfs_pools", pool.Id)
+	assert.NoError(t, err)
+	pool.Set("health", "FAULTED")
+	err = hub.Save(pool)
+	assert.NoError(t, err)
+	time.Sleep(50 * time.Millisecond)
+	assert.EqualValues(t, 2, hub.TestMailer.TotalSend(), "expected alerts for initial DEGRADED state and DEGRADED -> FAULTED")
+
+	// Recovery back to ONLINE must not send a new alert
+	pool, err = hub.FindRecordById("zfs_pools", pool.Id)
+	assert.NoError(t, err)
+	pool.Set("health", "ONLINE")
+	err = hub.Save(pool)
+	assert.NoError(t, err)
+	time.Sleep(50 * time.Millisecond)
+	assert.EqualValues(t, 2, hub.TestMailer.TotalSend(), "recovery should not send a new alert")
+
+	// And the open history entry should have been resolved
+	history, err := hub.FindRecordsByFilter("alerts_history", "alert_id={:alert_id}", "", 0, 0, map[string]any{"alert_id": pool.Id})
+	assert.NoError(t, err)
+	requireHistoryResolved(t, history)
+}
+
+func TestZfsPoolAlertUnknownHealthDoesNotResolve(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	system, err := beszelTests.CreateRecord(hub, "systems", map[string]any{
+		"name":  "test-system",
+		"users": []string{user.Id},
+		"host":  "127.0.0.1",
+	})
+	require.NoError(t, err)
+
+	pool, err := beszelTests.CreateRecord(hub, "zfs_pools", map[string]any{
+		"system": system.Id,
+		"name":   "tank",
+		"health": "DEGRADED",
+	})
+	require.NoError(t, err)
+	time.Sleep(50 * time.Millisecond)
+
+	pool, err = hub.FindRecordById("zfs_pools", pool.Id)
+	require.NoError(t, err)
+	pool.Set("health", "")
+	require.NoError(t, hub.Save(pool))
+	time.Sleep(50 * time.Millisecond)
+
+	history, err := hub.FindRecordsByFilter("alerts_history", "alert_id={:alert_id} && resolved=null", "", 0, 0, map[string]any{"alert_id": pool.Id})
+	require.NoError(t, err)
+	require.Len(t, history, 1, "unknown health must not resolve an active alert")
+}
+
+func TestZfsPoolAlertUnknownToFaulted(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	system, err := beszelTests.CreateRecord(hub, "systems", map[string]any{
+		"name":  "test-system",
+		"users": []string{user.Id},
+		"host":  "127.0.0.1",
+	})
+	assert.NoError(t, err)
+
+	pool, err := beszelTests.CreateRecord(hub, "zfs_pools", map[string]any{
+		"system": system.Id,
+		"name":   "tank",
+		"health": "",
+	})
+	assert.NoError(t, err)
+
+	pool, err = hub.FindRecordById("zfs_pools", pool.Id)
+	assert.NoError(t, err)
+
+	pool.Set("health", "FAULTED")
+	err = hub.Save(pool)
+	assert.NoError(t, err)
+
+	time.Sleep(50 * time.Millisecond)
+
+	assert.EqualValues(t, 1, hub.TestMailer.TotalSend(), "should alert when a previously unknown pool becomes FAULTED")
+}
+
+func TestZfsPoolAlertOnInitialUnhealthyState(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	system, err := beszelTests.CreateRecord(hub, "systems", map[string]any{
+		"name":  "test-system",
+		"users": []string{user.Id},
+		"host":  "127.0.0.1",
+	})
+	require.NoError(t, err)
+
+	pool, err := beszelTests.CreateRecord(hub, "zfs_pools", map[string]any{
+		"system": system.Id,
+		"name":   "tank",
+		"health": "DEGRADED",
+	})
+	require.NoError(t, err)
+	time.Sleep(50 * time.Millisecond)
+
+	require.EqualValues(t, 1, hub.TestMailer.TotalSend())
+	assert.Contains(t, hub.TestMailer.LastMessage().Text, "first observed as DEGRADED")
+
+	pool, err = hub.FindRecordById("zfs_pools", pool.Id)
+	require.NoError(t, err)
+	require.NoError(t, hub.Save(pool))
+	time.Sleep(50 * time.Millisecond)
+	assert.EqualValues(t, 1, hub.TestMailer.TotalSend(), "unchanged unhealthy health must not duplicate alerts")
+}
+
+func TestZfsPoolAlertWritesHistory(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	system, err := beszelTests.CreateRecord(hub, "systems", map[string]any{
+		"name":  "test-system",
+		"users": []string{user.Id},
+		"host":  "127.0.0.1",
+	})
+	assert.NoError(t, err)
+
+	pool, err := beszelTests.CreateRecord(hub, "zfs_pools", map[string]any{
+		"system": system.Id,
+		"name":   "tank",
+		"health": "ONLINE",
+	})
+	assert.NoError(t, err)
+
+	pool, err = hub.FindRecordById("zfs_pools", pool.Id)
+	assert.NoError(t, err)
+	pool.Set("health", "FAULTED")
+	err = hub.Save(pool)
+	assert.NoError(t, err)
+
+	time.Sleep(50 * time.Millisecond)
+
+	history, err := hub.FindRecordsByFilter("alerts_history", "alert_id={:alert_id}", "", 0, 0, map[string]any{"alert_id": pool.Id})
+	assert.NoError(t, err)
+	require.Len(t, history, 1, "expected one history entry per user")
+	assert.Equal(t, "ZFS Pool: tank", history[0].GetString("name"))
+	assert.Equal(t, system.Id, history[0].GetString("system"))
+}
+
+func TestZfsPoolAlertResolvedOnRecordDelete(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	system, err := beszelTests.CreateRecord(hub, "systems", map[string]any{
+		"name":  "test-system",
+		"users": []string{user.Id},
+		"host":  "127.0.0.1",
+	})
+	assert.NoError(t, err)
+
+	pool, err := beszelTests.CreateRecord(hub, "zfs_pools", map[string]any{
+		"system": system.Id,
+		"name":   "tank",
+		"health": "ONLINE",
+	})
+	assert.NoError(t, err)
+
+	// Trigger an alert so an open history entry exists.
+	pool, err = hub.FindRecordById("zfs_pools", pool.Id)
+	assert.NoError(t, err)
+	pool.Set("health", "FAULTED")
+	err = hub.Save(pool)
+	assert.NoError(t, err)
+	time.Sleep(50 * time.Millisecond)
+
+	history, err := hub.FindRecordsByFilter("alerts_history", "alert_id={:alert_id} && resolved=null", "", 0, 0, map[string]any{"alert_id": pool.Id})
+	assert.NoError(t, err)
+	require.Len(t, history, 1, "expected one open history entry")
+
+	// Deleting the pool record must resolve the open entry.
+	err = hub.Delete(pool)
+	assert.NoError(t, err)
+	time.Sleep(50 * time.Millisecond)
+
+	history, err = hub.FindRecordsByFilter("alerts_history", "alert_id={:alert_id}", "", 0, 0, map[string]any{"alert_id": pool.Id})
+	assert.NoError(t, err)
+	require.Len(t, history, 1)
+	requireHistoryResolved(t, history)
+}
+
+func requireHistoryResolved(t *testing.T, history []*core.Record) {
+	t.Helper()
+	for _, record := range history {
+		assert.False(t, record.GetDateTime("resolved").Time().IsZero(), "expected history entry to be resolved")
+	}
+}

internal/cmd/agent/agent.go

@@ -195,6 +195,6 @@ func main() {
 	}
 
 	if err := a.Start(serverConfig); err != nil {
-		log.Fatal("Failed to start server: ", err)
+		log.Fatal("Failed to start: ", err)
 	}
 }