mirrors/seaweedfs
1
0
Fork 0
mirror of https://github.com/seaweedfs/seaweedfs.git synced 2026-08-28 11:56:07 +00:00
Code Issues Packages Projects Releases Wiki Activity

telemetry: let the dashboard pick the confirmation window (#10904)

Browse Source 
* telemetry: let the dashboard pick the confirmation window

* telemetry: cover the serialized threshold map through the stats handler
This commit is contained in:
Chris Lu
2026-08-23 21:48:33 -07:00
committed by GitHub
parent 9c8d3b6a81
commit 71a8c77a36
5 changed files with 100 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
telemetry/server/api/handlers_test.go
+44 -2
View File
@@ -2,6 +2,7 @@ package api
import (
"bytes"
"encoding/json"
"net/http"
"net/http/httptest"
"testing"
@@ -11,6 +12,9 @@ import (
protobuf "google.golang.org/protobuf/proto"
)
// promauto registers on the global registry: one storage per test binary.
var testHandler = NewHandler(storage.NewPrometheusStorage())
func validReport() *proto.TelemetryData {
return &proto.TelemetryData{
TopologyId: "38422678-6a0d-4482-aa33-65b90010ac47",
@@ -43,8 +47,7 @@ func marshalReport(t *testing.T, data *proto.TelemetryData) []byte {
}
func TestCollectTelemetryValidation(t *testing.T) {
// promauto registers on the global registry: one storage per test binary.
h := NewHandler(storage.NewPrometheusStorage())
h := testHandler
t.Run("valid report accepted", func(t *testing.T) {
if w := postCollect(t, h, marshalReport(t, validReport()), "application/x-protobuf"); w.Code != http.StatusOK {
@@ -96,3 +99,42 @@ func TestCollectTelemetryValidation(t *testing.T) {
}
})
}
// The dashboard looks confirmation windows up by the select's string value,
// so the stats JSON must key confirmed_by_days by decimal strings and carry
// unmet thresholds as zeros.
func TestStatsSerializedThresholds(t *testing.T) {
data := validReport()
data.TopologyId = "49533789-7b1e-4593-bb44-76ca1121bd58"
if w := postCollect(t, testHandler, marshalReport(t, data), "application/x-protobuf"); w.Code != http.StatusOK {
t.Fatalf("collect: got %d: %s", w.Code, w.Body.String())
}
req := httptest.NewRequest(http.MethodGet, "/api/stats", nil)
w := httptest.NewRecorder()
testHandler.GetStats(w, req)
if w.Code != http.StatusOK {
t.Fatalf("stats: got %d", w.Code)
}
var stats struct {
ConfirmedByDays map[string]int `json:"confirmed_by_days"`
}
if err := json.Unmarshal(w.Body.Bytes(), &stats); err != nil {
t.Fatalf("decode: %v", err)
}
if len(stats.ConfirmedByDays) != 5 {
t.Fatalf("confirmed_by_days = %v, want the 5 thresholds", stats.ConfirmedByDays)
}
for _, key := range []string{"1", "3", "7", "14", "30"} {
if _, ok := stats.ConfirmedByDays[key]; !ok {
t.Errorf("confirmed_by_days missing %q: %v", key, stats.ConfirmedByDays)
}
}
if stats.ConfirmedByDays["1"] < 1 {
t.Errorf("fresh cluster missing from the 1-day count: %v", stats.ConfirmedByDays)
}
if stats.ConfirmedByDays["30"] != 0 {
t.Errorf("unmet threshold not zero: %v", stats.ConfirmedByDays)
}
}
telemetry/server/dashboard/dashboard.go
+30 -2
View File
@@ -57,6 +57,15 @@ func (h *Handler) ServeIndex(w http.ResponseWriter, r *http.Request) {
color: #666;
margin-top: 5px;
}
.stat-label select {
border: none;
background: none;
color: inherit;
font: inherit;
padding: 0;
cursor: pointer;
text-decoration: underline dotted;
}
.chart-container {
background: white;
padding: 20px;
@@ -129,7 +138,14 @@ func (h *Handler) ServeIndex(w http.ResponseWriter, r *http.Request) {
</div>
<div class="stat-card">
<div class="stat-value" id="confirmedInstances">-</div>
<div class="stat-label">Confirmed Clusters (7+ days)</div>
<div class="stat-label">Confirmed Clusters
(<select id="confirmDays" onchange="updateConfirmed()">
<option value="1">1+</option>
<option value="3">3+</option>
<option value="7" selected>7+</option>
<option value="14">14+</option>
<option value="30">30+</option>
</select> days)</div>
</div>
<div class="stat-card">
<div class="stat-value" id="totalVersions">-</div>
@@ -229,14 +245,26 @@ func (h *Handler) ServeIndex(w http.ResponseWriter, r *http.Request) {
}
}
let latestStats = {};
function updateStats(stats) {
latestStats = stats;
document.getElementById('totalInstances').textContent = stats.total_instances || 0;
document.getElementById('activeInstances').textContent = stats.active_instances || 0;
document.getElementById('confirmedInstances').textContent = stats.confirmed_instances || 0;
updateConfirmed();
document.getElementById('totalVersions').textContent = Object.keys(stats.versions || {}).length;
document.getElementById('totalOS').textContent = Object.keys(stats.os_distribution || {}).length;
}
// Servers from before confirmed_by_days fall back to the fixed
// 7-day count.
function updateConfirmed() {
const days = document.getElementById('confirmDays').value;
const byDays = latestStats.confirmed_by_days || {};
const count = byDays[days] !== undefined ? byDays[days] : latestStats.confirmed_instances;
document.getElementById('confirmedInstances').textContent = count || 0;
}
function updateCharts(stats) {
createPieChart('versionChart', 'Version Distribution', stats.versions || {});
createPieChart('osChart', 'Operating System Distribution', stats.os_distribution || {});
telemetry/server/storage/confirmed_test.go
+10
View File
@@ -74,4 +74,14 @@ func TestConfirmedClusters(t *testing.T) {
if _, ok := v["9.99"]; ok {
t.Errorf("one-shot cluster polluted the distribution: %v", v)
}
// Cluster A meets the 1/3/7-day thresholds, B only the 1-day one, and
// unmet thresholds are present as zero so the dashboard can show them.
byDays := stats["confirmed_by_days"].(map[int]int)
want := map[int]int{1: 2, 3: 1, 7: 1, 14: 0, 30: 0}
for threshold, expected := range want {
if got, ok := byDays[threshold]; !ok || got != expected {
t.Errorf("confirmed_by_days[%d] = %v (present=%v), want %d", threshold, got, ok, expected)
}
}
}
telemetry/server/storage/history.go
+4
View File
@@ -10,6 +10,10 @@ import (
// on before it counts as confirmed in the aggregated stats.
const confirmDays = 7
// confirmThresholds are the confirmation windows the dashboard lets the
// viewer pick between; confirmDays is the one everything else is built on.
var confirmThresholds = []int{1, 3, 7, 14, 30}
// activeDays is how recently a cluster must have reported to count as active.
const activeDays = 7
telemetry/server/storage/prometheus.go
+12 -1
View File
@@ -207,6 +207,10 @@ func (s *PrometheusStorage) updateStats() {
totalInstances := 0
activeInstances := 0
confirmedInstances := 0
confirmedByDays := make(map[int]int, len(confirmThresholds))
for _, threshold := range confirmThresholds {
confirmedByDays[threshold] = 0
}
versionsAll := make(map[string]int)
osAll := make(map[string]int)
versionsConfirmed := make(map[string]int)
@@ -223,7 +227,13 @@ func (s *PrometheusStorage) updateStats() {
// A cluster is confirmed once seen on confirmDays distinct UTC
// days (histories hold one sample per day), so short-lived
// clusters can't skew the distributions below.
if len(s.histories[instance.TelemetryData.TopologyId]) >= confirmDays {
daysSeen := len(s.histories[instance.TelemetryData.TopologyId])
for _, threshold := range confirmThresholds {
if daysSeen >= threshold {
confirmedByDays[threshold]++
}
}
if daysSeen >= confirmDays {
confirmedInstances++
versionsConfirmed[instance.TelemetryData.Version]++
osConfirmed[instance.TelemetryData.Os]++
@@ -249,6 +259,7 @@ func (s *PrometheusStorage) updateStats() {
"total_instances": totalInstances,
"active_instances": activeInstances,
"confirmed_instances": confirmedInstances,
"confirmed_by_days": confirmedByDays,
"versions": versions,
"os_distribution": osDistribution,
}