Skip to content

metrics: add canary label to metrics and clean up stale series on canary/namespace removal #1961

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
dkulchinsky wants to merge 3 commits into
base: main
Choose a base branch
from
Open

metrics: add canary label to metrics and clean up stale series on canary/namespace removal #1961

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
fb03c3c
metrics: add name label to canary_weight and clean up stale series on…
dkulchinsky Aug 23, 2026
b521c9f
refactor to add common canary label to all metrics indicating the can…
dkulchinsky Aug 25, 2026
260a2ad
emit flagger_canary_weight in steady state
dkulchinsky Aug 26, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 6 additions & 0 deletions pkg/controller/controller.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -64,6 +64,10 @@ type Controller struct {
logger *zap.SugaredLogger
canaries *sync.Map
jobs map[string]CanaryJob
// namespaces we have previously reported a canary_total value for;
// used to remove stale per-namespace series when a namespace no
// longer contains any Canary resources.
trackedNamespaces map[string]struct{}
recorder metrics.Recorder
notifier notifier.Interface
canaryFactory *canary.Factory
Expand Down Expand Up @@ -123,6 +127,7 @@ func NewController(
logger: logger,
canaries: new(sync.Map),
jobs: map[string]CanaryJob{},
trackedNamespaces: map[string]struct{}{},
flaggerWindow: flaggerWindow,
observerFactory: observerFactory,
recorder: recorder,
Expand Down Expand Up @@ -180,6 +185,7 @@ func NewController(
if ok {
ctrl.logger.Infof("Deleting %s.%s from cache", r.Name, r.Namespace)
ctrl.canaries.Delete(fmt.Sprintf("%s.%s", r.Name, r.Namespace))
ctrl.recorder.DeleteFor(r.Name, r.Namespace)
}
},
})
Expand Down
21 changes: 21 additions & 0 deletions pkg/controller/scheduler.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -153,6 +153,17 @@ func (c *Controller) scheduleCanaries() {
for k, v := range stats {
c.recorder.SetTotal(k, v)
}

// remove canary_total series for namespaces that no longer contain any canaries
for ns := range c.trackedNamespaces {
if _, ok := stats[ns]; !ok {
c.recorder.DeleteTotalFor(ns)
delete(c.trackedNamespaces, ns)
}
}
for ns := range stats {
c.trackedNamespaces[ns] = struct{}{}
}
}

func (c *Controller) advanceCanary(name string, namespace string) {
Expand Down Expand Up @@ -286,6 +297,12 @@ func (c *Controller) advanceCanary(name string, namespace string) {

if !shouldAdvance {
c.recorder.SetStatus(cd, cd.Status.Phase)
// keep the weight gauge populated in steady state (e.g. after a
// controller restart or right after initialization), otherwise it
// is only emitted once a canary starts progressing
if primaryWeight, canaryWeight, _, err := meshRouter.GetRoutes(cd); err == nil {
c.recorder.SetWeight(cd, primaryWeight, canaryWeight)
}
return
}

Expand Down Expand Up @@ -410,6 +427,7 @@ func (c *Controller) advanceCanary(name string, namespace string) {

c.recorder.SetStatus(cd, flaggerv1.CanaryPhaseSucceeded)
c.recorder.IncSuccesses(metrics.CanaryMetricLabels{
Canary: cd.Name,
Name: cd.Spec.TargetRef.Name,
Namespace: cd.Namespace,
DeploymentStrategy: cd.DeploymentStrategy(),
Expand Down Expand Up @@ -832,6 +850,7 @@ func (c *Controller) shouldSkipAnalysis(canary *flaggerv1.Canary, canaryControll
// notify
c.recorder.SetStatus(canary, flaggerv1.CanaryPhaseSucceeded)
c.recorder.IncSuccesses(metrics.CanaryMetricLabels{
Canary: canary.Name,
Name: canary.Spec.TargetRef.Name,
Namespace: canary.Namespace,
DeploymentStrategy: canary.DeploymentStrategy(),
Expand Down Expand Up @@ -989,6 +1008,7 @@ func (c *Controller) rollback(canary *flaggerv1.Canary, canaryController canary.

c.recorder.SetStatus(canary, flaggerv1.CanaryPhaseFailed)
c.recorder.IncFailures(metrics.CanaryMetricLabels{
Canary: canary.Name,
Name: canary.Spec.TargetRef.Name,
Namespace: canary.Namespace,
DeploymentStrategy: canary.DeploymentStrategy(),
Expand Down Expand Up @@ -1026,6 +1046,7 @@ func (c *Controller) handleFailedPromotion(canary *flaggerv1.Canary, canaryContr

c.recorder.SetStatus(canary, flaggerv1.CanaryPhaseFailed)
c.recorder.IncFailures(metrics.CanaryMetricLabels{
Canary: canary.Name,
Name: canary.Spec.TargetRef.Name,
Namespace: canary.Namespace,
DeploymentStrategy: canary.DeploymentStrategy(),
Expand Down
48 changes: 36 additions & 12 deletions pkg/controller/scheduler_metrics_test.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -194,7 +194,7 @@ func TestController_MetricsStateTransition(t *testing.T) {
mocks.makePrimaryReady(t)
mocks.ctrl.advanceCanary("podinfo", "default")

actualStatus := testutil.ToFloat64(mocks.ctrl.recorder.GetStatusMetric().WithLabelValues("podinfo", "default"))
actualStatus := testutil.ToFloat64(mocks.ctrl.recorder.GetStatusMetric().WithLabelValues("podinfo", "podinfo", "default"))
assert.Equal(t, float64(1), actualStatus)

actualTotal := testutil.ToFloat64(mocks.ctrl.recorder.GetTotalMetric().WithLabelValues("default"))
Expand All @@ -207,11 +207,11 @@ func TestController_MetricsStateTransition(t *testing.T) {
mocks.makeCanaryReady(t)
mocks.ctrl.advanceCanary("podinfo", "default")

actualStatus = testutil.ToFloat64(mocks.ctrl.recorder.GetStatusMetric().WithLabelValues("podinfo", "default"))
actualStatus = testutil.ToFloat64(mocks.ctrl.recorder.GetStatusMetric().WithLabelValues("podinfo", "podinfo", "default"))
assert.Equal(t, float64(0), actualStatus)

actualPrimaryWeight := testutil.ToFloat64(mocks.ctrl.recorder.GetWeightMetric().WithLabelValues("podinfo-primary", "default"))
actualCanaryWeight := testutil.ToFloat64(mocks.ctrl.recorder.GetWeightMetric().WithLabelValues("podinfo", "default"))
actualPrimaryWeight := testutil.ToFloat64(mocks.ctrl.recorder.GetWeightMetric().WithLabelValues("podinfo", "podinfo-primary", "default"))
actualCanaryWeight := testutil.ToFloat64(mocks.ctrl.recorder.GetWeightMetric().WithLabelValues("podinfo", "podinfo", "default"))

t.Logf("Progression weights - Primary: %f, Canary: %f", actualPrimaryWeight, actualCanaryWeight)
assert.GreaterOrEqual(t, actualPrimaryWeight, float64(50))
Expand All @@ -232,7 +232,7 @@ func TestController_MetricsStateTransition(t *testing.T) {
}
}

successCount := testutil.ToFloat64(mocks.ctrl.recorder.GetSuccessesMetric().WithLabelValues("podinfo", "default", "canary", "completed"))
successCount := testutil.ToFloat64(mocks.ctrl.recorder.GetSuccessesMetric().WithLabelValues("podinfo", "podinfo", "default", "canary", "completed"))
assert.Equal(t, float64(1), successCount)
})

Expand Down Expand Up @@ -267,15 +267,15 @@ func TestController_MetricsStateTransition(t *testing.T) {
mocks.ctrl.advanceCanary("podinfo", "default")
mocks.ctrl.advanceCanary("podinfo", "default")

actualStatus := testutil.ToFloat64(mocks.ctrl.recorder.GetStatusMetric().WithLabelValues("podinfo", "default"))
actualStatus := testutil.ToFloat64(mocks.ctrl.recorder.GetStatusMetric().WithLabelValues("podinfo", "podinfo", "default"))
assert.Equal(t, float64(2), actualStatus)

actualPrimaryWeight := testutil.ToFloat64(mocks.ctrl.recorder.GetWeightMetric().WithLabelValues("podinfo-primary", "default"))
actualCanaryWeight := testutil.ToFloat64(mocks.ctrl.recorder.GetWeightMetric().WithLabelValues("podinfo", "default"))
actualPrimaryWeight := testutil.ToFloat64(mocks.ctrl.recorder.GetWeightMetric().WithLabelValues("podinfo", "podinfo-primary", "default"))
actualCanaryWeight := testutil.ToFloat64(mocks.ctrl.recorder.GetWeightMetric().WithLabelValues("podinfo", "podinfo", "default"))
assert.Equal(t, float64(100), actualPrimaryWeight)
assert.Equal(t, float64(0), actualCanaryWeight)

failureCount := testutil.ToFloat64(mocks.ctrl.recorder.GetFailuresMetric().WithLabelValues("podinfo", "default", "canary", "completed"))
failureCount := testutil.ToFloat64(mocks.ctrl.recorder.GetFailuresMetric().WithLabelValues("podinfo", "podinfo", "default", "canary", "completed"))
assert.Equal(t, float64(1), failureCount)
})

Expand Down Expand Up @@ -303,9 +303,33 @@ func TestController_MetricsStateTransition(t *testing.T) {
mocks.makeCanaryReady(t)
mocks.ctrl.advanceCanary("podinfo", "default")

successCount := testutil.ToFloat64(mocks.ctrl.recorder.GetSuccessesMetric().WithLabelValues("podinfo", "default", "canary", "skipped"))
successCount := testutil.ToFloat64(mocks.ctrl.recorder.GetSuccessesMetric().WithLabelValues("podinfo", "podinfo", "default", "canary", "skipped"))
assert.Equal(t, float64(1), successCount)
})

t.Run("weight is emitted for an initialized canary before progression", func(t *testing.T) {
mocks := newDeploymentFixture(nil)

// drive the canary to the Initialized phase
mocks.ctrl.advanceCanary("podinfo", "default")
mocks.makePrimaryReady(t)
mocks.ctrl.advanceCanary("podinfo", "default")

c, err := mocks.flaggerClient.FlaggerV1beta1().Canaries("default").Get(context.TODO(), "podinfo", metav1.GetOptions{})
require.NoError(t, err)
require.Equal(t, flaggerv1.CanaryPhaseInitialized, c.Status.Phase)

// a steady-state reconcile (no changes, shouldAdvance == false) must still
// populate the weight gauge, otherwise it is absent until the canary progresses
mocks.ctrl.advanceCanary("podinfo", "default")

assert.Equal(t, 2, testutil.CollectAndCount(mocks.ctrl.recorder.GetWeightMetric(), "flagger_canary_weight"))

primaryWeight := testutil.ToFloat64(mocks.ctrl.recorder.GetWeightMetric().WithLabelValues("podinfo", "podinfo-primary", "default"))
canaryWeight := testutil.ToFloat64(mocks.ctrl.recorder.GetWeightMetric().WithLabelValues("podinfo", "podinfo", "default"))
assert.Equal(t, float64(100), primaryWeight)
assert.Equal(t, float64(0), canaryWeight)
})
}

func TestController_AnalysisMetricsRecording(t *testing.T) {
Expand Down Expand Up @@ -347,10 +371,10 @@ func TestController_AnalysisMetricsRecording(t *testing.T) {
result := mocks.ctrl.runMetricChecks(canary)
assert.True(t, result)

successRateMetric := mocks.ctrl.recorder.GetAnalysisMetric().WithLabelValues("podinfo", "default", "request-success-rate")
successRateMetric := mocks.ctrl.recorder.GetAnalysisMetric().WithLabelValues("podinfo", "podinfo", "default", "request-success-rate")
assert.NotNil(t, successRateMetric)

durationMetric := mocks.ctrl.recorder.GetAnalysisMetric().WithLabelValues("podinfo", "default", "request-duration")
durationMetric := mocks.ctrl.recorder.GetAnalysisMetric().WithLabelValues("podinfo", "podinfo", "default", "request-duration")
assert.NotNil(t, durationMetric)
})
}
46 changes: 34 additions & 12 deletions pkg/metrics/recorder.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -39,6 +39,9 @@ const (

// CanaryMetricLabels holds labels for canary metrics
type CanaryMetricLabels struct {
// Canary is the name of the Canary resource
Canary string
// Name is the name of the target workload (spec.targetRef.name)
Name string
Namespace string
DeploymentStrategy string
Expand All @@ -47,7 +50,7 @@ type CanaryMetricLabels struct {

// Values returns label values as a slice for Prometheus metrics
func (c CanaryMetricLabels) Values() []string {
return []string{c.Name, c.Namespace, c.DeploymentStrategy, c.AnalysisStatus}
return []string{c.Canary, c.Name, c.Namespace, c.DeploymentStrategy, c.AnalysisStatus}
}

// Recorder records the canary analysis as Prometheus metrics
Expand Down Expand Up @@ -75,7 +78,7 @@ func NewRecorder(controller string, register bool) Recorder {
Name: "canary_duration_seconds",
Help: "Seconds spent performing canary analysis.",
Buckets: prometheus.DefBuckets,
}, []string{"name", "namespace"})
}, []string{"canary", "name", "namespace"})

total := prometheus.NewGaugeVec(prometheus.GaugeOpts{
Subsystem: controller,
Expand All @@ -88,31 +91,31 @@ func NewRecorder(controller string, register bool) Recorder {
Subsystem: controller,
Name: "canary_status",
Help: "Last canary analysis result",
}, []string{"name", "namespace"})
}, []string{"canary", "name", "namespace"})

weight := prometheus.NewGaugeVec(prometheus.GaugeOpts{
Subsystem: controller,
Name: "canary_weight",
Help: "The virtual service destination weight current value",
}, []string{"workload", "namespace"})
}, []string{"canary", "workload", "namespace"})

analysis := prometheus.NewGaugeVec(prometheus.GaugeOpts{
Subsystem: controller,
Name: "canary_metric_analysis",
Help: "Last canary analysis result per metric",
}, []string{"name", "namespace", "metric"})
}, []string{"canary", "name", "namespace", "metric"})

successes := prometheus.NewCounterVec(prometheus.CounterOpts{
Subsystem: controller,
Name: "canary_successes_total",
Help: "Total number of canary successes",
}, []string{"name", "namespace", "deployment_strategy", "analysis_status"})
}, []string{"canary", "name", "namespace", "deployment_strategy", "analysis_status"})

failures := prometheus.NewCounterVec(prometheus.CounterOpts{
Subsystem: controller,
Name: "canary_failures_total",
Help: "Total number of canary failures",
}, []string{"name", "namespace", "deployment_strategy", "analysis_status"})
}, []string{"canary", "name", "namespace", "deployment_strategy", "analysis_status"})

if register {
prometheus.MustRegister(info)
Expand Down Expand Up @@ -144,7 +147,7 @@ func (cr *Recorder) SetInfo(version string, meshProvider string) {

// SetDuration sets the time spent in seconds performing canary analysis
func (cr *Recorder) SetDuration(cd *flaggerv1.Canary, duration time.Duration) {
cr.duration.WithLabelValues(cd.Spec.TargetRef.Name, cd.Namespace).Observe(duration.Seconds())
cr.duration.WithLabelValues(cd.Name, cd.Spec.TargetRef.Name, cd.Namespace).Observe(duration.Seconds())
}

// SetTotal sets the total number of canaries per namespace
Expand All @@ -153,7 +156,7 @@ func (cr *Recorder) SetTotal(namespace string, total int) {
}

func (cr *Recorder) SetAnalysis(cd *flaggerv1.Canary, metricTemplateName string, val float64) {
cr.analysis.WithLabelValues(cd.Spec.TargetRef.Name, cd.Namespace, metricTemplateName).Set(val)
cr.analysis.WithLabelValues(cd.Name, cd.Spec.TargetRef.Name, cd.Namespace, metricTemplateName).Set(val)
}

// SetStatus sets the last known canary analysis status
Expand All @@ -167,13 +170,13 @@ func (cr *Recorder) SetStatus(cd *flaggerv1.Canary, phase flaggerv1.CanaryPhase)
default:
status = 1
}
cr.status.WithLabelValues(cd.Spec.TargetRef.Name, cd.Namespace).Set(float64(status))
cr.status.WithLabelValues(cd.Name, cd.Spec.TargetRef.Name, cd.Namespace).Set(float64(status))
}

// SetWeight sets the weight values for primary and canary destinations
func (cr *Recorder) SetWeight(cd *flaggerv1.Canary, primary int, canary int) {
cr.weight.WithLabelValues(fmt.Sprintf("%s-primary", cd.Spec.TargetRef.Name), cd.Namespace).Set(float64(primary))
cr.weight.WithLabelValues(cd.Spec.TargetRef.Name, cd.Namespace).Set(float64(canary))
cr.weight.WithLabelValues(cd.Name, fmt.Sprintf("%s-primary", cd.Spec.TargetRef.Name), cd.Namespace).Set(float64(primary))
cr.weight.WithLabelValues(cd.Name, cd.Spec.TargetRef.Name, cd.Namespace).Set(float64(canary))
}

// IncSuccesses increments the total number of canary successes
Expand All @@ -186,6 +189,25 @@ func (cr *Recorder) IncFailures(labels CanaryMetricLabels) {
cr.failures.WithLabelValues(labels.Values()...).Inc()
}

// DeleteFor removes all per-canary metric series matching the given
// name and namespace. Called when a Canary resource is deleted so that
// stale series do not linger in the Prometheus registry.
func (cr *Recorder) DeleteFor(name, namespace string) {
labels := prometheus.Labels{"canary": name, "namespace": namespace}
cr.duration.DeletePartialMatch(labels)
cr.status.DeletePartialMatch(labels)
cr.weight.DeletePartialMatch(labels)
cr.analysis.DeletePartialMatch(labels)
cr.successes.DeletePartialMatch(labels)
cr.failures.DeletePartialMatch(labels)
}

// DeleteTotalFor removes the per-namespace canary_total series. Called
// once a namespace no longer contains any Canary resources.
func (cr *Recorder) DeleteTotalFor(namespace string) {
cr.total.DeleteLabelValues(namespace)
}

// GetStatusMetric returns the status metric
func (cr *Recorder) GetStatusMetric() *prometheus.GaugeVec {
return cr.status
Expand Down
Loading