Extend workflow job run metric (#167)

Signed-off-by: peterhalasz <peter.halasz@live.com>
cpanato · Feb 6, 2024 · 08d40ff · 08d40ff
1 parent 6129f95
commit 08d40ff
Show file tree

Hide file tree

Showing 4 changed files with 168 additions and 103 deletions.
diff --git a/internal/server/metrics.go b/internal/server/metrics.go
@@ -8,21 +8,21 @@ var (
 		Help:    "Time that a workflow job took to reach a given state.",
 		Buckets: prometheus.ExponentialBuckets(1, 1.4, 30),
 	},
-		[]string{"org", "repo", "state", "runner_group"},
+		[]string{"org", "repo", "state", "runner_group", "workflow_name", "job_name"},
 	)
 
 	workflowJobDurationCounter = prometheus.NewCounterVec(prometheus.CounterOpts{
 		Name: "workflow_job_duration_seconds_total",
 		Help: "The total duration of jobs.",
 	},
-		[]string{"org", "repo", "status", "conclusion", "runner_group"},
+		[]string{"org", "repo", "status", "conclusion", "runner_group", "workflow_name", "job_name"},
 	)
 
 	workflowJobStatusCounter = prometheus.NewCounterVec(prometheus.CounterOpts{
 		Name: "workflow_job_status_count",
 		Help: "Count of workflow job events.",
 	},
-		[]string{"org", "repo", "status", "conclusion", "runner_group"},
+		[]string{"org", "repo", "status", "conclusion", "runner_group", "workflow_name", "job_name"},
 	)
 
 	workflowRunHistogramVec = prometheus.NewHistogramVec(prometheus.HistogramOpts{
@@ -83,9 +83,10 @@ func init() {
 }
 
 type WorkflowObserver interface {
-	ObserveWorkflowJobDuration(org, repo, state, runnerGroup string, seconds float64)
-	CountWorkflowJobStatus(org, repo, status, conclusion, runnerGroup string)
-	CountWorkflowJobDuration(org, repo, status, conclusion, runnerGroup string, seconds float64)
+	ObserveWorkflowJobDuration(org, repo, state, runnerGroup, workflowName, jobName string, seconds float64)
+	CountWorkflowJobStatus(org, repo, status, conclusion, runnerGroup, workflowName, jobName string)
+	CountWorkflowJobDuration(org, repo, status, conclusion, runnerGroup, workflowName, jobName string, seconds float64)
+
 	ObserveWorkflowRunDuration(org, repo, workflow, conclusion string, seconds float64)
 	CountWorkflowRunStatus(org, repo, status, conclusion, workflow string)
 }
@@ -94,17 +95,17 @@ var _ WorkflowObserver = (*PrometheusObserver)(nil)
 
 type PrometheusObserver struct{}
 
-func (o *PrometheusObserver) ObserveWorkflowJobDuration(org, repo, state, runnerGroup string, seconds float64) {
-	workflowJobHistogramVec.WithLabelValues(org, repo, state, runnerGroup).
+func (o *PrometheusObserver) ObserveWorkflowJobDuration(org, repo, state, runnerGroup, workflowName, jobName string, seconds float64) {
+	workflowJobHistogramVec.WithLabelValues(org, repo, state, runnerGroup, workflowName, jobName).
 		Observe(seconds)
 }
 
-func (o *PrometheusObserver) CountWorkflowJobStatus(org, repo, status, conclusion, runnerGroup string) {
-	workflowJobStatusCounter.WithLabelValues(org, repo, status, conclusion, runnerGroup).Inc()
+func (o *PrometheusObserver) CountWorkflowJobStatus(org, repo, status, conclusion, runnerGroup, workflowName, jobName string) {
+	workflowJobStatusCounter.WithLabelValues(org, repo, status, conclusion, runnerGroup, workflowName, jobName).Inc()
 }
 
-func (o *PrometheusObserver) CountWorkflowJobDuration(org, repo, status, conclusion, runnerGroup string, seconds float64) {
-	workflowJobDurationCounter.WithLabelValues(org, repo, status, conclusion, runnerGroup).Add(seconds)
+func (o *PrometheusObserver) CountWorkflowJobDuration(org, repo, status, conclusion, runnerGroup, workflowName, jobName string, seconds float64) {
+	workflowJobDurationCounter.WithLabelValues(org, repo, status, conclusion, runnerGroup, workflowName, jobName).Add(seconds)
 }
 
 func (o *PrometheusObserver) ObserveWorkflowRunDuration(org, repo, workflowName, conclusion string, seconds float64) {

diff --git a/internal/server/server_test.go b/internal/server/server_test.go
@@ -92,6 +92,8 @@ func Test_Server_MetricsRouteAfterWorkflowJob(t *testing.T) {
 	jobStartedAt := time.Unix(1650308740, 0)
 	completedAt := jobStartedAt.Add(time.Duration(expectedDuration) * time.Second)
 	runnerGroupName := "runner-group"
+	workflowName := "Build and test"
+	jobName := "Test"
 
 	event := github.WorkflowJobEvent{
 		Action: github.String("completed"),
@@ -107,6 +109,8 @@ func Test_Server_MetricsRouteAfterWorkflowJob(t *testing.T) {
 			StartedAt:       &github.Timestamp{Time: jobStartedAt},
 			CompletedAt:     &github.Timestamp{Time: completedAt},
 			RunnerGroupName: &runnerGroupName,
+			WorkflowName:    &workflowName,
+			Name:            &jobName,
 		},
 	}
 	req := testWebhookRequest(t, "http://localhost:8001/webhook", "workflow_job", event)
@@ -125,6 +129,6 @@ func Test_Server_MetricsRouteAfterWorkflowJob(t *testing.T) {
 
 	payload, err := io.ReadAll(metricsRes.Body)
 	require.NoError(t, err)
-	assert.Contains(t, string(payload), `workflow_job_duration_seconds_bucket{org="someone",repo="some-repo",runner_group="runner-group",state="in_progress",le="10.541350399999995"} 1`)
-	assert.Contains(t, string(payload), `workflow_job_duration_seconds_total{conclusion="success",org="someone",repo="some-repo",runner_group="runner-group",status="completed"} 10`)
+	assert.Contains(t, string(payload), `workflow_job_duration_seconds_bucket{job_name="Test",org="someone",repo="some-repo",runner_group="runner-group",state="in_progress",workflow_name="Build and test",le="10.541350399999995"} 1`)
+	assert.Contains(t, string(payload), `workflow_job_duration_seconds_total{conclusion="success",job_name="Test",org="someone",repo="some-repo",runner_group="runner-group",status="completed",workflow_name="Build and test"} 10`)
 }
diff --git a/internal/server/workflow_metrics_exporter.go b/internal/server/workflow_metrics_exporter.go
@@ -74,7 +74,13 @@ func (c *WorkflowMetricsExporter) HandleGHWebHook(w http.ResponseWriter, r *http
 		return
 	case "workflow_job":
 		event := model.WorkflowJobEventFromJSON(io.NopCloser(bytes.NewBuffer(buf)))
-		_ = level.Info(c.Logger).Log("msg", "got workflow_job event", "org", event.GetRepo().GetOwner().GetLogin(), "repo", event.GetRepo().GetName(), "runId", event.GetWorkflowJob().GetRunID(), "action", event.GetAction())
+		_ = level.Info(c.Logger).Log("msg", "got workflow_job event",
+			"org", event.GetRepo().GetOwner().GetLogin(),
+			"repo", event.GetRepo().GetName(),
+			"runId", event.GetWorkflowJob().GetRunID(),
+			"action", event.GetAction(),
+			"workflow_name", event.GetWorkflowJob().GetWorkflowName(),
+			"job_name", event.GetWorkflowJob().GetName())
 		go c.CollectWorkflowJobEvent(event)
 	case "workflow_run":
 		event := model.WorkflowRunEventFromJSON(io.NopCloser(bytes.NewBuffer(buf)))
@@ -93,37 +99,40 @@ func (c *WorkflowMetricsExporter) HandleGHWebHook(w http.ResponseWriter, r *http
 func (c *WorkflowMetricsExporter) CollectWorkflowJobEvent(event *github.WorkflowJobEvent) {
 	repo := event.GetRepo().GetName()
 	org := event.GetRepo().GetOwner().GetLogin()
-	runnerGroup := event.WorkflowJob.GetRunnerGroupName()
-
 	action := event.GetAction()
-	conclusion := event.GetWorkflowJob().GetConclusion()
-	status := event.GetWorkflowJob().GetStatus()
+
+	workflowJob := event.GetWorkflowJob()
+	runnerGroup := workflowJob.GetRunnerGroupName()
+	conclusion := workflowJob.GetConclusion()
+	status := workflowJob.GetStatus()
+	workflowName := workflowJob.GetWorkflowName()
+	jobName := workflowJob.GetName()
 
 	switch action {
 	case "queued":
 		// Do nothing.
 	case "in_progress":
 
-		if len(event.WorkflowJob.Steps) == 0 {
+		if len(workflowJob.Steps) == 0 {
 			_ = level.Debug(c.Logger).Log("msg", "unable to calculate job duration of in_progress event as event has no steps")
 			break
 		}
 
-		firstStep := event.WorkflowJob.Steps[0]
-		queuedSeconds := firstStep.StartedAt.Time.Sub(event.WorkflowJob.StartedAt.Time).Seconds()
-		c.PrometheusObserver.ObserveWorkflowJobDuration(org, repo, "queued", runnerGroup, math.Max(0, queuedSeconds))
+		firstStep := workflowJob.Steps[0]
+		queuedSeconds := firstStep.StartedAt.Time.Sub(workflowJob.GetStartedAt().Time).Seconds()
+		c.PrometheusObserver.ObserveWorkflowJobDuration(org, repo, "queued", runnerGroup, workflowName, jobName, math.Max(0, queuedSeconds))
 	case "completed":
-		if event.WorkflowJob.StartedAt == nil || event.WorkflowJob.CompletedAt == nil {
+		if workflowJob.StartedAt == nil || workflowJob.CompletedAt == nil {
 			_ = level.Debug(c.Logger).Log("msg", "unable to calculate job duration of completed event steps are missing timestamps")
 			break
 		}
 
-		jobSeconds := math.Max(0, event.WorkflowJob.GetCompletedAt().Time.Sub(event.WorkflowJob.GetStartedAt().Time).Seconds())
-		c.PrometheusObserver.ObserveWorkflowJobDuration(org, repo, "in_progress", runnerGroup, jobSeconds)
-		c.PrometheusObserver.CountWorkflowJobDuration(org, repo, status, conclusion, runnerGroup, jobSeconds)
+		jobSeconds := math.Max(0, workflowJob.GetCompletedAt().Time.Sub(workflowJob.GetStartedAt().Time).Seconds())
+		c.PrometheusObserver.ObserveWorkflowJobDuration(org, repo, "in_progress", runnerGroup, workflowName, jobName, jobSeconds)
+		c.PrometheusObserver.CountWorkflowJobDuration(org, repo, status, conclusion, runnerGroup, workflowName, jobName, jobSeconds)
 	}
 
-	c.PrometheusObserver.CountWorkflowJobStatus(org, repo, status, conclusion, runnerGroup)
+	c.PrometheusObserver.CountWorkflowJobStatus(org, repo, status, conclusion, runnerGroup, workflowName, jobName)
 }
 
 func (c *WorkflowMetricsExporter) CollectWorkflowRunEvent(event *github.WorkflowRunEvent) {