feat: Add metrics per specific step in the CI job (#20)

* feat: Add metrics per specific step in the CI job

In order to get more detailed overview of the CI runs, we want to add a new metric github_workflow_job_step_run_duration_seconds which is give the run stats for each job in each workflow.

This way we will be able to see the full picture of the runs and see what takes
the most of the time

* only track succeeded jobs

* only check client repo

Author: n-srg

pkg/actionsmetrics/event_reader.go

@@ -74,8 +74,10 @@ func (reader *EventReader) ProcessWorkflowJobEvent(ctx context.Context, event in
 	labels["job_name"] = *e.WorkflowJob.Name
 	keysAndValues = append(keysAndValues, "job_name", *e.WorkflowJob.Name)
 
+	var repoName = ""
 	if e.Repo != nil {
 		if n := e.Repo.Name; n != nil {
+			repoName = *n
 			labels["repository"] = *n
 			keysAndValues = append(keysAndValues, "repository", *n)
 		}
@@ -210,6 +212,21 @@ func (reader *EventReader) ProcessWorkflowJobEvent(ctx context.Context, event in
 				githubWorkflowJobRunDurationSeconds.With(extraLabel("job_conclusion", *e.WorkflowJob.Conclusion, labels)).Observe(*runTimeSeconds)
 			}
 		}
+
+		if *e.WorkflowJob.Conclusion == "success" && repoName == "client" {
+			for _, step := range e.WorkflowJob.Steps {
+				stepLabels := extraLabel("step_name", *step.Name, labels)
+				stepLabels["step_number"] = fmt.Sprint(step.Number)
+				stepLabels["step_conclusion"] = *step.Conclusion
+				stepLabels["step_status"] = *step.Status
+
+				stepDuration := step.CompletedAt.Sub(step.StartedAt.Time)
+
+				githubWorkflowJobStepDurationSeconds.With(stepLabels).Observe(stepDuration.Seconds())
+
+				log.Info("processed step in the repository", "repo", repoName, "step_name", *step.Name, "step_conclusion", *step.Conclusion)
+			}
+		}
 	}
 }
 

pkg/actionsmetrics/metrics.go

@@ -42,10 +42,12 @@ func init() {
 
 	githubWorkflowJobQueueDurationSeconds = initGithubWorkflowJobQueueDurationSeconds(queueBuckets)
 	githubWorkflowJobRunDurationSeconds = initGithubWorkflowJobRunDurationSeconds(runBuckets)
+	githubWorkflowJobStepDurationSeconds = initGithubWorkflowJobStepDurationSeconds(runBuckets)
 
 	metrics.Registry.MustRegister(
 		githubWorkflowJobQueueDurationSeconds,
 		githubWorkflowJobRunDurationSeconds,
+		githubWorkflowJobStepDurationSeconds,
 		githubWorkflowJobConclusionsTotal,
 		githubWorkflowJobsQueuedTotal,
 		githubWorkflowJobsStartedTotal,
@@ -145,10 +147,22 @@ func initGithubWorkflowJobRunDurationSeconds(buckets []float64) *prometheus.Hist
 	)
 }
 
+func initGithubWorkflowJobStepDurationSeconds(buckets []float64) *prometheus.HistogramVec {
+	return prometheus.NewHistogramVec(
+		prometheus.HistogramOpts{
+			Name:    "github_workflow_job_step_run_duration_seconds",
+			Help:    "Run times for the steps in workflow jobs in seconds",
+			Buckets: buckets,
+		},
+		metricLabels("step_name", "step_number", "step_conclusion", "step_status"),
+	)
+}
+
 var (
 	commonLabels                          = []string{"runs_on", "job_name", "organization", "repository", "owner", "workflow_name", "is_main_branch"}
 	githubWorkflowJobQueueDurationSeconds *prometheus.HistogramVec
 	githubWorkflowJobRunDurationSeconds   *prometheus.HistogramVec
+	githubWorkflowJobStepDurationSeconds  *prometheus.HistogramVec
 	githubWorkflowJobConclusionsTotal     = prometheus.NewCounterVec(
 		prometheus.CounterOpts{
 			Name: "github_workflow_job_conclusions_total",