Merge branch 'main' into fix-rc-delete

Author: arybolovlev

docs/metrics.md

@@ -12,8 +12,8 @@ Starting with version `2.10.0`, the operator introduces HCP Terraform–specific
 
 | Metric name | Type | Description | Controller | Status |
 |-------------|------|-------------|------------|--------|
-| `hcp_tf_runs{run_status="<HCP Terraform Run Status>"}` | Gauge | Pending runs by statuses. | RunsCollector | Alpha |
-| `hcp_tf_runs_total` | Gauge | Total number of pending Runs by statuses. | RunsCollector | Alpha |
+| `hcp_tf_runs{run_status, agent_pool_id, agent_pool_name}` | Gauge | Pending runs by statuses. | RunsCollector | Alpha |
+| `hcp_tf_runs_total{agent_pool_id, agent_pool_name}` | Gauge | Total number of pending Runs. | RunsCollector | Alpha |
 
 ## Scraping Metrics
 

internal/controller/metrics.go

@@ -17,11 +17,11 @@ var (
 		},
 		// TODO:
 		// - Add a status label to indicate whether the metric is up or down
-		//   (for example, when an endpoint is unreachable or the CR is suspended).
-		// - Add agent_pool_name as label.
-		// - Add agent_pool_id as label.
+		//   (for example, when an endpoint is unreachable or the CR is suspended/paused).
 		[]string{
 			"run_status",
+			"agent_pool_id",
+			"agent_pool_name",
 		},
 	)
 	metricRunsTotal = prometheus.NewGaugeVec(
@@ -31,10 +31,11 @@ var (
 		},
 		// TODO:
 		// - Add a status label to indicate whether the metric is up or down
-		//   (for example, when an endpoint is unreachable or the CR is suspended).
-		// - Add agent_pool_name as label.
-		// - Add agent_pool_id as label.
-		[]string{},
+		//   (for example, when an endpoint is unreachable or the CR is suspended/paused).
+		[]string{
+			"agent_pool_id",
+			"agent_pool_name",
+		},
 	)
 	// TODO:
 	// - Add a metric to track associated Workspaces.

internal/controller/runscollector_controller.go

@@ -309,11 +309,18 @@ func (r *RunsCollectorReconciler) reconcileRuns(ctx context.Context, rc *runsCol
 	}
 
 	for _, status := range runStatuses {
-		metricRuns.WithLabelValues(string(status)).Set(float64(runs[status]))
+		metricRuns.WithLabelValues(
+			string(status),                    // run_status
+			rc.instance.Status.AgentPool.ID,   // agent_pool_id
+			rc.instance.Status.AgentPool.Name, // agent_pool_name
+		).Set(float64(runs[status]))
 	}
 
 	rc.log.Info("Reconcile Runs Collector", "msg", fmt.Sprintf("Total Runs: %.0f", runsTotal))
-	metricRunsTotal.WithLabelValues().Set(runsTotal)
+	metricRunsTotal.WithLabelValues(
+		rc.instance.Status.AgentPool.ID,   // agent_pool_id
+		rc.instance.Status.AgentPool.Name, // agent_pool_name
+	).Set(runsTotal)
 
 	rc.instance.Status.ObservedGeneration = rc.instance.Generation