PMM-14404 - Enhance PBM backup metrics collection by including cluste…

exporter/pbm_collector.go

@@ -175,6 +175,12 @@ func (p *pbmCollector) pbmAgentMetrics(ctx context.Context, pbmClient *sdk.Clien
 }
 
 func (p *pbmCollector) pbmBackupsMetrics(ctx context.Context, pbmClient *sdk.Client, l *slog.Logger) []prometheus.Metric {
+	currentNode, err := util.MyRole(ctx, p.base.client)
+	if err != nil {
+		l.Error("failed to get current node info", "error", err.Error())
+		return nil
+	}
+
 	backupsList, err := pbmClient.GetAllBackups(ctx)
 	if err != nil {
 		l.Error("failed to get PBM backup list", "error", err.Error())
@@ -184,42 +190,48 @@ func (p *pbmCollector) pbmBackupsMetrics(ctx context.Context, pbmClient *sdk.Cli
 	metrics := make([]prometheus.Metric, 0, len(backupsList))
 
 	for _, backup := range backupsList {
-		metrics = append(metrics, createPBMMetric("backup_size_bytes",
-			"Size of PBM backup",
-			float64(backup.Size), map[string]string{
-				"opid":   backup.OPID,
-				"status": string(backup.Status),
-				"name":   backup.Name,
-			}),
-		)
-
-		// Add backup_last_transition_ts metric
-		metrics = append(metrics, createPBMMetric("backup_last_transition_ts",
-			"Last transition timestamp of PBM backup (seconds since epoch)",
-			float64(backup.LastTransitionTS), map[string]string{
-				"opid":   backup.OPID,
-				"status": string(backup.Status),
-				"name":   backup.Name,
-			}),
-		)
-
-		var endTime int64
-		switch pbmAgentStatus(backup.Status) {
-		case statusDone, statusCancelled, statusError, statusDown:
-			endTime = backup.LastTransitionTS
-		default:
-			endTime = time.Now().Unix()
-		}
+		// Iterate through replsets in the backup metadata
+		for _, replset := range backup.Replsets {
+			// Determine if this is the current node
+			self := "0"
+			if replset.Node == currentNode.Me {
+				self = "1"
+			}
 
-		duration := time.Unix(endTime-backup.StartTS, 0).Unix()
-		metrics = append(metrics, createPBMMetric("backup_duration_seconds",
-			"Duration of PBM backup",
-			float64(duration), map[string]string{
-				"opid":   backup.OPID,
-				"status": string(backup.Status),
-				"name":   backup.Name,
-			}),
-		)
+			labels := map[string]string{
+				"opid":        backup.OPID,
+				"status":      string(backup.Status),
+				"name":        backup.Name,
+				"host":        replset.Node,
+				"replica_set": replset.Name,
+				"self":        self,
+			}
+
+			metrics = append(metrics, createPBMMetric("backup_size_bytes",
+				"Size of PBM backup",
+				float64(backup.Size), labels),
+			)
+
+			// Add backup_last_transition_ts metric
+			metrics = append(metrics, createPBMMetric("backup_last_transition_ts",
+				"Last transition timestamp of PBM backup (seconds since epoch)",
+				float64(backup.LastTransitionTS), labels),
+			)
+
+			var endTime int64
+			switch pbmAgentStatus(backup.Status) {
+			case statusDone, statusCancelled, statusError, statusDown:
+				endTime = backup.LastTransitionTS
+			default:
+				endTime = time.Now().Unix()
+			}
+
+			duration := time.Unix(endTime-backup.StartTS, 0).Unix()
+			metrics = append(metrics, createPBMMetric("backup_duration_seconds",
+				"Duration of PBM backup",
+				float64(duration), labels),
+			)
+		}
 	}
 	return metrics
 }