Register metrics manager for node server and begin emitting metrics.

Author: hajiler

cmd/gce-pd-csi-driver/main.go

@@ -155,13 +155,22 @@ func handle() {
 	}
 
 	var metricsManager *metrics.MetricsManager = nil
-	if *runControllerService && *httpEndpoint != "" {
+	runServiceWithMetrics := *runControllerService || *runNodeService
+	if runServiceWithMetrics && *httpEndpoint != "" {
 		mm := metrics.NewMetricsManager()
 		mm.InitializeHttpHandler(*httpEndpoint, *metricsPath)
-		mm.RegisterPDCSIMetric()
 
-		if metrics.IsGKEComponentVersionAvailable() {
-			mm.EmitGKEComponentVersion()
+		switch {
+		case *runControllerService:
+			mm.RegisterPDCSIMetric()
+			if metrics.IsGKEComponentVersionAvailable() {
+				mm.EmitGKEComponentVersion()
+			}
+		case *runNodeService:
+			if err := mm.EmmitProcessStartTime(); err != nil {
+				klog.Errorf("Failed to emit process start time: %v", err.Error())
+			}
+			mm.RegisterMountMetric()
 		}
 		metricsManager = &mm
 	}
@@ -255,6 +264,7 @@ func handle() {
 			EnableDeviceInUseCheck: *enableDeviceInUseCheck,
 			DeviceInUseTimeout:     *deviceInUseTimeout,
 			EnableDataCache:        *enableDataCacheFlag,
+			MetricsManager:         metricsManager,
 		}
 		nodeServer = driver.NewNodeServer(gceDriver, mounter, deviceUtils, meta, statter, nsArgs)
 		if *maxConcurrentFormatAndMount > 0 {

pkg/gce-pd-csi-driver/gce-pd-driver.go

@@ -156,6 +156,7 @@ func NewNodeServer(gceDriver *GCEDriver, mounter *mount.SafeFormatAndMount, devi
 		enableDeviceInUseCheck: args.EnableDeviceInUseCheck,
 		deviceInUseErrors:      newDeviceErrMap(args.DeviceInUseTimeout),
 		EnableDataCache:        args.EnableDataCache,
+		metricsManager:         args.MetricsManager,
 	}
 }
 

pkg/gce-pd-csi-driver/node.go

@@ -36,6 +36,7 @@ import (
 	"sigs.k8s.io/gcp-compute-persistent-disk-csi-driver/pkg/common"
 	"sigs.k8s.io/gcp-compute-persistent-disk-csi-driver/pkg/deviceutils"
 	metadataservice "sigs.k8s.io/gcp-compute-persistent-disk-csi-driver/pkg/gce-cloud-provider/metadata"
+	"sigs.k8s.io/gcp-compute-persistent-disk-csi-driver/pkg/metrics"
 	mountmanager "sigs.k8s.io/gcp-compute-persistent-disk-csi-driver/pkg/mount-manager"
 	"sigs.k8s.io/gcp-compute-persistent-disk-csi-driver/pkg/resizefs"
 )
@@ -71,6 +72,8 @@ type GCENodeServer struct {
 	// Embed UnimplementedNodeServer to ensure the driver returns Unimplemented for any
 	// new RPC methods that might be introduced in future versions of the spec.
 	csi.UnimplementedNodeServer
+
+	metricsManager *metrics.MetricsManager
 }
 
 type NodeServerArgs struct {
@@ -81,6 +84,8 @@ type NodeServerArgs struct {
 	DeviceInUseTimeout time.Duration
 
 	EnableDataCache bool
+
+	MetricsManager *metrics.MetricsManager
 }
 
 var _ csi.NodeServer = &GCENodeServer{}
@@ -406,6 +411,10 @@ func (ns *GCENodeServer) NodeStageVolume(ctx context.Context, req *csi.NodeStage
 				return &csi.NodeStageVolumeResponse{}, nil
 			}
 		}
+
+		if ns.metricsManager != nil {
+			ns.metricsManager.RecordMountErrorMetric(err)
+		}
 		return nil, status.Error(codes.Internal,
 			fmt.Sprintf("Failed to format and mount device from (%q) to (%q) with fstype (%q) and options (%q): %v",
 				devicePath, stagingTargetPath, fstype, options, err.Error()))