[release-1.28] feat: add node metrics

charts/latest/azurefile-csi-driver/templates/csi-azurefile-node-windows-hostprocess.yaml

@@ -124,6 +124,7 @@ spec:
             - "--allow-empty-cloud-config={{ .Values.node.allowEmptyCloudConfig }}"
             - "--enable-get-volume-stats={{ .Values.feature.enableGetVolumeStats }}"
             - "--enable-windows-host-process=true"
+            - "--metrics-address=0.0.0.0:{{ .Values.node.metricsPort }}"
           env:
             - name: AZURE_CREDENTIAL_FILE
               valueFrom:

charts/latest/azurefile-csi-driver/templates/csi-azurefile-node-windows.yaml

@@ -137,6 +137,7 @@ spec:
             - "--allow-empty-cloud-config={{ .Values.node.allowEmptyCloudConfig }}"
             - "--enable-get-volume-stats={{ .Values.feature.enableGetVolumeStats }}"
             - "--allow-inline-volume-key-access-with-identity={{ .Values.node.allowInlineVolumeKeyAccessWithIdentity }}"
+            - "--metrics-address=0.0.0.0:{{ .Values.node.metricsPort }}"
           ports:
             - containerPort: {{ .Values.node.livenessProbe.healthPort }}
               name: healthz

charts/latest/azurefile-csi-driver/templates/csi-azurefile-node.yaml

@@ -135,6 +135,7 @@ spec:
             - "--enable-get-volume-stats={{ .Values.feature.enableGetVolumeStats }}"
             - "--mount-permissions={{ .Values.linux.mountPermissions }}"
             - "--allow-inline-volume-key-access-with-identity={{ .Values.node.allowInlineVolumeKeyAccessWithIdentity }}"
+            - "--metrics-address=0.0.0.0:{{ .Values.node.metricsPort }}"
           ports:
             - containerPort: {{ .Values.node.livenessProbe.healthPort }}
               name: healthz

charts/latest/azurefile-csi-driver/values.yaml

@@ -124,6 +124,7 @@ node:
   cloudConfigSecretNamespace: kube-system
   allowEmptyCloudConfig: true
   allowInlineVolumeKeyAccessWithIdentity: false
+  metricsPort: 29615
   livenessProbe:
     healthPort: 29613
   logLevel: 5

deploy/csi-azurefile-node-windows-hostprocess.yaml

@@ -87,6 +87,7 @@ spec:
             - --endpoint=$(CSI_ENDPOINT)
             - --nodeid=$(KUBE_NODE_NAME)
             - --enable-windows-host-process=true
+            - --metrics-address="0.0.0.0:29615"
           env:
             - name: AZURE_CREDENTIAL_FILE
               valueFrom:

deploy/csi-azurefile-node-windows.yaml

@@ -100,6 +100,7 @@ spec:
             - --v=5
             - --endpoint=$(CSI_ENDPOINT)
             - --nodeid=$(KUBE_NODE_NAME)
+            - --metrics-address="0.0.0.0:29615"
           ports:
             - containerPort: 29613
               name: healthz

deploy/csi-azurefile-node.yaml

@@ -91,6 +91,7 @@ spec:
             - "--v=5"
             - "--endpoint=$(CSI_ENDPOINT)"
             - "--nodeid=$(KUBE_NODE_NAME)"
+            - "--metrics-address=0.0.0.0:29615"
           ports:
             - containerPort: 29613
               name: healthz

pkg/azurefile/nodeserver.go

@@ -37,6 +37,7 @@ import (
 	"golang.org/x/net/context"
 
 	azcache "sigs.k8s.io/cloud-provider-azure/pkg/cache"
+	"sigs.k8s.io/cloud-provider-azure/pkg/metrics"
 )
 
 // NodePublishVolume mount the volume from staging to target path
@@ -158,6 +159,12 @@ func (d *Driver) NodeStageVolume(ctx context.Context, req *csi.NodeStageVolumeRe
 	volumeMountGroup := req.GetVolumeCapability().GetMount().GetVolumeMountGroup()
 	gidPresent := checkGidPresentInMountFlags(mountFlags)
 
+	mc := metrics.NewMetricContext(azureFileCSIDriverName, "node_stage_volume", d.cloud.ResourceGroup, "", d.Name)
+	isOperationSucceeded := false
+	defer func() {
+		mc.ObserveOperationWithResult(isOperationSucceeded, VolumeID, volumeID)
+	}()
+
 	_, accountName, accountKey, fileShareName, diskName, _, err := d.GetAccountInfo(ctx, volumeID, req.GetSecrets(), context)
 	if err != nil {
 		return nil, status.Error(codes.InvalidArgument, fmt.Sprintf("GetAccountInfo(%s) failed with error: %v", volumeID, err))
@@ -369,6 +376,8 @@ func (d *Driver) NodeStageVolume(ctx context.Context, req *csi.NodeStageVolumeRe
 			}
 		}
 	}
+
+	isOperationSucceeded = true
 	return &csi.NodeStageVolumeResponse{}, nil
 }
 
@@ -388,6 +397,12 @@ func (d *Driver) NodeUnstageVolume(ctx context.Context, req *csi.NodeUnstageVolu
 	}
 	defer d.volumeLocks.Release(volumeID)
 
+	mc := metrics.NewMetricContext(azureFileCSIDriverName, "node_unstage_volume", d.cloud.ResourceGroup, "", d.Name)
+	isOperationSucceeded := false
+	defer func() {
+		mc.ObserveOperationWithResult(isOperationSucceeded, VolumeID, volumeID)
+	}()
+
 	klog.V(2).Infof("NodeUnstageVolume: CleanupMountPoint volume %s on %s", volumeID, stagingTargetPath)
 	if err := CleanupMountPoint(d.mounter, stagingTargetPath, true /*extensiveMountPointCheck*/); err != nil {
 		return nil, status.Errorf(codes.Internal, "failed to unmount staging target %s: %v", stagingTargetPath, err)
@@ -400,6 +415,7 @@ func (d *Driver) NodeUnstageVolume(ctx context.Context, req *csi.NodeUnstageVolu
 	}
 	klog.V(2).Infof("NodeUnstageVolume: unmount volume %s on %s successfully", volumeID, stagingTargetPath)
 
+	isOperationSucceeded = true
 	return &csi.NodeUnstageVolumeResponse{}, nil
 }
 
@@ -472,6 +488,12 @@ func (d *Driver) NodeGetVolumeStats(ctx context.Context, req *csi.NodeGetVolumeS
 		klog.V(6).Infof("NodeGetVolumeStats: begin to get VolumeStats on volume %s path %s", req.VolumeId, req.VolumePath)
 	}
 
+	mc := metrics.NewMetricContext(azureFileCSIDriverName, "node_get_volume_stats", d.cloud.ResourceGroup, "", d.Name)
+	isOperationSucceeded := false
+	defer func() {
+		mc.ObserveOperationWithResult(isOperationSucceeded, VolumeID, req.VolumeId)
+	}()
+
 	resp, err := GetVolumeStats(req.VolumePath, d.enableWindowsHostProcess)
 	if err == nil && resp != nil {
 		if d.printVolumeStatsCallLogs {
@@ -485,6 +507,7 @@ func (d *Driver) NodeGetVolumeStats(ctx context.Context, req *csi.NodeGetVolumeS
 			d.volStatsCache.Set(newVolID, *resp)
 		}
 	}
+	isOperationSucceeded = true
 	return resp, err
 }