chore: Add CSI-specific Prometheus metrics

deploy/example/metrics/README.md

@@ -7,8 +7,8 @@ kubectl apply -f https://raw.githubusercontent.com/kubernetes-sigs/azurefile-csi
 2. Get `EXTERNAL-IP` of service `csi-azurefile-controller`
 ```console
 $ kubectl get svc csi-azurefile-controller -n kube-system
-NAME                       TYPE        CLUSTER-IP   EXTERNAL-IP   PORT(S)     AGE
-csi-azurefile-controller   ClusterIP   10.0.184.0   20.39.21.132  29614/TCP   47m
+NAME                       TYPE           CLUSTER-IP   EXTERNAL-IP    PORT(S)           AGE
+csi-azurefile-controller   LoadBalancer   10.0.184.0   20.39.21.132   29614:30563/TCP   47m
 ```
 
 3. Run following command to get cloudprovider_azure operation metrics
@@ -17,8 +17,58 @@ ip=`kubectl get svc csi-azurefile-controller -n kube-system | grep file | awk '{
 curl http://$ip:29614/metrics | grep cloudprovider_azure | grep file | grep -e sum -e count
 ```
 
+4. Run following command to get CSI-specific operation metrics
+```console
+ip=`kubectl get svc csi-azurefile-controller -n kube-system | grep file | awk '{print $4}'`
+curl http://$ip:29614/metrics | grep azurefile_csi_driver_operation | grep -e sum -e count
+```
+
+
+## CSI Driver Metrics
+
+The Azure File CSI driver exposes the following custom metrics:
+
+### Controller Metrics (port 29614)
+
+| Metric | Type | Labels | Description |
+|--------|------|--------|-------------|
+| `azurefile_csi_driver_operation_duration_seconds` | Histogram | `operation`, `success` | Duration of CSI operations in seconds |
+| `azurefile_csi_driver_operation_duration_seconds_labeled` | Histogram | `operation`, `success`, `protocol`, `storage_account_type` | Duration of CSI operations with additional labels |
+| `azurefile_csi_driver_operations_total` | Counter | `operation`, `success` | Total number of CSI operations |
+
+**Label Values:**
+- `operation`: `controller_create_volume`, `controller_delete_volume`, `controller_create_snapshot`, `controller_delete_snapshot`, `controller_expand_volume`
+- `success`: `true`, `false`
+- `protocol`: `SMB`, `NFS`
+- `storage_account_type`: `Premium_LRS`, `Premium_ZRS`, `Standard_LRS`, `StandardV2_LRS`, `Standard_GRS`, `Standard_ZRS`, etc.
+
+### Node Metrics (port 29615)
+
+| Metric | Type | Labels | Description |
+|--------|------|--------|-------------|
+| `azurefile_csi_driver_operation_duration_seconds` | Histogram | `operation`, `success` | Duration of CSI operations in seconds |
+| `azurefile_csi_driver_operations_total` | Counter | `operation`, `success` | Total number of CSI operations |
+
+**Label Values:**
+- `operation`: `node_stage_volume`, `node_unstage_volume`, `node_publish_volume`, `node_unpublish_volume`
+- `success`: `true`, `false`
+
+### Azure Cloud Provider Metrics
+
+The CSI driver also exposes Azure cloud provider metrics from the underlying Azure SDK operations:
+
+| Metric | Type | Labels | Description |
+|--------|------|--------|-------------|
+| `cloudprovider_azure_api_request_duration_seconds` | Histogram | `request`, `resource_group`, `subscription_id`, `source`, `result` | Latency of Azure API calls |
+| `cloudprovider_azure_api_request_throttled_count` | Counter | `request`, `resource_group`, `subscription_id`, `source` | Number of throttled Azure API requests |
+| `cloudprovider_azure_api_request_errors` | Counter | `request`, `resource_group`, `subscription_id`, `source` | Number of errors in Azure API requests |
+
+These metrics help monitor Azure API performance, throttling, and error rates for file share operations.
+
 ## Get Prometheus metrics from CSI driver node pod
 
 ```console
-kubectl get --raw /api/v1/namespaces/kube-system/pods/csi-azurefile-node-hfgrn:29615/proxy/metrics
+kubectl get --raw /api/v1/namespaces/kube-system/pods/csi-azurefile-node-xxxxx:29615/proxy/metrics
 ```
+
+> **Note:** Replace `csi-azurefile-node-xxxxx` with an actual pod name from `kubectl get pods -n kube-system -l app=csi-azurefile-node`

pkg/azurefile/controllerserver.go

@@ -41,6 +41,7 @@ import (
 	"k8s.io/apimachinery/pkg/util/wait"
 	"k8s.io/klog/v2"
 	"k8s.io/utils/ptr"
+	csiMetrics "sigs.k8s.io/azurefile-csi-driver/pkg/metrics"
 	azcache "sigs.k8s.io/cloud-provider-azure/pkg/cache"
 	"sigs.k8s.io/cloud-provider-azure/pkg/metrics"
 	"sigs.k8s.io/cloud-provider-azure/pkg/provider/storage"
@@ -515,6 +516,15 @@ func (d *Driver) CreateVolume(ctx context.Context, req *csi.CreateVolumeRequest)
 			requestName = "controller_create_volume_from_volume"
 		}
 	}
+
+	csiMC := csiMetrics.NewCSIMetricContext(requestName)
+	isOperationSucceeded := false
+	defer func() {
+		csiMC.ObserveWithLabels(isOperationSucceeded,
+			"protocol", string(shareProtocol),
+			"storage_account_type", sku)
+	}()
+
 	if sourceID != "" {
 		_, srcAccountName, _, _, _, _, err = GetFileShareInfo(sourceID) //nolint:dogsled
 		if err != nil {
@@ -558,7 +568,6 @@ func (d *Driver) CreateVolume(ctx context.Context, req *csi.CreateVolumeRequest)
 	}
 
 	mc := metrics.NewMetricContext(azureFileCSIDriverName, requestName, d.cloud.ResourceGroup, subsID, d.Name)
-	isOperationSucceeded := false
 	defer func() {
 		mc.ObserveOperationWithResult(isOperationSucceeded, VolumeID, volumeID)
 	}()
@@ -781,7 +790,13 @@ func (d *Driver) CreateVolume(ctx context.Context, req *csi.CreateVolumeRequest)
 }
 
 // DeleteVolume delete an azure file
-func (d *Driver) DeleteVolume(ctx context.Context, req *csi.DeleteVolumeRequest) (*csi.DeleteVolumeResponse, error) {
+func (d *Driver) DeleteVolume(ctx context.Context, req *csi.DeleteVolumeRequest) (resp *csi.DeleteVolumeResponse, returnedErr error) {
+	requestName := "controller_delete_volume"
+	csiMC := csiMetrics.NewCSIMetricContext(requestName)
+	defer func() {
+		csiMC.Observe(returnedErr == nil)
+	}()
+
 	volumeID := req.GetVolumeId()
 	if len(volumeID) == 0 {
 		return nil, status.Error(codes.InvalidArgument, "Volume ID missing in request")
@@ -826,10 +841,9 @@ func (d *Driver) DeleteVolume(ctx context.Context, req *csi.DeleteVolumeRequest)
 		secret = createStorageAccountSecret(accountName, accountKey)
 	}
 
-	mc := metrics.NewMetricContext(azureFileCSIDriverName, "controller_delete_volume", resourceGroupName, subsID, d.Name)
-	isOperationSucceeded := false
+	mc := metrics.NewMetricContext(azureFileCSIDriverName, requestName, resourceGroupName, subsID, d.Name)
 	defer func() {
-		mc.ObserveOperationWithResult(isOperationSucceeded, VolumeID, volumeID)
+		mc.ObserveOperationWithResult(returnedErr == nil, VolumeID, volumeID)
 	}()
 
 	if err := d.DeleteFileShare(ctx, subsID, resourceGroupName, accountName, fileShareName, secret, useDataPlaneAPI); err != nil {
@@ -840,7 +854,6 @@ func (d *Driver) DeleteVolume(ctx context.Context, req *csi.DeleteVolumeRequest)
 		klog.Warningf("RemoveStorageAccountTag(%s) under rg(%s) account(%s) failed with %v", storage.SkipMatchingTag, resourceGroupName, accountName, err)
 	}
 
-	isOperationSucceeded = true
 	return &csi.DeleteVolumeResponse{}, nil
 }
 
@@ -936,7 +949,13 @@ func (d *Driver) ControllerUnpublishVolume(_ context.Context, _ *csi.ControllerU
 }
 
 // CreateSnapshot create a snapshot
-func (d *Driver) CreateSnapshot(ctx context.Context, req *csi.CreateSnapshotRequest) (*csi.CreateSnapshotResponse, error) {
+func (d *Driver) CreateSnapshot(ctx context.Context, req *csi.CreateSnapshotRequest) (resp *csi.CreateSnapshotResponse, returnedErr error) {
+	requestName := "controller_create_snapshot"
+	csiMC := csiMetrics.NewCSIMetricContext(requestName)
+	defer func() {
+		csiMC.Observe(returnedErr == nil)
+	}()
+
 	sourceVolumeID := req.GetSourceVolumeId()
 	snapshotName := req.Name
 	if len(snapshotName) == 0 {
@@ -974,10 +993,9 @@ func (d *Driver) CreateSnapshot(ctx context.Context, req *csi.CreateSnapshotRequ
 		useDataPlaneAPI = d.useDataPlaneAPI(ctx, sourceVolumeID, accountName)
 	}
 
-	mc := metrics.NewMetricContext(azureFileCSIDriverName, "controller_create_snapshot", rgName, subsID, d.Name)
-	isOperationSucceeded := false
+	mc := metrics.NewMetricContext(azureFileCSIDriverName, requestName, rgName, subsID, d.Name)
 	defer func() {
-		mc.ObserveOperationWithResult(isOperationSucceeded, SourceResourceID, sourceVolumeID, SnapshotName, snapshotName)
+		mc.ObserveOperationWithResult(returnedErr == nil, SourceResourceID, sourceVolumeID, SnapshotName, snapshotName)
 	}()
 
 	exists, itemSnapshot, itemSnapshotTime, itemSnapshotQuota, err := d.snapshotExists(ctx, sourceVolumeID, snapshotName, req.GetSecrets(), useDataPlaneAPI)
@@ -1071,7 +1089,7 @@ func (d *Driver) CreateSnapshot(ctx context.Context, req *csi.CreateSnapshotRequ
 		}
 	}
 
-	createResp := &csi.CreateSnapshotResponse{
+	resp = &csi.CreateSnapshotResponse{
 		Snapshot: &csi.Snapshot{
 			SizeBytes:      util.GiBToBytes(int64(itemSnapshotQuota)),
 			SnapshotId:     sourceVolumeID + "#" + itemSnapshot + "#" + subsID,
@@ -1082,12 +1100,18 @@ func (d *Driver) CreateSnapshot(ctx context.Context, req *csi.CreateSnapshotRequ
 		},
 	}
 
-	isOperationSucceeded = true
-	return createResp, nil
+	return resp, nil
 }
 
 // DeleteSnapshot delete a snapshot (todo)
 func (d *Driver) DeleteSnapshot(ctx context.Context, req *csi.DeleteSnapshotRequest) (*csi.DeleteSnapshotResponse, error) {
+	requestName := "controller_delete_snapshot"
+	csiMC := csiMetrics.NewCSIMetricContext(requestName)
+	isOperationSucceeded := false
+	defer func() {
+		csiMC.Observe(isOperationSucceeded)
+	}()
+
 	if len(req.SnapshotId) == 0 {
 		return nil, status.Error(codes.InvalidArgument, "Snapshot ID must be provided")
 	}
@@ -1109,8 +1133,7 @@ func (d *Driver) DeleteSnapshot(ctx context.Context, req *csi.DeleteSnapshotRequ
 		subsID = d.cloud.SubscriptionID
 	}
 
-	mc := metrics.NewMetricContext(azureFileCSIDriverName, "controller_delete_snapshot", rgName, subsID, d.Name)
-	isOperationSucceeded := false
+	mc := metrics.NewMetricContext(azureFileCSIDriverName, requestName, rgName, subsID, d.Name)
 	defer func() {
 		mc.ObserveOperationWithResult(isOperationSucceeded, SnapshotID, req.SnapshotId)
 	}()
@@ -1280,6 +1303,13 @@ func (d *Driver) execAzcopyCopy(srcPath, dstPath string, azcopyCopyOptions, auth
 
 // ControllerExpandVolume controller expand volume
 func (d *Driver) ControllerExpandVolume(ctx context.Context, req *csi.ControllerExpandVolumeRequest) (*csi.ControllerExpandVolumeResponse, error) {
+	requestName := "controller_expand_volume"
+	csiMC := csiMetrics.NewCSIMetricContext(requestName)
+	isOperationSucceeded := false
+	defer func() {
+		csiMC.Observe(isOperationSucceeded)
+	}()
+
 	volumeID := req.GetVolumeId()
 	if len(volumeID) == 0 {
 		return nil, status.Error(codes.InvalidArgument, "Volume ID missing in request")
@@ -1318,8 +1348,7 @@ func (d *Driver) ControllerExpandVolume(ctx context.Context, req *csi.Controller
 		}
 	}
 
-	mc := metrics.NewMetricContext(azureFileCSIDriverName, "controller_expand_volume", resourceGroupName, subsID, d.Name)
-	isOperationSucceeded := false
+	mc := metrics.NewMetricContext(azureFileCSIDriverName, requestName, resourceGroupName, subsID, d.Name)
 	defer func() {
 		mc.ObserveOperationWithResult(isOperationSucceeded, VolumeID, volumeID)
 	}()
@@ -1402,9 +1431,6 @@ func (d *Driver) snapshotExists(ctx context.Context, sourceVolumeID, snapshotNam
 
 		// List share snapshots.
 		listSnapshot := serviceURL.NewListSharesPager(&service.ListSharesOptions{Include: service.ListSharesInclude{Metadata: true, Snapshots: true}})
-		if err != nil {
-			return false, "", time.Time{}, 0, err
-		}
 		for listSnapshot.More() {
 			response, err := listSnapshot.NextPage(ctx)
 			if err != nil {

pkg/azurefile/nodeserver.go

@@ -39,6 +39,7 @@ import (
 	"golang.org/x/net/context"
 	"google.golang.org/grpc"
 	mount_azurefile "sigs.k8s.io/azurefile-csi-driver/pkg/azurefile-proxy/pb"
+	csiMetrics "sigs.k8s.io/azurefile-csi-driver/pkg/metrics"
 	volumehelper "sigs.k8s.io/azurefile-csi-driver/pkg/util"
 	azcache "sigs.k8s.io/cloud-provider-azure/pkg/cache"
 	"sigs.k8s.io/cloud-provider-azure/pkg/metrics"
@@ -58,7 +59,12 @@ func NewMountClient(cc *grpc.ClientConn) *MountClient {
 }
 
 // NodePublishVolume mount the volume from staging to target path
-func (d *Driver) NodePublishVolume(ctx context.Context, req *csi.NodePublishVolumeRequest) (*csi.NodePublishVolumeResponse, error) {
+func (d *Driver) NodePublishVolume(ctx context.Context, req *csi.NodePublishVolumeRequest) (resp *csi.NodePublishVolumeResponse, returnedErr error) {
+	csiMC := csiMetrics.NewCSIMetricContext("node_publish_volume")
+	defer func() {
+		csiMC.Observe(returnedErr == nil)
+	}()
+
 	volCap := req.GetVolumeCapability()
 	if volCap == nil {
 		return nil, status.Error(codes.InvalidArgument, "Volume capability missing in request")
@@ -197,7 +203,12 @@ func (d *Driver) NodePublishVolume(ctx context.Context, req *csi.NodePublishVolu
 }
 
 // NodeUnpublishVolume unmount the volume from the target path
-func (d *Driver) NodeUnpublishVolume(_ context.Context, req *csi.NodeUnpublishVolumeRequest) (*csi.NodeUnpublishVolumeResponse, error) {
+func (d *Driver) NodeUnpublishVolume(_ context.Context, req *csi.NodeUnpublishVolumeRequest) (resp *csi.NodeUnpublishVolumeResponse, returnedErr error) {
+	csiMC := csiMetrics.NewCSIMetricContext("node_unpublish_volume")
+	defer func() {
+		csiMC.Observe(returnedErr == nil)
+	}()
+
 	if len(req.GetVolumeId()) == 0 {
 		return nil, status.Error(codes.InvalidArgument, "Volume ID missing in request")
 	}
@@ -224,14 +235,19 @@ func (d *Driver) NodeUnpublishVolume(_ context.Context, req *csi.NodeUnpublishVo
 			return nil, status.Errorf(codes.Internal, "failed to direct volume remove mount info %s: %v", targetPath, err)
 		}
 	}
-
 	klog.V(2).Infof("NodeUnpublishVolume: unmount volume %s on %s successfully", volumeID, targetPath)
 
 	return &csi.NodeUnpublishVolumeResponse{}, nil
 }
 
 // NodeStageVolume mount the volume to a staging path
-func (d *Driver) NodeStageVolume(ctx context.Context, req *csi.NodeStageVolumeRequest) (*csi.NodeStageVolumeResponse, error) {
+func (d *Driver) NodeStageVolume(ctx context.Context, req *csi.NodeStageVolumeRequest) (resp *csi.NodeStageVolumeResponse, returnedErr error) {
+	requestName := "node_stage_volume"
+	csiMC := csiMetrics.NewCSIMetricContext(requestName)
+	defer func() {
+		csiMC.Observe(returnedErr == nil)
+	}()
+
 	if len(req.GetVolumeId()) == 0 {
 		return nil, status.Error(codes.InvalidArgument, "Volume ID missing in request")
 	}
@@ -261,10 +277,9 @@ func (d *Driver) NodeStageVolume(ctx context.Context, req *csi.NodeStageVolumeRe
 		klog.V(2).Infof("CSI volume is read-only, mounting with extra option ro")
 	}
 
-	mc := metrics.NewMetricContext(azureFileCSIDriverName, "node_stage_volume", d.cloud.ResourceGroup, "", d.Name)
-	isOperationSucceeded := false
+	mc := metrics.NewMetricContext(azureFileCSIDriverName, requestName, d.cloud.ResourceGroup, "", d.Name)
 	defer func() {
-		mc.ObserveOperationWithResult(isOperationSucceeded, VolumeID, volumeID)
+		mc.ObserveOperationWithResult(returnedErr == nil, VolumeID, volumeID)
 	}()
 
 	_, accountName, accountKey, fileShareName, diskName, _, err := d.GetAccountInfo(ctx, volumeID, req.GetSecrets(), context)
@@ -573,12 +588,18 @@ func (d *Driver) NodeStageVolume(ctx context.Context, req *csi.NodeStageVolumeRe
 		}
 	}
 
-	isOperationSucceeded = true
 	return &csi.NodeStageVolumeResponse{}, nil
 }
 
 // NodeUnstageVolume unmount the volume from the staging path
 func (d *Driver) NodeUnstageVolume(_ context.Context, req *csi.NodeUnstageVolumeRequest) (*csi.NodeUnstageVolumeResponse, error) {
+	requestName := "node_unstage_volume"
+	csiMC := csiMetrics.NewCSIMetricContext(requestName)
+	isOperationSucceeded := false
+	defer func() {
+		csiMC.Observe(isOperationSucceeded)
+	}()
+
 	volumeID := req.GetVolumeId()
 	if len(volumeID) == 0 {
 		return nil, status.Error(codes.InvalidArgument, "Volume ID missing in request")
@@ -594,8 +615,7 @@ func (d *Driver) NodeUnstageVolume(_ context.Context, req *csi.NodeUnstageVolume
 	}
 	defer d.volumeLocks.Release(lockKey)
 
-	mc := metrics.NewMetricContext(azureFileCSIDriverName, "node_unstage_volume", d.cloud.ResourceGroup, "", d.Name)
-	isOperationSucceeded := false
+	mc := metrics.NewMetricContext(azureFileCSIDriverName, requestName, d.cloud.ResourceGroup, "", d.Name)
 	defer func() {
 		mc.ObserveOperationWithResult(isOperationSucceeded, VolumeID, volumeID)
 	}()