Merge pull request kubernetes-sigs#1069 from andyzhangx/add-node-metrics

feat: add node metrics
cvvz · Oct 18, 2023 · 87a48b6 · 87a48b6
2 parents c84d27d + a74154b
commit 87a48b6
Show file tree

Hide file tree

Showing 6 changed files with 27 additions and 0 deletions.
diff --git a/charts/latest/blob-csi-driver-v0.0.0.tgz b/charts/latest/blob-csi-driver-v0.0.0.tgz
diff --git a/charts/latest/blob-csi-driver/templates/csi-blob-node.yaml b/charts/latest/blob-csi-driver/templates/csi-blob-node.yaml
@@ -171,6 +171,7 @@ spec:
             - "--mount-permissions={{ .Values.node.mountPermissions }}"
             - "--allow-inline-volume-key-access-with-idenitity={{ .Values.node.allowInlineVolumeKeyAccessWithIdentity }}"
             - "--enable-aznfs-mount={{ .Values.node.enableAznfsMount }}"
+            - "--metrics-address=0.0.0.0:{{ .Values.node.metricsPort }}"
           ports:
             - containerPort: {{ .Values.node.livenessProbe.healthPort }}
               name: healthz

diff --git a/charts/latest/blob-csi-driver/values.yaml b/charts/latest/blob-csi-driver/values.yaml
@@ -111,6 +111,7 @@ node:
   allowEmptyCloudConfig: true
   allowInlineVolumeKeyAccessWithIdentity: false
   maxUnavailable: 1
+  metricsPort: 29635
   livenessProbe:
     healthPort: 29633
   logLevel: 5

diff --git a/deploy/csi-blob-node.yaml b/deploy/csi-blob-node.yaml
@@ -127,6 +127,7 @@ spec:
             - "--blobfuse-proxy-endpoint=$(BLOBFUSE_PROXY_ENDPOINT)"
             - "--nodeid=$(KUBE_NODE_NAME)"
             - "--user-agent-suffix=OSS-kubectl"
+            - "--metrics-address=0.0.0.0:29635"
           ports:
             - containerPort: 29633
               name: healthz

diff --git a/pkg/blob/blob_test.go b/pkg/blob/blob_test.go
@@ -60,6 +60,7 @@ func NewFakeDriver() *Driver {
 	driver.Name = fakeDriverName
 	driver.Version = vendorVersion
 	driver.subnetLockMap = util.NewLockMap()
+	driver.cloud = &azure.Cloud{}
 	return driver
 }
 
@@ -92,6 +93,7 @@ func TestNewDriver(t *testing.T) {
 	fakedriver.accountSearchCache = driver.accountSearchCache
 	fakedriver.dataPlaneAPIVolCache = driver.dataPlaneAPIVolCache
 	fakedriver.volStatsCache = driver.volStatsCache
+	fakedriver.cloud = driver.cloud
 	assert.Equal(t, driver, fakedriver)
 }
 

diff --git a/pkg/blob/nodeserver.go b/pkg/blob/nodeserver.go
@@ -28,6 +28,7 @@ import (
 
 	volumehelper "sigs.k8s.io/blob-csi-driver/pkg/util"
 	azcache "sigs.k8s.io/cloud-provider-azure/pkg/cache"
+	"sigs.k8s.io/cloud-provider-azure/pkg/metrics"
 
 	"github.com/Azure/azure-sdk-for-go/storage"
 	"github.com/container-storage-interface/spec/lib/go/csi"
@@ -238,6 +239,12 @@ func (d *Driver) NodeStageVolume(ctx context.Context, req *csi.NodeStageVolumeRe
 	attrib := req.GetVolumeContext()
 	secrets := req.GetSecrets()
 
+	mc := metrics.NewMetricContext(blobCSIDriverName, "node_stage_volume", d.cloud.ResourceGroup, "", d.Name)
+	isOperationSucceeded := false
+	defer func() {
+		mc.ObserveOperationWithResult(isOperationSucceeded, VolumeID, volumeID)
+	}()
+
 	var serverAddress, storageEndpointSuffix, protocol, ephemeralVolMountOptions string
 	var ephemeralVol, isHnsEnabled bool
 
@@ -340,6 +347,7 @@ func (d *Driver) NodeStageVolume(ctx context.Context, req *csi.NodeStageVolumeRe
 			klog.V(2).Infof("skip chmod on targetPath(%s) since mountPermissions is set as 0", targetPath)
 		}
 
+		isOperationSucceeded = true
 		klog.V(2).Infof("volume(%s) mount %s on %s succeeded", volumeID, source, targetPath)
 		return &csi.NodeStageVolumeResponse{}, nil
 	}
@@ -442,13 +450,20 @@ func (d *Driver) NodeUnstageVolume(ctx context.Context, req *csi.NodeUnstageVolu
 	}
 	defer d.volumeLocks.Release(volumeID)
 
+	mc := metrics.NewMetricContext(blobCSIDriverName, "node_unstage_volume", d.cloud.ResourceGroup, "", d.Name)
+	isOperationSucceeded := false
+	defer func() {
+		mc.ObserveOperationWithResult(isOperationSucceeded, VolumeID, volumeID)
+	}()
+
 	klog.V(2).Infof("NodeUnstageVolume: volume %s unmounting on %s", volumeID, stagingTargetPath)
 	err := mount.CleanupMountPoint(stagingTargetPath, d.mounter, true /*extensiveMountPointCheck*/)
 	if err != nil {
 		return nil, status.Errorf(codes.Internal, "failed to unmount staging target %q: %v", stagingTargetPath, err)
 	}
 	klog.V(2).Infof("NodeUnstageVolume: volume %s unmount on %s successfully", volumeID, stagingTargetPath)
 
+	isOperationSucceeded = true
 	return &csi.NodeUnstageVolumeResponse{}, nil
 }
 
@@ -491,6 +506,12 @@ func (d *Driver) NodeGetVolumeStats(ctx context.Context, req *csi.NodeGetVolumeS
 		return &resp, nil
 	}
 
+	mc := metrics.NewMetricContext(blobCSIDriverName, "node_get_volume_stats", d.cloud.ResourceGroup, "", d.Name)
+	isOperationSucceeded := false
+	defer func() {
+		mc.ObserveOperationWithResult(isOperationSucceeded, VolumeID, req.VolumeId)
+	}()
+
 	if _, err := os.Lstat(req.VolumePath); err != nil {
 		if os.IsNotExist(err) {
 			return nil, status.Errorf(codes.NotFound, "path %s does not exist", req.VolumePath)
@@ -548,6 +569,7 @@ func (d *Driver) NodeGetVolumeStats(ctx context.Context, req *csi.NodeGetVolumeS
 		},
 	}
 
+	isOperationSucceeded = true
 	klog.V(6).Infof("NodeGetVolumeStats: volume stats for volume %s path %s is %v", req.VolumeId, req.VolumePath, resp)
 	// cache the volume stats per volume
 	d.volStatsCache.Set(req.VolumeId, *resp)