Liveness (#5)

* New sdk adopted

* e2fsprogs added in apk install

* deviice used for mount logic

* detachvolume implemented with new api

* SDK added to project path to fix faulty parts

* AppVersion taken optionally from chart.yml

* Probe and liveness sidecar implemented

* liveness probe configured

* device check added

* Chart version updated

---------

Co-authored-by: syed.shafin <syed.shafin@bracits.com>

Author: shafinhasnat

Dockerfile

@@ -12,11 +12,14 @@ RUN CGO_ENABLED=0 GOOS=linux GOARCH=amd64 \
         -X main.version=${VERSION} \
         -X k8s.io/csi-hyperstack/pkg/driver.DriverName=hyperstack.csi.nexgencloud.com \
         -X k8s.io/csi-hyperstack/pkg/driver.DriverVersion=${VERSION}" \
-      -o /csi-hyperstack .
+      -o /csi-hyperstack
 
 FROM alpine:3.20 as runtime
 RUN apk add --no-cache --update e2fsprogs
-
+RUN wget "https://github.com/fullstorydev/grpcurl/releases/download/v1.7.0/grpcurl_1.7.0_linux_x86_64.tar.gz" \
+ && tar -xvf grpcurl_1.7.0_linux_x86_64.tar.gz -C /usr/local/bin grpcurl \
+ && chmod +x /usr/local/bin/grpcurl \
+ && rm -rf grpcurl_1.7.0_linux_x86_64.tar.gz
 COPY --from=build /csi-hyperstack .
 RUN chmod +x csi-hyperstack
 RUN mkdir -p /csi

charts/csi-hyperstack/Chart.yaml

@@ -3,5 +3,5 @@ name: csi-hyperstack
 description: |
   HELM chart for Hyperstack CSI drvier deployment
 type: application
-version: 0.0.5
+version: 0.0.3
 appVersion: v0.0.4

charts/csi-hyperstack/templates/daemonset-node.yaml

@@ -32,9 +32,32 @@ spec:
           args:
             - "start"
             - "--endpoint=unix:///csi/csi.sock"
-            - "--hyperstack-api-address={{ .Values.hyperstack.apiKey }}"
-            - "--hyperstack-api-key={{ .Values.hyperstack.apiAddress }}"
-            - "--service-node-enabled"
+            - "--hyperstack-api-address={{ .Values.hyperstack.apiAddress }}"
+            - "--hyperstack-api-key={{ .Values.hyperstack.apiKey }}"
+            - "--service-node-enabled=true"
+          livenessProbe:
+            exec:
+              command:
+                - /bin/sh
+                - -c
+                - |
+                  grpcurl -plaintext -unix /csi/csi.sock csi.v1.Identity.GetPluginInfo >/dev/null 2>&1 \
+                  && grpcurl -plaintext -unix /csi/csi.sock csi.v1.Node.NodeGetInfo >/dev/null 2>&1
+            # initialDelaySeconds: 20
+            # periodSeconds: 10
+          readinessProbe:
+            exec:
+              command:
+                - /bin/sh
+                - -c
+                - |
+                  grpcurl -plaintext -unix /csi/csi.sock csi.v1.Identity.GetPluginInfo >/dev/null 2>&1 \
+                  && grpcurl -plaintext -unix /csi/csi.sock csi.v1.Node.NodeGetInfo >/dev/null 2>&1
+            # initialDelaySeconds: 20
+            # periodSeconds: 10
+            # timeoutSeconds: 5
+            # failureThreshold: 3
+            # successThreshold: 1
           env:
             - name: DRIVER_NAME
               value: "hyperstack.csi.nexgencloud.com"
@@ -61,6 +84,12 @@ spec:
             - --v=5
             - --csi-address=/csi/csi.sock
             - --kubelet-registration-path=/var/lib/kubelet/plugins/hyperstack.csi.nexgencloud.com/csi.sock
+          livenessProbe:
+            exec:
+              command:
+              - /csi-node-driver-registrar
+              - --kubelet-registration-path=/var/lib/kubelet/plugins/hyperstack.csi.nexgencloud.com/csi.sock
+              - --mode=kubelet-registration-probe
           securityContext:
             privileged: true
           env:

charts/csi-hyperstack/templates/deployment-controller.yaml

@@ -25,12 +25,30 @@ spec:
           image: {{ .Values.components.csiHyperstack.image }}:{{ .Values.components.csiHyperstack.tag | default .Chart.AppVersion }}
           imagePullPolicy: {{ .Values.components.csiHyperstack.imagePullPolicy }}
           # command: ["csi-hyperstack"]
+          livenessProbe:
+            exec:
+              command:
+                - /bin/sh
+                - -c
+                - ls -l /csi/csi.sock
+            # initialDelaySeconds: 20
+            # periodSeconds: 10
+          # readinessProbe:
+          #   exec:
+          #     command:
+          #       - /bin/sh
+          #       - -c
+          #       - |
+          #         grpcurl -plaintext -unix /csi/csi.sock csi.v1.Identity.GetPluginInfo >/dev/null 2>&1 \
+          #         && grpcurl -plaintext -unix /csi/csi.sock csi.v1.Controller.ControllerGetCapabilities >/dev/null 2>&1
+            # initialDelaySeconds: 20
+            # periodSeconds: 10
           args:
             - "start"
             - "--endpoint=unix:///csi/csi.sock"
             - "--hyperstack-api-address={{ .Values.hyperstack.apiAddress }}"
             - "--hyperstack-api-key={{ .Values.hyperstack.apiKey }}"
-            - "--service-controller-enabled"
+            - "--service-controller-enabled=true"
           env:
             - name: CSI_ENDPOINT
               value: unix:///csi/csi.sock

pkg/driver/controllerserver.go

@@ -201,11 +201,14 @@ func (cs *controllerServer) ControllerPublishVolume(ctx context.Context, req *cs
 	klog.Infof("ControllerPublishVolume: GetVolume succeeded -\nStatus: %s\nName: %s\nID: %d\nSize:%d", *getVolume.Status, *getVolume.Name, *getVolume.Id, *getVolume.Size)
 	if *getVolume.Status == "in-use" { //Volume is already attached
 		klog.Infof("ControllerPublishVolume: Volume %s is already in use", *getVolume.Name)
-		return &csi.ControllerPublishVolumeResponse{
-			PublishContext: map[string]string{
-				volNameKeyFromControllerPublishVolume: *(*getVolume.Attachments)[0].Device,
-			},
-		}, nil
+		if len(*getVolume.Attachments) > 0 {
+			klog.Infof("ControllerPublishVolume: Volume %s is already attached to node %s", *getVolume.Name, *(*getVolume.Attachments)[0].InstanceId)
+			return &csi.ControllerPublishVolumeResponse{
+				PublishContext: map[string]string{
+					volNameKeyFromControllerPublishVolume: *(*getVolume.Attachments)[0].Device,
+				},
+			}, nil
+		}
 	}
 	if *getVolume.Status == "available" {
 		attachVolume, err := cloud.AttachVolumeToNode(ctx, vmId, volumeIDInt)

pkg/driver/nodeserver.go

@@ -36,6 +36,9 @@ func (ns *nodeServer) NodeStageVolume(ctx context.Context, req *csi.NodeStageVol
 	klog.Infof("\n==============NodeStageVolume: called================\n")
 	klog.Infof("NodeStageVolume: called with args %+v", protosanitizer.StripSecrets(*req))
 	devicename := req.PublishContext[volNameKeyFromControllerPublishVolume]
+	if devicename == "" {
+		return nil, status.Error(codes.InvalidArgument, "Device name not found in publish context. Please wait for volume to be attached.")
+	}
 	klog.Infof("NodeStageVolume: devicename from publish context: %s", devicename)
 	err := formateAndMakeFS(devicename, "ext4")
 	if err != nil {
@@ -127,17 +130,11 @@ func (ns *nodeServer) NodePublishVolume(ctx context.Context, req *csi.NodePublis
 	klog.Infof("==============NodePublishVolume: called================\n")
 	klog.Infof("NodePublishVolume: called with args %+v", protosanitizer.StripSecrets(*req))
 
-	// make sure the requried fields are set and not empty
-
 	options := []string{"bind"}
 	if req.Readonly {
 		options = append(options, "ro")
 	}
 
-	// get req.VolumeCaps and make sure that you handle request for block mode as well
-	// here we are just handling request for filesystem mode
-	// in case of block mode, the source is going to be the device dir where volume was attached form ControllerPubVolume RPC
-
 	fsType := "ext4"
 	if req.VolumeCapability.GetMount().FsType != "" {
 		fsType = req.VolumeCapability.GetMount().FsType
@@ -146,8 +143,6 @@ func (ns *nodeServer) NodePublishVolume(ctx context.Context, req *csi.NodePublis
 	source := req.StagingTargetPath
 	target := req.TargetPath
 
-	// we want to run mount -t fstype source target -o bind,ro
-
 	err := mountDevice(source, target, fsType, options)
 	if err != nil {
 		return nil, status.Error(codes.Internal, fmt.Sprintf("Error %s, mounting the volume from staging dir to target dir", err.Error()))