Merge pull request #155 from gnufied/add-node-registrar-healthcheck

OCPBUGS-26924: Add healthcheck for node-driver-registrar container
openshift · Feb 7, 2024 · 57812c5 · 57812c5
2 parents f9760de + 005e6d4
commit 57812c5
Show file tree

Hide file tree

Showing 9 changed files with 75 additions and 0 deletions.
diff --git a/assets/common/sidecars/node_driver_registrar.yaml b/assets/common/sidecars/node_driver_registrar.yaml
@@ -12,6 +12,7 @@ spec:
           args:
             - --csi-address=/csi/csi.sock
             - --kubelet-registration-path=/var/lib/kubelet/plugins/${DRIVER_NAME}/csi.sock
+            - --http-endpoint=:${NODE_DRIVER_REGISTRAR_HEALTH_PORT}
             - --v=${LOG_LEVEL}
           # Empty env. instead of nil for json patch to append new env. vars there
           env: []
@@ -24,6 +25,18 @@ spec:
               mountPath: /csi
             - name: registration-dir
               mountPath: /registration
+          ports:
+            - containerPort: ${NODE_DRIVER_REGISTRAR_HEALTH_PORT}
+              name: rhealthz
+              protocol: TCP
+          livenessProbe:
+            httpGet:
+              path: /healthz
+              port: rhealthz
+            initialDelaySeconds: 10
+            timeoutSeconds: 3
+            periodSeconds: 10
+            failureThreshold: 5
           resources:
             requests:
               memory: 50Mi

diff --git a/assets/overlays/aws-ebs/generated/hypershift/node.yaml b/assets/overlays/aws-ebs/generated/hypershift/node.yaml
@@ -75,6 +75,7 @@ spec:
       - args:
         - --csi-address=/csi/csi.sock
         - --kubelet-registration-path=/var/lib/kubelet/plugins/ebs.csi.aws.com/csi.sock
+        - --http-endpoint=:10304
         - --v=${LOG_LEVEL}
         env: []
         image: ${NODE_DRIVER_REGISTRAR_IMAGE}
@@ -86,7 +87,19 @@ spec:
               - /bin/sh
               - -c
               - rm -rf /registration/ebs.csi.aws.com-reg.sock /csi/csi.sock
+        livenessProbe:
+          failureThreshold: 5
+          httpGet:
+            path: /healthz
+            port: rhealthz
+          initialDelaySeconds: 10
+          periodSeconds: 10
+          timeoutSeconds: 3
         name: csi-node-driver-registrar
+        ports:
+        - containerPort: 10304
+          name: rhealthz
+          protocol: TCP
         resources:
           requests:
             cpu: 10m

diff --git a/assets/overlays/aws-ebs/generated/standalone/node.yaml b/assets/overlays/aws-ebs/generated/standalone/node.yaml
@@ -75,6 +75,7 @@ spec:
       - args:
         - --csi-address=/csi/csi.sock
         - --kubelet-registration-path=/var/lib/kubelet/plugins/ebs.csi.aws.com/csi.sock
+        - --http-endpoint=:10304
         - --v=${LOG_LEVEL}
         env: []
         image: ${NODE_DRIVER_REGISTRAR_IMAGE}
@@ -86,7 +87,19 @@ spec:
               - /bin/sh
               - -c
               - rm -rf /registration/ebs.csi.aws.com-reg.sock /csi/csi.sock
+        livenessProbe:
+          failureThreshold: 5
+          httpGet:
+            path: /healthz
+            port: rhealthz
+          initialDelaySeconds: 10
+          periodSeconds: 10
+          timeoutSeconds: 3
         name: csi-node-driver-registrar
+        ports:
+        - containerPort: 10304
+          name: rhealthz
+          protocol: TCP
         resources:
           requests:
             cpu: 10m

diff --git a/assets/overlays/azure-disk/generated/hypershift/node.yaml b/assets/overlays/azure-disk/generated/hypershift/node.yaml
@@ -92,6 +92,7 @@ spec:
       - args:
         - --csi-address=/csi/csi.sock
         - --kubelet-registration-path=/var/lib/kubelet/plugins/disk.csi.azure.com/csi.sock
+        - --http-endpoint=:10304
         - --v=${LOG_LEVEL}
         env: []
         image: ${NODE_DRIVER_REGISTRAR_IMAGE}
@@ -103,7 +104,19 @@ spec:
               - /bin/sh
               - -c
               - rm -rf /registration/disk.csi.azure.com-reg.sock /csi/csi.sock
+        livenessProbe:
+          failureThreshold: 5
+          httpGet:
+            path: /healthz
+            port: rhealthz
+          initialDelaySeconds: 10
+          periodSeconds: 10
+          timeoutSeconds: 3
         name: csi-node-driver-registrar
+        ports:
+        - containerPort: 10304
+          name: rhealthz
+          protocol: TCP
         resources:
           requests:
             cpu: 10m

diff --git a/assets/overlays/azure-disk/generated/standalone/node.yaml b/assets/overlays/azure-disk/generated/standalone/node.yaml
@@ -92,6 +92,7 @@ spec:
       - args:
         - --csi-address=/csi/csi.sock
         - --kubelet-registration-path=/var/lib/kubelet/plugins/disk.csi.azure.com/csi.sock
+        - --http-endpoint=:10304
         - --v=${LOG_LEVEL}
         env: []
         image: ${NODE_DRIVER_REGISTRAR_IMAGE}
@@ -103,7 +104,19 @@ spec:
               - /bin/sh
               - -c
               - rm -rf /registration/disk.csi.azure.com-reg.sock /csi/csi.sock
+        livenessProbe:
+          failureThreshold: 5
+          httpGet:
+            path: /healthz
+            port: rhealthz
+          initialDelaySeconds: 10
+          periodSeconds: 10
+          timeoutSeconds: 3
         name: csi-node-driver-registrar
+        ports:
+        - containerPort: 10304
+          name: rhealthz
+          protocol: TCP
         resources:
           requests:
             cpu: 10m

diff --git a/pkg/driver/aws-ebs/aws_ebs.go b/pkg/driver/aws-ebs/aws_ebs.go
@@ -101,6 +101,8 @@ func GetAWSEBSGeneratorConfig() *generator.CSIDriverGeneratorConfig {
 		GuestConfig: &generator.GuestConfig{
 			DaemonSetTemplateAssetName: "overlays/aws-ebs/patches/node_add_driver.yaml",
 			LivenessProbePort:          10300,
+			// 10305 is used for healthcheck of efs-operator
+			NodeRegistrarHealthCheckPort: 10304,
 			Sidecars: []generator.SidecarConfig{
 				commongenerator.DefaultNodeDriverRegistrar,
 				commongenerator.DefaultLivenessProbe.WithExtraArguments(

diff --git a/pkg/driver/azure-disk/azure_disk.go b/pkg/driver/azure-disk/azure_disk.go
@@ -121,6 +121,8 @@ func GetAzureDiskGeneratorConfig() *generator.CSIDriverGeneratorConfig {
 		GuestConfig: &generator.GuestConfig{
 			DaemonSetTemplateAssetName: "overlays/azure-disk/patches/node_add_driver.yaml",
 			LivenessProbePort:          10300,
+			// 10303 port is taken by azurefile stuff and hence we must use 10304 here
+			NodeRegistrarHealthCheckPort: 10304,
 			Sidecars: []generator.SidecarConfig{
 				commongenerator.DefaultNodeDriverRegistrar,
 				commongenerator.DefaultLivenessProbe.WithExtraArguments(

diff --git a/pkg/generator/asset_generator.go b/pkg/generator/asset_generator.go
@@ -254,6 +254,10 @@ func (gen *AssetGenerator) generateDaemonSet() error {
 		extraReplacements = append(extraReplacements, "${LIVENESS_PROBE_PORT}", strconv.Itoa(int(cfg.LivenessProbePort)))
 	}
 
+	if cfg.NodeRegistrarHealthCheckPort > 0 {
+		extraReplacements = append(extraReplacements, "${NODE_DRIVER_REGISTRAR_HEALTH_PORT}", strconv.Itoa(int(cfg.NodeRegistrarHealthCheckPort)))
+	}
+
 	err = gen.applyAssetPatch(dsYAML, cfg.DaemonSetTemplateAssetName, extraReplacements)
 	if err != nil {
 		return err

diff --git a/pkg/generator/types.go b/pkg/generator/types.go
@@ -116,6 +116,8 @@ type GuestConfig struct {
 	// MetricsPorts               []MetricsPort
 	// Liveness probe TCP port number exposed by the driver itself, i.e. by DaemonSetTemplateAssetName.
 	LivenessProbePort uint16
+	// port where node-registrar will expose health check endpoint
+	NodeRegistrarHealthCheckPort uint16
 	// Configuration of all sidecars to inject into the guest DaemonSet.
 	Sidecars []SidecarConfig