Merge branch 'main' into feature/resiliency_module

dell · May 22, 2023 · 0c8f926 · 0c8f926
2 parents c3b29e2 + e76c52d
commit 0c8f926
Show file tree

Hide file tree

Showing 34 changed files with 1,601 additions and 123 deletions.
diff --git a/bundle/manifests/dell-csm-operator.clusterserviceversion.yaml b/bundle/manifests/dell-csm-operator.clusterserviceversion.yaml
@@ -387,18 +387,14 @@ metadata:
           "apiVersion": "storage.dell.com/v1",
           "kind": "ContainerStorageModule",
           "metadata": {
-            "name": "test-unity",
-            "namespace": "test-unity"
+            "name": "unity",
+            "namespace": "unity"
           },
           "spec": {
             "driver": {
-              "authSecret": "test-unity-creds",
+              "authSecret": "unity-creds",
               "common": {
                 "envs": [
-                  {
-                    "name": "X_CSI_UNITY_NODENAME_PREFIX",
-                    "value": "csi-node"
-                  },
                   {
                     "name": "X_CSI_UNITY_ALLOW_MULTI_POD_ACCESS",
                     "value": "false"

diff --git a/config/samples/storage_v1_csm_unity.yaml b/config/samples/storage_v1_csm_unity.yaml
@@ -1,8 +1,8 @@
 apiVersion: storage.dell.com/v1
 kind: ContainerStorageModule
 metadata:
-  name: test-unity
-  namespace: test-unity
+  name: unity
+  namespace: unity
 spec:
   driver:
     csiDriverType: "unity"
@@ -15,8 +15,8 @@ spec:
     configVersion: v2.7.0
     # authSecret: This is the secret used to validate the default Unity secret used for installation
     # Allowed values: <metadataName specified in the Manifest>-creds
-    # For example: If the metadataName is set to test-unity, authSecret value should be set to test-unity-creds
-    authSecret: test-unity-creds
+    # For example: If the metadataName is set to unity, authSecret value should be set to unity-creds
+    authSecret: unity-creds
     # Controller count
     replicas: 2
     dnsPolicy: ClusterFirstWithHostNet
@@ -27,8 +27,6 @@ spec:
       image: "dellemc/csi-unity:v2.7.0"
       imagePullPolicy: IfNotPresent
       envs:
-        - name: X_CSI_UNITY_NODENAME_PREFIX
-          value: "csi-node"
           # X_CSI_UNITY_ALLOW_MULTI_POD_ACCESS - Flag to enable sharing of volumes across multiple pods within the same node in RWO access mode.
           # Allowed values: boolean
           # Default value: "false"

diff --git a/controllers/csm_controller.go b/controllers/csm_controller.go
@@ -16,6 +16,7 @@ import (
 	"context"
 	"encoding/json"
 	"fmt"
+	"strconv"
 	"sync/atomic"
 	"time"
 
@@ -357,18 +358,25 @@ func (r *ContainerStorageModuleReconciler) handleDeploymentUpdate(oldObj interfa
 
 	csm := new(csmv1.ContainerStorageModule)
 	err := r.Client.Get(ctx, namespacedName, csm)
+
 	if err != nil {
 		log.Error("deployment get csm", "error", err.Error())
 	}
 
 	newStatus := csm.GetCSMStatus()
+
+	// Updating controller status manually as controller runtime API is not updating csm object with latest data
+	// TODO: Can remove this once the controller runtime repo has a fix for updating the object passed
+	newStatus.ControllerStatus.Available = strconv.Itoa(int(available))
+	newStatus.ControllerStatus.Desired = strconv.Itoa(int(desired))
+	newStatus.ControllerStatus.Failed = strconv.Itoa(int(numberUnavailable))
+
 	err = utils.UpdateStatus(ctx, csm, r, newStatus)
 	if err != nil {
 		log.Debugw("deployment status ", "pods", err.Error())
 	} else {
 		r.EventRecorder.Eventf(csm, corev1.EventTypeNormal, csmv1.EventCompleted, "Driver deployment running OK")
 	}
-
 }
 
 func (r *ContainerStorageModuleReconciler) handlePodsUpdate(oldObj interface{}, obj interface{}) {

diff --git a/operatorconfig/driverconfig/unity/v2.5.0/node.yaml b/operatorconfig/driverconfig/unity/v2.5.0/node.yaml
@@ -101,8 +101,6 @@ spec:
                 fieldRef:
                   apiVersion: v1
                   fieldPath: spec.nodeName
-            - name: X_CSI_UNITY_NODENAME_PREFIX
-              value: <X_CSI_UNITY_NODENAME_PREFIX>
             - name: SSL_CERT_DIR
               value: /certs
             - name: X_CSI_UNITY_SYNC_NODEINFO_INTERVAL

diff --git a/operatorconfig/driverconfig/unity/v2.6.0/node.yaml b/operatorconfig/driverconfig/unity/v2.6.0/node.yaml
@@ -101,8 +101,6 @@ spec:
                 fieldRef:
                   apiVersion: v1
                   fieldPath: spec.nodeName
-            - name: X_CSI_UNITY_NODENAME_PREFIX
-              value: <X_CSI_UNITY_NODENAME_PREFIX>
             - name: SSL_CERT_DIR
               value: /certs
             - name: X_CSI_UNITY_SYNC_NODEINFO_INTERVAL

diff --git a/operatorconfig/driverconfig/unity/v2.7.0/node.yaml b/operatorconfig/driverconfig/unity/v2.7.0/node.yaml
@@ -101,8 +101,6 @@ spec:
                 fieldRef:
                   apiVersion: v1
                   fieldPath: spec.nodeName
-            - name: X_CSI_UNITY_NODENAME_PREFIX
-              value: <X_CSI_UNITY_NODENAME_PREFIX>
             - name: SSL_CERT_DIR
               value: /certs
             - name: X_CSI_UNITY_SYNC_NODEINFO_INTERVAL

diff --git a/pkg/drivers/common_test.go b/pkg/drivers/common_test.go
@@ -290,9 +290,6 @@ func csmWithUnity(driver csmv1.DriverType, version string, certProvided bool) cs
 	}
 	res.Spec.Driver.Common.Envs = []corev1.EnvVar{envVar1, envVar2, envVar3, envVar4, envVar5, envVar6, envVar7}
 
-	// Add node name prefix to cover some code in GetNode
-	// nodeNamePrefix := corev1.EnvVar{Name: "X_CSI_UNITY_NODENAME_PREFIX"}
-
 	// Add node fields specific to unity
 	healthMonitor := corev1.EnvVar{Name: "X_CSI_HEALTH_MONITOR_ENABLED", Value: "true"}
 	res.Spec.Driver.Node.Envs = []corev1.EnvVar{healthMonitor}

diff --git a/pkg/drivers/unity.go b/pkg/drivers/unity.go
@@ -36,9 +36,6 @@ const (
 	// UnityConfigParamsVolumeMount - Volume mount
 	UnityConfigParamsVolumeMount = "csi-unity-config-params"
 
-	// CsiUnityNodeNamePrefix - Node Name Prefix
-	CsiUnityNodeNamePrefix = "<X_CSI_UNITY_NODENAME_PREFIX>"
-
 	// CsiLogLevel - Defines the log level
 	CsiLogLevel = "<CSI_LOG_LEVEL>"
 
@@ -116,24 +113,17 @@ func PrecheckUnity(ctx context.Context, cr *csmv1.ContainerStorageModule, operat
 // ModifyUnityCR - Configuring CR parameters
 func ModifyUnityCR(yamlString string, cr csmv1.ContainerStorageModule, fileType string) string {
 	// Parameters to initialise CR values
-	nodePrefix := ""
 	healthMonitorNode := ""
 	healthMonitorController := ""
 
 	switch fileType {
 	case "Node":
-		for _, env := range cr.Spec.Driver.Common.Envs {
-			if env.Name == "X_CSI_UNITY_NODENAME_PREFIX" {
-				nodePrefix = env.Value
-			}
-		}
 		for _, env := range cr.Spec.Driver.Node.Envs {
 
 			if env.Name == "X_CSI_HEALTH_MONITOR_ENABLED" {
 				healthMonitorNode = env.Value
 			}
 		}
-		yamlString = strings.ReplaceAll(yamlString, CsiUnityNodeNamePrefix, nodePrefix)
 		yamlString = strings.ReplaceAll(yamlString, CsiHealthMonitorEnabled, healthMonitorNode)
 	case "Controller":
 		for _, env := range cr.Spec.Driver.Controller.Envs {

diff --git a/pkg/utils/status.go b/pkg/utils/status.go
@@ -128,6 +128,8 @@ func getDeploymentStatus(ctx context.Context, instance *csmv1.ContainerStorageMo
 		err = errors.New(msg)
 	}
 
+	log.Infof("Deployment totalReplicas count %d totalReadyPods %d totalFailedCount %d", totalReplicas, totalReadyPods, totalFailedCount)
+
 	return totalReplicas, csmv1.PodStatus{
 		Available: fmt.Sprintf("%d", totalReadyPods),
 		Desired:   fmt.Sprintf("%d", totalReplicas),
@@ -221,33 +223,46 @@ func getDaemonSetStatus(ctx context.Context, instance *csmv1.ContainerStorageMod
 func calculateState(ctx context.Context, instance *csmv1.ContainerStorageModule, r ReconcileCSM, newStatus *csmv1.ContainerStorageModuleStatus) (bool, error) {
 	log := logger.GetLogger(ctx)
 	running := false
-	controllerReplicas, controllerStatus, controllerErr := getDeploymentStatus(ctx, instance, r)
+	// TODO: Currently commented this block of code as the API used to get the latest deployment status is not working as expected
+	// TODO: Can be uncommented once this issues gets sorted out
+	/* controllerReplicas, controllerStatus, controllerErr := getDeploymentStatus(ctx, instance, r)
+	expected, nodeStatus, daemonSetErr := getDaemonSetStatus(ctx, instance, r)
 	newStatus.ControllerStatus = controllerStatus
+	newStatus.NodeStatus = nodeStatus */
 	expected, nodeStatus, daemonSetErr := getDaemonSetStatus(ctx, instance, r)
 	newStatus.NodeStatus = nodeStatus
+	controllerReplicas := newStatus.ControllerStatus.Desired
+	controllerStatus := newStatus.ControllerStatus
 
 	newStatus.State = constants.Failed
-	log.Infof("deployment controllerReplicas [%d]", controllerReplicas)
+	log.Infof("deployment controllerReplicas [%s]", controllerReplicas)
 	log.Infof("deployment controllerStatus.Available [%s]", controllerStatus.Available)
 
 	log.Infof("daemonset expected [%d]", expected)
 	log.Infof("daemonset nodeStatus.Available [%s]", nodeStatus.Available)
 
-	if (fmt.Sprintf("%d", controllerReplicas) == controllerStatus.Available) && (fmt.Sprintf("%d", expected) == nodeStatus.Available) {
+	if (controllerReplicas == controllerStatus.Available) && (fmt.Sprintf("%d", expected) == nodeStatus.Available) {
 		running = true
 		newStatus.State = constants.Succeeded
 	}
 	log.Infof("calculate overall state [%s]", newStatus.State)
-	var err error
-	if controllerErr != nil {
-		err = controllerErr
-	}
+	var err error = nil
+	// TODO: Uncomment this when the controller runtime API gets fixed
+	/*
+		if controllerErr != nil {
+			err = controllerErr
+		}
+		if daemonSetErr != nil {
+			err = daemonSetErr
+		}
+		if daemonSetErr != nil && controllerErr != nil {
+			err = fmt.Errorf("ControllerError: %s, Daemonseterror: %s", controllerErr.Error(), daemonSetErr.Error())
+			log.Infof("calculate overall error msg [%s]", err.Error())
+		} */
+
 	if daemonSetErr != nil {
 		err = daemonSetErr
-	}
-	if daemonSetErr != nil && controllerErr != nil {
-		err = fmt.Errorf("ControllerError: %s, Daemonseterror: %s", controllerErr.Error(), daemonSetErr.Error())
-		log.Infof("calculate overall error msg [%s]", err.Error())
+		log.Infof("calculate Daemonseterror msg [%s]", daemonSetErr.Error())
 	}
 	SetStatus(ctx, r, instance, newStatus)
 	return running, err
@@ -284,6 +299,12 @@ func UpdateStatus(ctx context.Context, instance *csmv1.ContainerStorageModule, r
 	if err != nil {
 		return err
 	}
+
+	log.Infow("instance - new controller Status", "desired", instance.Status.ControllerStatus.Desired)
+	log.Infow("instance - new controller Status", "Available", instance.Status.ControllerStatus.Available)
+	log.Infow("instance - new controller Status", "numberUnavailable", instance.Status.ControllerStatus.Failed)
+	log.Infow("instance - new controller Status", "State", instance.Status.State)
+
 	csm.Status = instance.Status
 	err = r.GetClient().Status().Update(ctx, csm)
 	if err != nil {

diff --git a/samples/storage_csm_unity_v250.yaml b/samples/storage_csm_unity_v250.yaml
@@ -1,8 +1,8 @@
 apiVersion: storage.dell.com/v1
 kind: ContainerStorageModule
 metadata:
-  name: test-unity
-  namespace: test-unity
+  name: unity
+  namespace: unity
 spec:
   driver:
     csiDriverType: "unity"
@@ -15,8 +15,8 @@ spec:
     configVersion: v2.5.0
     # authSecret: This is the secret used to validate the default Unity secret used for installation
     # Allowed values: <metadataName specified in the Manifest>-creds
-    # For example: If the metadataName is set to test-unity, authSecret value should be set to test-unity-creds
-    authSecret: test-unity-creds
+    # For example: If the metadataName is set to unity, authSecret value should be set to unity-creds
+    authSecret: unity-creds
     # Controller count
     replicas: 2
     dnsPolicy: ClusterFirstWithHostNet
@@ -27,8 +27,6 @@ spec:
       image: "dellemc/csi-unity:v2.5.0"
       imagePullPolicy: IfNotPresent
       envs:
-        - name: X_CSI_UNITY_NODENAME_PREFIX
-          value: "csi-node"
           # X_CSI_UNITY_ALLOW_MULTI_POD_ACCESS - Flag to enable sharing of volumes across multiple pods within the same node in RWO access mode.
           # Allowed values: boolean
           # Default value: "false"

diff --git a/samples/storage_csm_unity_v260.yaml b/samples/storage_csm_unity_v260.yaml
@@ -1,8 +1,8 @@
 apiVersion: storage.dell.com/v1
 kind: ContainerStorageModule
 metadata:
-  name: test-unity
-  namespace: test-unity
+  name: unity
+  namespace: unity
 spec:
   driver:
     csiDriverType: "unity"
@@ -15,8 +15,8 @@ spec:
     configVersion: v2.6.0
     # authSecret: This is the secret used to validate the default Unity secret used for installation
     # Allowed values: <metadataName specified in the Manifest>-creds
-    # For example: If the metadataName is set to test-unity, authSecret value should be set to test-unity-creds
-    authSecret: test-unity-creds
+    # For example: If the metadataName is set to unity, authSecret value should be set to unity-creds
+    authSecret: unity-creds
     # Controller count
     replicas: 2
     dnsPolicy: ClusterFirstWithHostNet
@@ -27,8 +27,6 @@ spec:
       image: "dellemc/csi-unity:v2.6.0"
       imagePullPolicy: IfNotPresent
       envs:
-        - name: X_CSI_UNITY_NODENAME_PREFIX
-          value: "csi-node"
           # X_CSI_UNITY_ALLOW_MULTI_POD_ACCESS - Flag to enable sharing of volumes across multiple pods within the same node in RWO access mode.
           # Allowed values: boolean
           # Default value: "false"

diff --git a/samples/storage_csm_unity_v270.yaml b/samples/storage_csm_unity_v270.yaml
@@ -1,8 +1,8 @@
 apiVersion: storage.dell.com/v1
 kind: ContainerStorageModule
 metadata:
-  name: test-unity
-  namespace: test-unity
+  name: unity
+  namespace: unity
 spec:
   driver:
     csiDriverType: "unity"
@@ -15,8 +15,8 @@ spec:
     configVersion: v2.7.0
     # authSecret: This is the secret used to validate the default Unity secret used for installation
     # Allowed values: <metadataName specified in the Manifest>-creds
-    # For example: If the metadataName is set to test-unity, authSecret value should be set to test-unity-creds
-    authSecret: test-unity-creds
+    # For example: If the metadataName is set to unity, authSecret value should be set to unity-creds
+    authSecret: unity-creds
     # Controller count
     replicas: 2
     dnsPolicy: ClusterFirstWithHostNet
@@ -27,8 +27,6 @@ spec:
       image: "dellemc/csi-unity:v2.7.0"
       imagePullPolicy: IfNotPresent
       envs:
-        - name: X_CSI_UNITY_NODENAME_PREFIX
-          value: "csi-node"
           # X_CSI_UNITY_ALLOW_MULTI_POD_ACCESS - Flag to enable sharing of volumes across multiple pods within the same node in RWO access mode.
           # Allowed values: boolean
           # Default value: "false"

diff --git a/tests/config/driverconfig/unity/v2.6.0/controller.yaml b/tests/config/driverconfig/unity/v2.6.0/controller.yaml
@@ -113,7 +113,7 @@ spec:
             topologyKey: "kubernetes.io/hostname"
       containers:
         - name: attacher
-          image: k8s.gcr.io/sig-storage/csi-attacher:v4.0.0
+          image: registry.k8s.io/sig-storage/csi-attacher:v4.0.0
           imagePullPolicy: IfNotPresent
           args:
             - "--csi-address=$(ADDRESS)"
@@ -126,7 +126,7 @@ spec:
             - name: socket-dir
               mountPath: /var/run/csi
         - name: provisioner
-          image: k8s.gcr.io/sig-storage/csi-provisioner:v3.3.0
+          image: registry.k8s.io/sig-storage/csi-provisioner:v3.3.0
           imagePullPolicy: IfNotPresent
           args:
             - "--csi-address=$(ADDRESS)"
@@ -147,7 +147,7 @@ spec:
             - name: socket-dir
               mountPath: /var/run/csi
         - name: snapshotter
-          image: k8s.gcr.io/sig-storage/csi-snapshotter:v6.1.0
+          image: registry.k8s.io/sig-storage/csi-snapshotter:v6.1.0
           imagePullPolicy: IfNotPresent
           args:
             - "--csi-address=$(ADDRESS)"
@@ -163,7 +163,7 @@ spec:
             - name: socket-dir
               mountPath: /var/run/csi
         - name: resizer
-          image: k8s.gcr.io/sig-storage/csi-resizer:v1.6.0
+          image: registry.k8s.io/sig-storage/csi-resizer:v1.6.0
           imagePullPolicy: IfNotPresent
           args:
             - "--csi-address=$(ADDRESS)"
@@ -176,7 +176,7 @@ spec:
             - name: socket-dir
               mountPath: /var/run/csi
         - name: external-health-monitor
-          image: k8s.gcr.io/sig-storage/csi-external-health-monitor-controller:v0.7.0
+          image: registry.k8s.io/sig-storage/csi-external-health-monitor-controller:v0.7.0
           imagePullPolicy: IfNotPresent
           args:
             - "--v=5"

diff --git a/tests/config/driverconfig/unity/v2.6.0/node.yaml b/tests/config/driverconfig/unity/v2.6.0/node.yaml
@@ -101,8 +101,6 @@ spec:
                 fieldRef:
                   apiVersion: v1
                   fieldPath: spec.nodeName
-            - name: X_CSI_UNITY_NODENAME_PREFIX
-              value: <X_CSI_UNITY_NODENAME_PREFIX>
             - name: SSL_CERT_DIR
               value: /certs
             - name: X_CSI_UNITY_SYNC_NODEINFO_INTERVAL
@@ -132,7 +130,7 @@ spec:
             - name: unity-secret
               mountPath: /unity-secret
         - name: registrar
-          image: k8s.gcr.io/sig-storage/csi-node-driver-registrar:v2.6.0
+          image: registry.k8s.io/sig-storage/csi-node-driver-registrar:v2.6.0
           args:
             - "--v=5"
             - "--csi-address=$(ADDRESS)"