Fix dell-replication-controller-config reconcile reset issue (#760)

controllers/csm_controller.go

@@ -811,6 +811,7 @@ func (r *ContainerStorageModuleReconciler) SyncCSM(ctx context.Context, cr csmv1
 
 				node.Rbac.ClusterRole = *clusterRoleForNode
 			case csmv1.Replication:
+				// This function adds replication sidecar to driver pods.
 				log.Info("Injecting CSM Replication")
 				dp, err := modules.ReplicationInjectDeployment(controller.Deployment, cr, operatorConfig)
 				if err != nil {
@@ -887,9 +888,17 @@ func (r *ContainerStorageModuleReconciler) SyncCSM(ctx context.Context, cr csmv1
 		}
 
 		if replicationEnabled {
+			// This will also create the dell-replication-controller namespace.
 			if err = modules.ReplicationManagerController(ctx, false, operatorConfig, cr, cluster.ClusterCTRLClient); err != nil {
 				return fmt.Errorf("failed to deploy replication controller: %v", err)
 			}
+
+			// Create ConfigMap if it does not already exist.
+			// ConfigMap requires namespace to be created.
+			_, err = modules.CreateReplicationConfigmap(ctx, cr, operatorConfig, ctrlClient)
+			if err != nil {
+				return fmt.Errorf("injecting replication into replication configmap: %v", err)
+			}
 		}
 
 		// if Observability is enabled, create or update obs components: topology, metrics of PowerScale and PowerFlex
@@ -1256,6 +1265,10 @@ func (r *ContainerStorageModuleReconciler) removeDriver(ctx context.Context, ins
 			if err = modules.ReplicationManagerController(ctx, true, operatorConfig, instance, cluster.ClusterCTRLClient); err != nil {
 				return err
 			}
+			log.Infow("Deleting Replication configmap")
+			if err = modules.DeleteReplicationConfigmap(cluster.ClusterCTRLClient); err != nil {
+				return err
+			}
 		}
 
 		// remove module observability

operatorconfig/moduleconfig/replication/v1.10.0/controller.yaml

@@ -214,17 +214,6 @@ subjects:
     namespace: dell-replication-controller
 ---
 apiVersion: v1
-data:
-  config.yaml: |
-    clusterId: ""
-    targets: []
-    CSI_LOG_LEVEL: "<REPLICATION_CTRL_LOG_LEVEL>"
-kind: ConfigMap
-metadata:
-  name: dell-replication-controller-config
-  namespace: dell-replication-controller
----
-apiVersion: v1
 kind: Service
 metadata:
   labels:

operatorconfig/moduleconfig/replication/v1.10.0/dell-replication-controller-config.yaml

@@ -0,0 +1,10 @@
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: dell-replication-controller-config
+  namespace: dell-replication-controller
+data:
+  config.yaml: |
+    clusterId: ""
+    targets: []
+    CSI_LOG_LEVEL: "debug"

operatorconfig/moduleconfig/replication/v1.8.1/controller.yaml

@@ -214,17 +214,6 @@ subjects:
     namespace: dell-replication-controller
 ---
 apiVersion: v1
-data:
-  config.yaml: |
-    clusterId: ""
-    targets: []
-    CSI_LOG_LEVEL: "<REPLICATION_CTRL_LOG_LEVEL>"
-kind: ConfigMap
-metadata:
-  name: dell-replication-controller-config
-  namespace: dell-replication-controller
----
-apiVersion: v1
 kind: Service
 metadata:
   labels:

operatorconfig/moduleconfig/replication/v1.8.1/dell-replication-controller-config.yaml

@@ -0,0 +1,10 @@
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: dell-replication-controller-config
+  namespace: dell-replication-controller
+data:
+  config.yaml: |
+    clusterId: ""
+    targets: []
+    CSI_LOG_LEVEL: "debug"

operatorconfig/moduleconfig/replication/v1.9.0/controller.yaml

@@ -214,17 +214,6 @@ subjects:
     namespace: dell-replication-controller
 ---
 apiVersion: v1
-data:
-  config.yaml: |
-    clusterId: ""
-    targets: []
-    CSI_LOG_LEVEL: "<REPLICATION_CTRL_LOG_LEVEL>"
-kind: ConfigMap
-metadata:
-  name: dell-replication-controller-config
-  namespace: dell-replication-controller
----
-apiVersion: v1
 kind: Service
 metadata:
   labels:

operatorconfig/moduleconfig/replication/v1.9.0/dell-replication-controller-config.yaml

@@ -0,0 +1,10 @@
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: dell-replication-controller-config
+  namespace: dell-replication-controller
+data:
+  config.yaml: |
+    clusterId: ""
+    targets: []
+    CSI_LOG_LEVEL: "debug"

pkg/modules/replication.go

@@ -17,6 +17,11 @@ import (
 	"fmt"
 	"strings"
 
+	corev1 "k8s.io/api/core/v1"
+	k8serrors "k8s.io/apimachinery/pkg/api/errors"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	t1 "k8s.io/apimachinery/pkg/types"
+
 	csmv1 "github.com/dell/csm-operator/api/v1"
 
 	"github.com/dell/csm-operator/pkg/drivers"
@@ -428,3 +433,50 @@ func ReplicationManagerController(ctx context.Context, isDeleting bool, op utils
 
 	return nil
 }
+
+func CreateReplicationConfigmap(ctx context.Context, cr csmv1.ContainerStorageModule, op utils.OperatorConfig, ctrlClient client.Client) ([]crclient.Object, error) {
+	replica, err := getReplicaModule(cr)
+	if err != nil {
+		return nil, err
+	}
+
+	buf, err := readConfigFile(replica, cr, op, "dell-replication-controller-config.yaml")
+	if err != nil {
+		return nil, err
+	}
+
+	var cm corev1.ConfigMap
+	if err := yaml.Unmarshal(buf, &cm); err != nil {
+		return nil, err
+	}
+
+	// Check if the ConfigMap already exists
+	foundConfigMap := &corev1.ConfigMap{}
+
+	err = ctrlClient.Get(ctx, t1.NamespacedName{Name: cm.Name, Namespace: cm.Namespace}, foundConfigMap)
+	if err != nil && k8serrors.IsNotFound(err) {
+		// ConfigMap doesn't exist, create it
+		if err := ctrlClient.Create(ctx, &cm); err != nil {
+			return nil, err
+		}
+	}
+	return []crclient.Object{&cm}, nil
+}
+
+func DeleteReplicationConfigmap(ctrlClient client.Client) error {
+	configMap := &corev1.ConfigMap{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      "dell-replication-controller-config",
+			Namespace: "dell-replication-controller",
+		},
+	}
+
+	if err := ctrlClient.Delete(context.Background(), configMap); err != nil {
+		if k8serrors.IsNotFound(err) {
+			return nil
+		}
+		return err
+	}
+
+	return nil
+}

pkg/modules/replication_test.go

@@ -18,8 +18,12 @@ import (
 	utils "github.com/dell/csm-operator/pkg/utils"
 	"github.com/dell/csm-operator/tests/shared"
 	"github.com/stretchr/testify/assert"
+	corev1 "k8s.io/api/core/v1"
 	rbacv1 "k8s.io/api/rbac/v1"
+	k8serrors "k8s.io/apimachinery/pkg/api/errors"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	"k8s.io/apimachinery/pkg/runtime"
+	t1 "k8s.io/apimachinery/pkg/types"
 	applyv1 "k8s.io/client-go/applyconfigurations/apps/v1"
 	"k8s.io/client-go/kubernetes"
 	"k8s.io/client-go/kubernetes/fake"
@@ -406,3 +410,71 @@ func TestReplicationManagerController(t *testing.T) {
 		})
 	}
 }
+
+func TestReplicationConfigmap(t *testing.T) {
+	// Create a fake client to use in the test
+	scheme := runtime.NewScheme()
+	_ = corev1.AddToScheme(scheme)
+	fakeClient := ctrlClientFake.NewClientBuilder().WithScheme(scheme).Build()
+
+	// Create a test ContainerStorageModule
+	cr, err := getCustomResource("./testdata/cr_powerscale_replica.yaml")
+	if err != nil {
+		panic(err)
+	}
+
+	// Call the function we want to test
+	// we can't use test config, as it doesn't have versionvalues
+	realConfig := utils.OperatorConfig{
+		ConfigDirectory: "../../operatorconfig",
+	}
+	objs, err := CreateReplicationConfigmap(context.Background(), cr, realConfig, fakeClient)
+	// Check that the function returned the expected results
+	if err != nil {
+		t.Errorf("CreateReplicationConfigmap returned an unexpected error: %v", err)
+	}
+
+	if len(objs) != 1 {
+		t.Errorf("CreateReplicationConfigmap returned the wrong number of objects: %d", len(objs))
+	}
+
+	cm, ok := objs[0].(*corev1.ConfigMap)
+	if !ok {
+		t.Errorf("CreateReplicationConfigmap returned the wrong type of object: %T", objs[0])
+	}
+
+	if cm.Name != "dell-replication-controller-config" {
+		t.Errorf("CreateReplicationConfigmap returned the wrong ConfigMap name: %s", cm.Name)
+	}
+
+	if cm.Namespace != "dell-replication-controller" {
+		t.Errorf("CreateReplicationConfigmap returned the wrong ConfigMap namespace: %s", cm.Namespace)
+	}
+
+	// Check that the ConfigMap was created in the fake client
+	foundConfigMap := &corev1.ConfigMap{}
+	err = fakeClient.Get(context.Background(), t1.NamespacedName{Name: "dell-replication-controller-config", Namespace: "dell-replication-controller"}, foundConfigMap)
+	if err != nil {
+		t.Errorf("ConfigMap was not created in the fake client: %v", err)
+	}
+
+	// Now verify that the ConfigMap can be deleted properly
+	// Call the function we want to test
+	if err := DeleteReplicationConfigmap(fakeClient); err != nil {
+		t.Errorf("DeleteReplicationConfigmap returned an unexpected error: %v", err)
+	}
+
+	// Check that the ConfigMap was deleted from the fake client
+	configMap := &corev1.ConfigMap{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      "dell-replication-controller-config",
+			Namespace: "dell-replication-controller",
+		},
+	}
+	err = fakeClient.Get(context.Background(), t1.NamespacedName{Name: cm.Name, Namespace: cm.Namespace}, configMap)
+	if err == nil {
+		t.Errorf("ConfigMap was not deleted from the fake client")
+	} else if !k8serrors.IsNotFound(err) {
+		t.Errorf("ConfigMap was not deleted from the fake client: %v", err)
+	}
+}

pkg/modules/testdata/dell-replication-controller-config.yaml

@@ -0,0 +1,10 @@
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: dell-replication-controller-config
+  namespace: dell-replication-controller
+data:
+  config.yaml: |
+    clusterId: ""
+    targets: []
+    CSI_LOG_LEVEL: "debug"

pkg/utils/status.go

@@ -116,6 +116,7 @@ func getDaemonSetStatus(ctx context.Context, instance *csmv1.ContainerStorageMod
 	}
 
 	for _, cluster := range clusterClients {
+		totalRunning = 0
 		log.Infof("\ndaemonset status for cluster: %s", cluster.ClusterID)
 		msg += fmt.Sprintf("error message for %s \n", cluster.ClusterID)
 

tests/e2e/testfiles/minimal-testfiles/storage_csm_powermax_replica.yaml

@@ -8,6 +8,8 @@ spec:
     csiDriverType: "powermax"
     configVersion: v2.12.0
     forceRemoveDriver: true
+    common:
+      image: "quay.io/dell/container-storage-modules/csi-powermax:nightly"
   modules:
     - name: replication
       enabled: true