Fix Robin integration in E2E tests

controllers/redis_manager.go

@@ -420,13 +420,6 @@ func (r *RedkeyClusterReconciler) doSlowUpgradeUpgrading(ctx context.Context, re
 		return err
 	}
 
-	// Reset node
-	err = redkeyRobin.ClusterResetNode(currentPartition)
-	if err != nil {
-		r.logError(redkeyCluster.NamespacedName(), err, "Error from Robin forgeting the node", "node index", currentPartition)
-		return err
-	}
-
 	err = r.updateClusterSubStatus(ctx, redkeyCluster, redkeyv1.SubstatusRollingConfig, strconv.Itoa(currentPartition))
 	if err != nil {
 		r.logError(redkeyCluster.NamespacedName(), err, "Error updating substatus")
@@ -460,6 +453,20 @@ func (r *RedkeyClusterReconciler) doSlowUpgradeRollingUpdate(ctx context.Context
 		return err
 	}
 
+	// Get the current partition
+	currentPartition, err := strconv.Atoi(redkeyCluster.Status.Substatus.UpgradingPartition)
+	if err != nil {
+		r.logError(redkeyCluster.NamespacedName(), err, "Error getting Upgrading Partition from RedkeyCluster object")
+		return err
+	}
+
+	// Reset node
+	err = redkeyRobin.ClusterResetNode(currentPartition)
+	if err != nil {
+		r.logError(redkeyCluster.NamespacedName(), err, "Error from Robin forgeting the node", "node index", currentPartition)
+		return err
+	}
+
 	// Check all cluster nodes are ready from Robin.
 	clusterNodes, err := redkeyRobin.GetClusterNodes()
 	if err != nil {
@@ -482,23 +489,6 @@ func (r *RedkeyClusterReconciler) doSlowUpgradeRollingUpdate(ctx context.Context
 		return nil // Cluster not ready --> keep waiting
 	}
 
-	// Get the current partition and update Upgrading Partition in RedkeyCluster Status if starting iterating over partitions.
-	var currentPartition int
-	if redkeyCluster.Status.Substatus.UpgradingPartition == "" {
-		currentPartition = int(*(existingStatefulSet.Spec.Replicas)) - 1
-		err := r.updateClusterSubStatus(ctx, redkeyCluster, redkeyv1.SubstatusSlowUpgrading, strconv.Itoa(currentPartition))
-		if err != nil {
-			r.logError(redkeyCluster.NamespacedName(), err, "Error updating substatus")
-			return err
-		}
-	} else {
-		currentPartition, err = strconv.Atoi(redkeyCluster.Status.Substatus.UpgradingPartition)
-		if err != nil {
-			r.logError(redkeyCluster.NamespacedName(), err, "Error getting Upgrading Partition from RedkeyCluster object")
-			return err
-		}
-	}
-
 	// If first partition reached, we can move to the next step.
 	// Else step over to the next partition.
 	if currentPartition == 0 {
@@ -820,10 +810,18 @@ func (r *RedkeyClusterReconciler) doFastScaling(ctx context.Context, redkeyClust
 			return true, err
 		}
 
+		err = robin.SetAndPersistRobinStatus(ctx, r.Client, redkeyCluster, redkeyv1.RobinStatusScalingUp)
+		if err != nil {
+			r.logError(redkeyCluster.NamespacedName(), err, "Error updating/persisting Robin status", "status", redkeyv1.RobinStatusScalingUp)
+			return true, err
+		}
+
 		return true, nil
 	case redkeyv1.SubstatusEndingFastScaling:
 		// Rebuilding the cluster after recreating all node pods. Check if the cluster is ready to end the Fast scaling.
 		logger := r.getHelperLogger(redkeyCluster.NamespacedName())
+
+		r.logInfo(redkeyCluster.NamespacedName(), "Finishing fast scaling")
 		robin, err := robin.NewRobin(ctx, r.Client, redkeyCluster, logger)
 		if err != nil {
 			r.logError(redkeyCluster.NamespacedName(), err, "Error getting Robin to check its readiness")
@@ -999,10 +997,10 @@ func (r *RedkeyClusterReconciler) scaleDownCluster(ctx context.Context, redkeyCl
 	existingPrimaryNodes := len(clusterNodes.GetPrimaryNodes())
 	existingReplicaNodes := len(clusterNodes.GetReplicaNodes())
 	expectedPrimaryNodes := int(redkeyCluster.Spec.Primaries)
-	expectedRelicaNodes := int(redkeyCluster.Spec.Primaries * redkeyCluster.Spec.ReplicasPerPrimary)
-	if existingPrimaryNodes != expectedPrimaryNodes || existingReplicaNodes != expectedRelicaNodes {
+	expectedReplicaNodes := int(redkeyCluster.Spec.Primaries * redkeyCluster.Spec.ReplicasPerPrimary)
+	if existingPrimaryNodes != expectedPrimaryNodes || existingReplicaNodes != expectedReplicaNodes {
 		r.logInfo(redkeyCluster.NamespacedName(), "Waiting for Robin to scale the cluster", "expected primary nodes", expectedPrimaryNodes,
-			"existing primary nodes", existingPrimaryNodes, "expected replica nodes", expectedRelicaNodes,
+			"existing primary nodes", existingPrimaryNodes, "expected replica nodes", expectedReplicaNodes,
 			"existing replica nodes", existingReplicaNodes)
 		return true, nil // Requeue
 	}

controllers/redkeycluster_manager.go

@@ -89,6 +89,7 @@ func (r *RedkeyClusterReconciler) ReconcileClusterObject(ctx context.Context, re
 		err = r.clusterScaledToZeroPrimaries(ctx, redkeyCluster)
 		if err != nil {
 			r.logError(redkeyCluster.NamespacedName(), err, "Error managing cluster scaled to 0 primaries")
+			return ctrl.Result{RequeueAfter: time.Second * ErrorRequeueTimeout}, err
 		}
 
 		// Requeue to recheck, reconciliation ends here!

controllers/robin.go

@@ -83,7 +83,8 @@ func (r *RedkeyClusterReconciler) handleRobinConfig(ctx context.Context, req ctr
 		return nil // No changes detected, no more to do here
 	}
 
-	// Robin configmap changed: update configmap
+	// Robin config changed: update configmap
+	r.logInfo(req.NamespacedName, "Robin's configuration has changed, updating ConfigMap")
 
 	// Serialize robin configuration to YAML
 	configYAML, err := yaml.Marshal(declaredConfig)
@@ -139,6 +140,12 @@ func (r *RedkeyClusterReconciler) handleRobinDeployment(ctx context.Context, req
 		replicas := int32(1)
 		deployment.Spec.Replicas = &replicas
 		changed = true
+
+		// Update replicas in robin configuration
+		err = robin.PersistRobinReplicas(ctx, r.Client, redkeyCluster, int(redkeyCluster.Spec.Primaries), int(redkeyCluster.Spec.ReplicasPerPrimary))
+		if err != nil {
+			r.logError(redkeyCluster.NamespacedName(), err, "Error persisting Robin primaries/replicasPerPrimary")
+		}
 	}
 
 	if !changed {

controllers/status.go

@@ -58,7 +58,7 @@ func (r *RedkeyClusterReconciler) updateClusterStatus(ctx context.Context, redke
 		// Update Robin status
 		// Do not update if we are switching to Initializing status because Robin needs some
 		// time to be ready to accept API requests.
-		if redkeyCluster.Status.Status != redkeyv1.StatusInitializing {
+		if redkeyCluster.Status.Status != redkeyv1.StatusInitializing && redkeyCluster.Spec.Primaries > 0 {
 			logger := r.getHelperLogger(redkeyCluster.NamespacedName())
 			robin, err := robin.NewRobin(ctx, r.Client, redkeyCluster, logger)
 			if err != nil {

internal/robin/robin.go

@@ -386,7 +386,11 @@ func (r *Robin) GetClusterStatus() (string, error) {
 }
 
 func doSimpleGet(url string) ([]byte, error) {
-	resp, err := http.Get(url)
+	client := &http.Client{
+		Timeout: time.Second * 10,
+	}
+
+	resp, err := client.Get(url)
 	if err != nil {
 		return nil, err
 	}
@@ -400,7 +404,9 @@ func doSimpleGet(url string) ([]byte, error) {
 }
 
 func doPut(url string, payload []byte) ([]byte, error) {
-	client := &http.Client{}
+	client := &http.Client{
+		Timeout: time.Second * 20,
+	}
 
 	req, err := http.NewRequest(http.MethodPut, url, bytes.NewBuffer(payload))
 	req.Header.Set("Content-Type", "application/json")
@@ -425,8 +431,15 @@ func doPut(url string, payload []byte) ([]byte, error) {
 func CompareConfigurations(a, b *Configuration) bool {
 	a2 := new(Configuration)
 	*a2 = *a
+
+	// Normalize nil vs empty slice for RedisInfoKeys. This is needed because comparing a nil slice with an empty slice returns false.
+	if len(b.Redis.Metrics.RedisInfoKeys) == 0 {
+		a2.Redis.Metrics.RedisInfoKeys = []string{}
+	}
+
 	a2.Redis.Cluster.Status = b.Redis.Cluster.Status
 	a2.Redis.Cluster.Primaries = b.Redis.Cluster.Primaries
+	a2.Redis.Cluster.ReplicasPerPrimary = b.Redis.Cluster.ReplicasPerPrimary
 	return reflect.DeepEqual(a2, b)
 }
 

test/cmd/k6scripts/test-300k.js

@@ -3,7 +3,7 @@
 // SPDX-License-Identifier: Apache-2.0
 
 /*
-  Test that imitates a redis-cluster that was causing rebalancing errors:
+  Test that imitates a redkey-cluster that was causing rebalancing errors:
   - Value Size: 1Byte - 300KBytes
   - Timeout: 30s
   - Some deletes

test/cmd/manifests/deployment-test-local/deployment-operator-local.yaml

@@ -5,7 +5,7 @@
 apiVersion: v1
 kind: ServiceAccount
 metadata:
-  name: redis-operator-sa
+  name: redkey-operator-sa
 ---
 apiVersion: rbac.authorization.k8s.io/v1
 kind: Role
@@ -37,7 +37,7 @@ rules:
 apiVersion: rbac.authorization.k8s.io/v1
 kind: Role
 metadata:
-  name: redis-operator-role
+  name: redkey-operator-role
 rules:
 - apiGroups:
   - ""
@@ -140,7 +140,7 @@ kind: ClusterRole
 metadata:
   labels:
     rbac.authorization.k8s.io/aggregate-to-admin: "true"
-  name: redis-operator-role
+  name: redkey-operator-role
 rules:
 - apiGroups:
   - ""
@@ -263,19 +263,19 @@ roleRef:
   name: leader-election-role
 subjects:
 - kind: ServiceAccount
-  name: redis-operator-sa
+  name: redkey-operator-sa
 ---
 apiVersion: rbac.authorization.k8s.io/v1
 kind: RoleBinding
 metadata:
-  name: redis-operator-rolebinding
+  name: redkey-operator-rolebinding
 roleRef:
   apiGroup: rbac.authorization.k8s.io
   kind: Role
-  name: redis-operator-role
+  name: redkey-operator-role
 subjects:
 - kind: ServiceAccount
-  name: redis-operator-sa
+  name: redkey-operator-sa
 ---
 apiVersion: v1
 data:
@@ -291,19 +291,19 @@ data:
       resourceName: db95d8a6.inditex.com
 kind: ConfigMap
 metadata:
-  name: redis-operator-config
+  name: redkey-operator-config
 ---
 apiVersion: apps/v1
 kind: Deployment
 metadata:
   labels:
-    control-plane: redis-operator
-  name: redis-operator
+    control-plane: redkey-operator
+  name: redkey-operator
 spec:
   primaries: 1
   selector:
     matchLabels:
-      control-plane: redis-operator
+      control-plane: redkey-operator
   template:
     metadata:
       annotations:
@@ -312,7 +312,7 @@ spec:
         prometheus.io/scheme: http
         prometheus.io/scrape: "true"
       labels:
-        control-plane: redis-operator
+        control-plane: redkey-operator
         domain: DOMAIN
         environment: ENVIRONMENT
         layer: middleware-redkeyoperator
@@ -333,9 +333,9 @@ spec:
           valueFrom:
             fieldRef:
               fieldPath: metadata.namespace
-        image: redis-operator:0.1.0-test
+        image: redkey-operator:0.1.0-test
         imagePullPolicy: IfNotPresent
-        name: redis-operator
+        name: redkey-operator
         resources:
           limits:
             cpu: 300m
@@ -347,5 +347,5 @@ spec:
           allowPrivilegeEscalation: false
       securityContext:
         runAsNonRoot: true
-      serviceAccountName: redis-operator-sa
+      serviceAccountName: redkey-operator-sa
       terminationGracePeriodSeconds: 10

test/cmd/manifests/rdcl-test-local.yml

@@ -8,16 +8,16 @@ metadata:
   annotations:
     amiga-admin: ""
     paas-details: |
-      {"workloads":[{"kind":"StatefulSet", "name":"redis-cluster-test"}]}
+      {"workloads":[{"kind":"StatefulSet", "name":"redkey-cluster-test"}]}
     technology: redis_7.4.0
   labels:
-    app: redis-cluster-test
+    app: redkey-cluster-test
     slot: test
     technology: redis_7.4.0
     version: 7.4.0
-    redkey-cluster-name: redis-cluster-test
+    redkey-cluster-name: redkey-cluster-test
     redis.redkeycluster.operator/component: redis
-  name: redis-cluster-test
+  name: redkey-cluster-test
 spec:
   config: |
     save ""
@@ -28,7 +28,7 @@ spec:
   image: redis/redis-stack-server:7.4.0-v3
   labels:
     application: greta
-    deployment: redis-cluster-test
+    deployment: redkey-cluster-test
     domain: comdistrib
     environment: pre
     slot: test

test/cmd/manifests/rdcl-test-loki-local.yml

@@ -8,20 +8,20 @@ metadata:
   annotations:
     amiga-admin: ""
     paas-details: |
-      {"workloads":[{"kind":"StatefulSet", "name":"redis-cluster-test"}]}
+      {"workloads":[{"kind":"StatefulSet", "name":"redkey-cluster-test"}]}
     technology: redis_7.4.0
     nodes-extra-annotations: |
       {"log-format": "lineformat"}
     nodes-extra-labels: |
       {"log-scrape": "true"}
   labels:
-    app: redis-cluster-test
+    app: redkey-cluster-test
     slot: test
     technology: redis_7.4.0
     version: 7.4.0
-    redkey-cluster-name: redis-cluster-test
+    redkey-cluster-name: redkey-cluster-test
     redis.redkeycluster.operator/component: redis
-  name: redis-cluster-test
+  name: redkey-cluster-test
 spec:
   config: |
     save ""
@@ -32,7 +32,7 @@ spec:
   image: redis/redis-stack-server:7.4.0-v3
   labels:
     application: redis-app
-    deployment: redis-cluster-test
+    deployment: redkey-cluster-test
     slot: test
     technology: redis_7.4.0
     version: 7.4.0

test/cmd/manifests/rdcl-test-size-local.yml

@@ -8,17 +8,17 @@ metadata:
   annotations:
     amiga-admin: ""
     paas-details: |
-      {"workloads":[{"kind":"StatefulSet", "name":"redis-cluster-test"}]}
+      {"workloads":[{"kind":"StatefulSet", "name":"redkey-cluster-test"}]}
     technology: redis_7.4.0
   labels:
-    app: redis-cluster-test
+    app: redkey-cluster-test
     application: redis-app
     slot: test
     technology: redis_7.4.0
     version: 7.4.0
-    redkey-cluster-name: redis-cluster-test
+    redkey-cluster-name: redkey-cluster-test
     redis.redkeycluster.operator/component: redis
-  name: redis-cluster-test
+  name: redkey-cluster-test
 spec:
   config: |
     save ""
@@ -30,7 +30,7 @@ spec:
   image: redis/redis-stack-server:7.4.0-v3
   labels:
     application: redis-app
-    deployment: redis-cluster-test
+    deployment: redkey-cluster-test
     slot: test
     technology: redis_7.4.0
     version: 7.4.0