Set roster before quiescing next node (#178)

aerospike · Jan 13, 2023 · 3f5e1f3 · 3f5e1f3
1 parent f8dbcda
commit 3f5e1f3
Show file tree

Hide file tree

Showing 4 changed files with 18 additions and 14 deletions.
diff --git a/controllers/aero_info_calls.go b/controllers/aero_info_calls.go
@@ -33,7 +33,7 @@ import (
 // The ignorablePods list should be a list of failed or pending pods that are going to be
 // deleted eventually and are safe to ignore in stability checks.
 func (r *SingleClusterReconciler) waitForMultipleNodesSafeStopReady(
-	pods []*corev1.Pod, ignorablePods []corev1.Pod,
+	pods []*corev1.Pod, ignorablePods []corev1.Pod, setRoster bool,
 ) reconcileResult {
 	if len(pods) == 0 {
 		return reconcileSuccess()
@@ -60,8 +60,16 @@ func (r *SingleClusterReconciler) waitForMultipleNodesSafeStopReady(
 		return res
 	}
 
-	if err := r.validateSCClusterState(policy, ignorablePods); err != nil {
-		return reconcileError(err)
+	if setRoster {
+		// Setup roster after migration.
+		if err = r.getAndSetRoster(policy, r.aeroCluster.Spec.RosterNodeBlockList, ignorablePods); err != nil {
+			r.Log.Error(err, "Failed to set roster for cluster")
+			return reconcileRequeueAfter(1)
+		}
+	} else {
+		if err := r.validateSCClusterState(policy, ignorablePods); err != nil {
+			return reconcileError(err)
+		}
 	}
 
 	if err := r.quiescePods(policy, allHostConns, pods); err != nil {

diff --git a/controllers/pod.go b/controllers/pod.go
@@ -157,7 +157,7 @@ func (r *SingleClusterReconciler) rollingRestartPods(
 
 	if len(activePods) != 0 {
 		r.Log.Info("Restart active pods", "pods", getPodNames(activePods))
-		if res := r.waitForMultipleNodesSafeStopReady(activePods, ignorablePods); !res.isSuccess {
+		if res := r.waitForMultipleNodesSafeStopReady(activePods, ignorablePods, false); !res.isSuccess {
 			return res
 		}
 		if res := r.restartPods(rackState, activePods, restartTypeMap); !res.isSuccess {
@@ -333,7 +333,7 @@ func (r *SingleClusterReconciler) safelyDeletePodsAndEnsureImageUpdated(
 
 	if len(activePods) != 0 {
 		r.Log.Info("Restart active pods with updated container image", "pods", getPodNames(activePods))
-		if res := r.waitForMultipleNodesSafeStopReady(activePods, ignorablePods); !res.isSuccess {
+		if res := r.waitForMultipleNodesSafeStopReady(activePods, ignorablePods, false); !res.isSuccess {
 			return res
 		}
 		if res := r.deletePodAndEnsureImageUpdated(rackState, activePods); !res.isSuccess {

diff --git a/controllers/rack.go b/controllers/rack.go
@@ -621,16 +621,10 @@ func (r *SingleClusterReconciler) scaleDownRack(
 
 	// Ignore safe stop check on pod not in running state.
 	if utils.IsPodRunningAndReady(pod) {
-		if res := r.waitForMultipleNodesSafeStopReady([]*corev1.Pod{pod}, ignorablePods); !res.isSuccess {
+		if res := r.waitForMultipleNodesSafeStopReady([]*corev1.Pod{pod}, ignorablePods, true); !res.isSuccess {
 			// The pod is running and is unsafe to terminate.
 			return found, res
 		}
-
-		// Setup roster after migration.
-		if err = r.getAndSetRoster(policy, r.aeroCluster.Spec.RosterNodeBlockList, ignorablePods); err != nil {
-			r.Log.Error(err, "Failed to set roster for cluster")
-			return found, reconcileRequeueAfter(1)
-		}
 	}
 
 	// set migrate-fill-delay to 0 across all nodes of cluster to scale down fast

diff --git a/controllers/reconciler.go b/controllers/reconciler.go
@@ -158,8 +158,10 @@ func (r *SingleClusterReconciler) Reconcile() (ctrl.Result, error) {
 	}
 
 	if asdbv1beta1.IsClusterSCEnabled(r.aeroCluster) {
-		if res := r.waitForClusterStability(policy, allHostConns); !res.isSuccess {
-			return res.result, res.err
+		if r.aeroCluster.Status.AerospikeConfig != nil {
+			if res := r.waitForClusterStability(policy, allHostConns); !res.isSuccess {
+				return res.result, res.err
+			}
 		}
 
 		// Setup roster