(bug) Fix Helm conflict race when chart ownership transfers between ClusterSummaries

When a new ClusterSummary (B) is created before an existing one (A) is deleted,
and both reference the same Helm chart, B may receive a NonRetriableError for chart
ownership conflict while A is still being withdrawn.

The existing recovery path handles the normal case correctly: when A is deleted, it
unregisters from the chart manager and requeues all conflicting ClusterSummaries so
they retry.

The race occurs when B's deploy worker is still running at that moment. IsInProgress(B)
returns true, the requeue is skipped, and nobody notifies B once A is gone and B has
become the rightful owner.

B's deploy worker eventually finishes and stores the NonRetriableError. When B's reconciler
reads it, the conflict is already resolved, A has been removed from the chart manager and B
now owns the chart. Without this fix, B is permanently marked FailedNonRetriable and
never retried, since NonRetriableError intentionally suppresses further reconciliation.

The fix detects this at the moment B processes the stale result: if all charts that were in
conflict can now be managed by B, the error is treated as resolved, status is reset to Provisioning,
and a requeue is returned. The next reconciliation cycle submits a fresh deploy.

Author: gianlucam76

controllers/clustersummary_deployer.go

@@ -30,6 +30,7 @@ import (
 	"sigs.k8s.io/controller-runtime/pkg/client"
 
 	configv1beta1 "github.com/projectsveltos/addon-controller/api/v1beta1"
+	"github.com/projectsveltos/addon-controller/controllers/chartmanager"
 	"github.com/projectsveltos/addon-controller/lib/clusterops"
 	"github.com/projectsveltos/addon-controller/pkg/scope"
 	libsveltosv1beta1 "github.com/projectsveltos/libsveltos/api/v1beta1"
@@ -182,9 +183,22 @@ func (r *ClusterSummaryReconciler) proceedDeployingFeature(ctx context.Context,
 
 		r.updateFeatureStatus(clusterSummaryScope, f.id, deployerStatus, currentHash, deployerError, logger)
 		if deployerError != nil {
-			shouldReturn, err := r.handleDeployerError(deployerError, clusterSummaryScope, f, currentHash, logger)
-			if shouldReturn {
-				return err
+			// Race: when a blocking profile is deleted while this CS's deploy was still in
+			// progress (IsInProgress=true prevented requeueClusterSummary from running), the
+			// conflict may already be gone by the time we process this result. If so, skip
+			// FailedNonRetriable and fall through to submit a new deploy immediately.
+			var nonRetriableError *configv1beta1.NonRetriableError
+			if f.id == libsveltosv1beta1.FeatureHelm && errors.As(deployerError, &nonRetriableError) &&
+				r.helmConflictResolved(ctx, clusterSummaryScope, logger) {
+
+				provisioning := libsveltosv1beta1.FeatureStatusProvisioning
+				r.updateFeatureStatus(clusterSummaryScope, f.id, &provisioning, currentHash, nil, logger)
+				return fmt.Errorf("helm conflict resolved, requeuing")
+			} else {
+				shouldReturn, err := r.handleDeployerError(deployerError, clusterSummaryScope, f, currentHash, logger)
+				if shouldReturn {
+					return err
+				}
 			}
 		}
 		if *deployerStatus == libsveltosv1beta1.FeatureStatusProvisioning {
@@ -873,6 +887,44 @@ func (r *ClusterSummaryReconciler) shouldRedeploy(ctx context.Context,
 	return true
 }
 
+// helmConflictResolved returns true when all Helm charts previously in HelmChartStatusConflict
+// can now be managed by this ClusterSummary according to the in-memory chart manager.
+// HelmReleaseSummaries contain instantiated release names, so chart-manager lookups are exact.
+func (r *ClusterSummaryReconciler) helmConflictResolved(ctx context.Context,
+	clusterSummaryScope *scope.ClusterSummaryScope, logger logr.Logger) bool {
+
+	cs := clusterSummaryScope.ClusterSummary
+
+	var conflicted []configv1beta1.HelmChartSummary
+	for i := range cs.Status.HelmReleaseSummaries {
+		if cs.Status.HelmReleaseSummaries[i].Status == configv1beta1.HelmChartStatusConflict {
+			conflicted = append(conflicted, cs.Status.HelmReleaseSummaries[i])
+		}
+	}
+	if len(conflicted) == 0 {
+		return false
+	}
+
+	chartMgr, err := chartmanager.GetChartManagerInstance(ctx, r.Client)
+	if err != nil {
+		logger.V(logs.LogInfo).Error(err, "failed to get chart manager instance")
+		return false
+	}
+
+	for i := range conflicted {
+		chart := &configv1beta1.HelmChart{
+			ReleaseNamespace: conflicted[i].ReleaseNamespace,
+			ReleaseName:      conflicted[i].ReleaseName,
+		}
+		if !chartMgr.CanManageChart(cs, chart) {
+			return false
+		}
+	}
+
+	logger.V(logs.LogDebug).Info("all previously conflicting helm charts can now be managed")
+	return true
+}
+
 // maxNumberOfConsecutiveFailureReached returns true if max number of consecutive failures has been reached.
 func (r *ClusterSummaryReconciler) maxNumberOfConsecutiveFailureReached(clusterSummaryScope *scope.ClusterSummaryScope, f feature,
 	logger logr.Logger) bool {