OCPBUGS-105207: Revert "OCPBUGS-89689: (karpenter) use completed release image for unpinned NodeClaims during CP upgrade"

karpenter-operator/controllers/karpenterignition/karpenterignition_controller.go

@@ -126,11 +126,8 @@ func (r *KarpenterIgnitionReconciler) Reconcile(ctx context.Context, req ctrl.Re
 		return ctrl.Result{}, fmt.Errorf("failed to get HostedCluster: %w", err)
 	}
 
-	releaseImage, version, requeue := currentClusterRelease(hostedCluster)
-	if requeue && openshiftEC2NodeClass.Spec.Version == "" {
-		log.Info("No version history available for unpinned NodeClass, requeueing")
-		return ctrl.Result{RequeueAfter: 5 * time.Second}, nil
-	}
+	releaseImage := hcp.Spec.ReleaseImage
+	version := currentClusterVersion(hostedCluster)
 
 	// When the user requests a version for the OpenshiftEC2NodeClass we perform further validation and lookup of the release image.
 	// We only detect skew and if the version is valid in this case, as under normal circumstances we can assume the release image
@@ -345,18 +342,14 @@ func (r *KarpenterIgnitionReconciler) createInMemoryNodePool(
 	}
 }
 
-// currentClusterRelease returns the release image and version of the most recently
-// completed update from the HostedCluster's version history, plus a requeue flag.
-// It searches history entries for the one with state=Completed and the most recent
-// CompletionTime. If no completed entries exist, it falls back to the first Partial
-// entry (the version currently being rolled out). If no history entries exist at all,
-// it signals the caller to requeue.
-// This ensures that during a control plane upgrade, unpinned Karpenter NodeClaims use the
-// completed version's release image rather than the desired version, preventing premature
-// drift detection and worker node replacement before the control plane is ready.
-func currentClusterRelease(hostedCluster *hyperv1.HostedCluster) (string, string, bool) {
+// currentClusterVersion returns the version of the most recently completed update from the
+// HostedCluster's version history. It searches history entries for the one with state=Completed
+// and the most recent CompletionTime. If no completed entries exist and there is exactly one
+// history entry, it falls back to the desired version. This handles the case where a cluster
+// is still rolling out its initial version.
+func currentClusterVersion(hostedCluster *hyperv1.HostedCluster) string {
 	if hostedCluster.Status.Version == nil {
-		return hostedCluster.Spec.Release.Image, "", false
+		return ""
 	}
 
 	var latest *configv1.UpdateHistory
@@ -369,33 +362,22 @@ func currentClusterRelease(hostedCluster *hyperv1.HostedCluster) (string, string
 			latest = entry
 			continue
 		}
-		switch {
-		case entry.CompletionTime == nil:
-			// keep latest; nothing to compare against
-		case latest.CompletionTime == nil:
-			latest = entry
-		case entry.CompletionTime.After(latest.CompletionTime.Time):
+		if entry.CompletionTime != nil && latest.CompletionTime != nil && entry.CompletionTime.After(latest.CompletionTime.Time) {
 			latest = entry
 		}
 	}
 
 	if latest != nil {
-		return latest.Image, latest.Version, false
+		return latest.Version
 	}
 
-	// No completed entries: fall back to the first Partial entry if available.
-	// CVO prepends newest entries, so the first Partial is the currently rolling-out version.
-	// This is safer than Spec.Release.Image which flips to the new (desired) version
-	// immediately on upgrade request.
-	for i := range hostedCluster.Status.Version.History {
-		entry := &hostedCluster.Status.Version.History[i]
-		if entry.State == configv1.PartialUpdate {
-			return entry.Image, entry.Version, false
-		}
+	// If there are no completed entries but exactly one history entry exists, the cluster
+	// is likely still rolling out its first version. Fall back to the desired version.
+	if len(hostedCluster.Status.Version.History) == 1 {
+		return hostedCluster.Status.Version.Desired.Version
 	}
 
-	// Empty history: signal caller to requeue — CVO hasn't reported any version yet.
-	return "", "", true
+	return hostedCluster.Status.Version.Desired.Version
 }
 
 // validateVersion checks whether the requested version is valid for the given HostedCluster.