Add AsyncOperationInProgress to ExternalObservation

When AsyncOperationInProgress is true, the managed reconciler sets
Synced=False with reason ReconcilePending instead of ReconcileSuccess.
This prevents a false Synced=True signal during long-running async
operations (e.g. MSK cluster instance type changes).

Fixes #941

Signed-off-by: Aj Nye <aj.nye@appian.com>

Author: ajnye

apis/common/condition.go

@@ -63,6 +63,7 @@ const (
 	ReasonReconcileSuccess ConditionReason = "ReconcileSuccess"
 	ReasonReconcileError   ConditionReason = "ReconcileError"
 	ReasonReconcilePaused  ConditionReason = "ReconcilePaused"
+	ReasonReconcilePending ConditionReason = "ReconcilePending"
 )
 
 // See https://github.com/kubernetes/community/blob/master/contributors/devel/sig-architecture/api-conventions.md#typical-status-properties
@@ -312,3 +313,16 @@ func ReconcilePaused() Condition {
 		Reason:             ReasonReconcilePaused,
 	}
 }
+
+// ReconcilePending returns a condition indicating that reconciliation is
+// deferred pending an in-flight async operation. Unlike ReconcileError, this
+// does not trigger exponential backoff.
+func ReconcilePending(msg string) Condition {
+	return Condition{
+		Type:               TypeSynced,
+		Status:             corev1.ConditionFalse,
+		LastTransitionTime: metav1.Now(),
+		Reason:             ReasonReconcilePending,
+		Message:            msg,
+	}
+}

apis/common/v1/condition.go

@@ -60,6 +60,7 @@ const (
 	ReasonReconcileSuccess = common.ReasonReconcileSuccess
 	ReasonReconcileError   = common.ReasonReconcileError
 	ReasonReconcilePaused  = common.ReasonReconcilePaused
+	ReasonReconcilePending = common.ReasonReconcilePending
 )
 
 // See https://github.com/kubernetes/community/blob/master/contributors/devel/sig-architecture/api-conventions.md#typical-status-properties
@@ -134,3 +135,9 @@ func ReconcileError(err error) Condition {
 func ReconcilePaused() Condition {
 	return common.ReconcilePaused()
 }
+
+// ReconcilePending returns a condition indicating that reconciliation is
+// deferred pending an in-flight async operation.
+func ReconcilePending(msg string) Condition {
+	return common.ReconcilePending(msg)
+}

pkg/reconciler/managed/reconciler.go

@@ -526,6 +526,13 @@ type ExternalObservation struct {
 	// finding where the observed diverges from the desired state.
 	// The string should be a cmp.Diff that details the difference.
 	Diff string
+
+	// AsyncOperationInProgress indicates that an asynchronous operation
+	// (e.g. a long-running cloud API call) is currently in progress for
+	// this resource. When true, the managed reconciler will set
+	// Synced=False with reason ReconcilePending instead of
+	// ReconcileSuccess, and will not call Update().
+	AsyncOperationInProgress bool
 }
 
 // An ExternalCreation is the result of the creation of an external resource.
@@ -1436,8 +1443,14 @@ func (r *Reconciler) Reconcile(ctx context.Context, req reconcile.Request) (resu
 		// https://github.com/crossplane/crossplane/issues/289
 		reconcileAfter := r.pollIntervalHook(managed, r.pollInterval)
 		log.Debug("External resource is up to date", "requeue-after", time.Now().Add(reconcileAfter))
-		status.MarkConditions(xpv1.ReconcileSuccess())
-		r.metricRecorder.recordFirstTimeReady(managed)
+
+		if observation.AsyncOperationInProgress {
+			log.Debug("Async operation in progress, setting ReconcilePending")
+			status.MarkConditions(xpv1.ReconcilePending("Async operation in progress"))
+		} else {
+			status.MarkConditions(xpv1.ReconcileSuccess())
+			r.metricRecorder.recordFirstTimeReady(managed)
+		}
 
 		// record that we intentionally did not update the managed resource
 		// because no drift was detected. We call this so late in the reconcile