Add waiting for dependent objects of ComputeDomain to be fully removed

Signed-off-by: Kevin Klues <[email protected]>
NVIDIA · Feb 9, 2025 · 0ea161c · 0ea161c
1 parent b8fa263
commit 0ea161c
Show file tree

Hide file tree

Showing 3 changed files with 40 additions and 0 deletions.
diff --git a/cmd/compute-domain-controller/computedomain.go b/cmd/compute-domain-controller/computedomain.go
@@ -231,10 +231,18 @@ func (m *ComputeDomainManager) onAddOrUpdate(ctx context.Context, obj any) error
 				return fmt.Errorf("error removing finalizer on ResourceClaimTemplate: %w", err)
 			}
 
+			if err := m.resourceClaimTemplateManager.AssertRemoved(ctx, string(cd.UID)); err != nil {
+				return fmt.Errorf("error asserting removal of ResourceClaimTemplate: %w", err)
+			}
+
 			if err := m.deploymentManager.RemoveFinalizer(ctx, string(cd.UID)); err != nil {
 				return fmt.Errorf("error removing finalizer on Deployment: %w", err)
 			}
 
+			if err := m.deploymentManager.AssertRemoved(ctx, string(cd.UID)); err != nil {
+				return fmt.Errorf("error asserting removal of Deployment: %w", err)
+			}
+
 			if err := m.RemoveFinalizer(ctx, string(cd.UID)); err != nil {
 				return fmt.Errorf("error removing finalizer: %w", err)
 			}

diff --git a/cmd/compute-domain-controller/deployment.go b/cmd/compute-domain-controller/deployment.go
@@ -255,6 +255,27 @@ func (m *DeploymentManager) RemoveFinalizer(ctx context.Context, cdUID string) e
 	return nil
 }
 
+func (m *DeploymentManager) AssertRemoved(ctx context.Context, cdUID string) error {
+	if err := m.resourceClaimTemplateManager.AssertRemoved(ctx, cdUID); err != nil {
+		return fmt.Errorf("error asserting ResourceClaimTemplate removed: %w", err)
+	}
+	if err := m.assertRemoved(ctx, cdUID); err != nil {
+		return fmt.Errorf("error asserting Deployment removal: %w", err)
+	}
+	return nil
+}
+
+func (m *DeploymentManager) assertRemoved(ctx context.Context, cdUID string) error {
+	ds, err := getByComputeDomainUID[*appsv1.Deployment](ctx, m.informer, cdUID)
+	if err != nil {
+		return fmt.Errorf("error retrieving Deployment: %w", err)
+	}
+	if len(ds) != 0 {
+		return fmt.Errorf("still exists")
+	}
+	return nil
+}
+
 func (m *DeploymentManager) removeFinalizer(ctx context.Context, cdUID string) error {
 	ds, err := getByComputeDomainUID[*appsv1.Deployment](ctx, m.informer, cdUID)
 	if err != nil {

diff --git a/cmd/compute-domain-controller/resourceclaimtemplate.go b/cmd/compute-domain-controller/resourceclaimtemplate.go
@@ -223,6 +223,17 @@ func (m *BaseResourceClaimTemplateManager) RemoveFinalizer(ctx context.Context,
 	return nil
 }
 
+func (m *BaseResourceClaimTemplateManager) AssertRemoved(ctx context.Context, cdUID string) error {
+	rcts, err := getByComputeDomainUID[*resourceapi.ResourceClaimTemplate](ctx, m.informer, cdUID)
+	if err != nil {
+		return fmt.Errorf("error retrieving ResourceClaimTemplate: %w", err)
+	}
+	if len(rcts) != 0 {
+		return fmt.Errorf("still exists")
+	}
+	return nil
+}
+
 func NewDeploymentResourceClaimTemplateManager(config *ManagerConfig) *DeploymentResourceClaimTemplateManager {
 	labelSelector := &metav1.LabelSelector{
 		MatchExpressions: []metav1.LabelSelectorRequirement{