update goldens.yaml

Author: DannyLiCom

goldens.yaml

@@ -1,6 +1,6 @@
 goldens:
   "NAP cluster-create with pathways":
-    command: xpk cluster create-pathways --project=golden-project --zone=us-central1-a --enable-autoprovisioning --cluster=golden-cluster --tpu-type=tpu7x-8 --on-demand --dry-run
+    command: xpk cluster create-pathways --project=golden-project --zone=us-central1-a --enable-autoprovisioning --cluster=golden-cluster --tpu-type=tpu7x-8 --on-demand --dry-run --managed-mldiagnostics
   "NAP cluster-create":
     command: xpk cluster create --project=golden-project --zone=us-central1-a --enable-autoprovisioning --cluster=golden-cluster --tpu-type=tpu7x-8 --on-demand --dry-run
   "Basic cluster create":

goldens/NAP_cluster-create_with_pathways.txt

@@ -219,17 +219,4 @@ kubectl apply -f a63aa3c4593c38ad90671fd8b067d1886f6313ad558379b364b51791aa50f4e
 kubectl apply -f 1d13ddebae3c90a05ba26b312df088982dd0df0edc4f4013b88384e476c20486
 [XPK] GKE commands done! Resources are created.
 [XPK] See your GKE Cluster here: https://console.cloud.google.com/kubernetes/clusters/details/us-central1/golden-cluster/details?project=golden-project
-Traceback (most recent call last):
-  File "/usr/local/google/home/lidanny/Desktop/Project/diagon_xpk/cienet_xpk/xpk/xpk.py", line 39, in <module>
-    main()
-    ~~~~^^
-  File "/usr/local/google/home/lidanny/Desktop/Project/diagon_xpk/cienet_xpk/xpk/src/xpk/main.py", line 77, in main
-    main_args.func(main_args)
-    ~~~~~~~~~~~~~~^^^^^^^^^^^
-  File "/usr/local/google/home/lidanny/Desktop/Project/diagon_xpk/cienet_xpk/xpk/src/xpk/commands/cluster.py", line 765, in cluster_create_pathways
-    cluster_create(args)
-    ~~~~~~~~~~~~~~^^^^^^
-  File "/usr/local/google/home/lidanny/Desktop/Project/diagon_xpk/cienet_xpk/xpk/src/xpk/commands/cluster.py", line 411, in cluster_create
-    if args.managed_mldiagnostics:
-       ^^^^^^^^^^^^^^^^^^^^^^^^^^
-AttributeError: 'Namespace' object has no attribute 'managed_mldiagnostics'
+[XPK] Exiting XPK cleanly

src/xpk/commands/cluster.py

@@ -1330,7 +1330,7 @@ def install_cert_manager(version: str = 'v1.13.0'):
   Returns:
     0 if successful and 1 otherwise.
   """
-  
+
   command = (
       f'kubectl apply -f https://github.com/cert-manager/cert-manager/releases/download/'
       f'{version}/cert-manager.yaml'
@@ -1358,19 +1358,19 @@ def download_mldiagnostics_yaml(package_name: str, version: str):
       f'--package={package_name} --version={version} --destination=./ '
       f'--project=ai-on-gke'
   )
-  
+
   return_code, return_output = run_command_for_value(
       command, f'Starting gcloud artifacts download for {package_name} {version}...'
   )
-  
+
   if return_code != 0:
     if 'already exists' in return_output:
       xpk_print(f'Artifact file for {package_name} {version} already exists locally. Skipping download.')
       return 0
     xpk_print(f'gcloud download returned with ERROR {return_code}.\n')
     xpk_exit(return_code)
 
-  xpk_print(f'Artifact download completed successfully.')
+  xpk_print('Artifact download completed successfully.')
   return return_code
 
 def create_mldiagnostics_namespace():
@@ -1380,21 +1380,21 @@ def create_mldiagnostics_namespace():
   Returns:
     0 if successful and 1 otherwise.
   """
-  
-  command = f'kubectl create namespace gke-diagon'
-  
+
+  command = 'kubectl create namespace gke-diagon'
+
   return_code, return_output = run_command_for_value(
-      command, f'Starting kubectl create namespace...'
+      command, 'Starting kubectl create namespace...'
   )
-  
+
   if return_code != 0:
     if 'already exists' in return_output:
       xpk_print('Namespace already exists locally. Skipping creation.')
       return 0
     xpk_print(f'Namespace creation returned with ERROR {return_code}.\n')
     xpk_exit(return_code)
 
-  xpk_print(f'gke-diagon Namespace created or already exists.')
+  xpk_print('gke-diagon Namespace created or already exists.')
   return return_code
 
 def install_mldiagnostics_yaml(artifact_filename: str):
@@ -1404,17 +1404,17 @@ def install_mldiagnostics_yaml(artifact_filename: str):
   Returns:
     0 if successful and 1 otherwise.
   """
-  
+
   command = f'kubectl apply -f {artifact_filename} -n gke-diagon'
-  
+
   return_code = run_command_with_updates(
       command, f'Starting kubectl apply -f {artifact_filename} -n gke-diagon...'
   )
-  
+
   if return_code != 0:
     xpk_print(f'kubectl apply returned with ERROR {return_code}.\n')
     xpk_exit(return_code)
-  
+
   xpk_print(f'{artifact_filename} applied successfully.')
 
   if os.path.exists(artifact_filename):
@@ -1424,13 +1424,10 @@ def install_mldiagnostics_yaml(artifact_filename: str):
 
     except PermissionError:
       xpk_print(f'Failed to delete file {artifact_filename} due to Permission Error.')
-      
-    except Exception as e:
-      xpk_print(f'Failed to delete file {artifact_filename}. Unexpected error: {e}')
 
   else:
     xpk_print(f'File {artifact_filename} does not exist locally. Skipping deletion (Cleanup assumed).')
-  
+
   return return_code
 
 def label_default_namespace_mldiagnostics():
@@ -1440,13 +1437,13 @@ def label_default_namespace_mldiagnostics():
   Returns:
     0 if successful and 1 otherwise.
   """
-  
-  command = f'kubectl label namespace default diagon-enabled=true'
-  
+
+  command = 'kubectl label namespace default diagon-enabled=true'
+
   return_code = run_command_with_updates(
-      command, f"Starting kubectl label namespace default with diagon-enabled=true..."
+      command, 'Starting kubectl label namespace default with diagon-enabled=true...'
   )
-  
+
   if return_code != 0:
     xpk_print(f'Namespace labeling returned with ERROR {return_code}.\n')
     xpk_exit(return_code)
@@ -1475,46 +1472,46 @@ def install_diagon_prerequisites():
     cert_webhook_ready = wait_for_deployment_ready(cert_webhook_deployment_name, cert_webhook_namespace_name)
     if cert_webhook_ready:    
 
-      webhook_package = "mldiagnostics-injection-webhook"
-      webhook_version = "v0.3.0"
-      webhook_filename = f"{webhook_package}-{webhook_version}.yaml"
+      webhook_package = 'mldiagnostics-injection-webhook'
+      webhook_version = 'v0.3.0'
+      webhook_filename = f'{webhook_package}-{webhook_version}.yaml'
 
       return_code = download_mldiagnostics_yaml(package_name=webhook_package, version=webhook_version)
       if return_code != 0:
         return return_code
-      
+
       return_code = create_mldiagnostics_namespace()
       if return_code != 0:
         return return_code
 
       return_code = install_mldiagnostics_yaml(artifact_filename=webhook_filename)
       if return_code != 0:
         return return_code
-      
+
       return_code = label_default_namespace_mldiagnostics()
       if return_code != 0:
         return return_code
 
       # --- Install Operator ---
-      operator_package = "mldiagnostics-connection-operator"
-      operator_version = "v0.3.0"
-      operator_filename = f"{operator_package}-{operator_version}.yaml"
-      
+      operator_package = 'mldiagnostics-connection-operator'
+      operator_version = 'v0.3.0'
+      operator_filename = f'{operator_package}-{operator_version}.yaml'
+
       return_code = download_mldiagnostics_yaml(package_name=operator_package, version=operator_version)
       if return_code != 0:
         return return_code
-        
+
       return_code = install_mldiagnostics_yaml(artifact_filename=operator_filename)
       if return_code != 0:
         return return_code
-      
-      xpk_print("All diagon installation and setup steps have been successfully completed!")
+
+      xpk_print('All diagon installation and setup steps have been successfully completed!')
       return return_code
     else:
-      xpk_print("The cert-manager-webhook installation failed.")
+      xpk_print('The cert-manager-webhook installation failed.')
       xpk_exit(1)
   else:
-    xpk_print(f"Application {deployment_name} failed to become ready within the timeout.")
+    xpk_print(f'Application {deployment_name} failed to become ready within the timeout.')
     xpk_exit(1)
 
 def wait_for_deployment_ready(deployment_name: str, namespace: str, timeout_seconds: int = 300) -> bool:
@@ -1530,28 +1527,22 @@ def wait_for_deployment_ready(deployment_name: str, namespace: str, timeout_seco
   Returns:
       bool: True if the Deployment successfully rolled out, False otherwise (timeout or error).
   """
-  
+
   command = (
       f'kubectl rollout status deployment/{deployment_name} -n {namespace}'
       f' --timeout={timeout_seconds}s'
   )
-  
-  print(f"Waiting for deployment {deployment_name} in namespace {namespace} to successfully roll out...")
 
-  try:
-    return_code, return_output = run_command_for_value(
-        command, f'Checking status of deployment {deployment_name}...'
-    )
+  print(f'Waiting for deployment {deployment_name} in namespace {namespace} to successfully roll out...')
 
-    if return_code != 0:
-      xpk_print(f"\nError: Deployment {deployment_name} failed to roll out.")
-      xpk_print(f"kubectl output: {return_output}")
-      return False
+  return_code, return_output = run_command_for_value(
+      command, f'Checking status of deployment {deployment_name}...'
+  )
 
-    xpk_print(f"Success: Deployment {deployment_name} successfully rolled out.")
-    return True
-      
-  except Exception as e:
-    xpk_print(f"\nUnexpected API request error while checking deployment status: {e}")
-    time.sleep(10)
+  if return_code != 0:
+    xpk_print(f'\nError: Deployment {deployment_name} failed to roll out.')
+    xpk_print(f'kubectl output: {return_output}')
     return False
+
+  xpk_print(f'Success: Deployment {deployment_name} successfully rolled out.')
+  return True