[UX] Rephrase service initialization timeout (#3176)

* fix

* Update sky/serve/serve_utils.py

Co-authored-by: Zongheng Yang <[email protected]>

* lint & rename vars

* change func name

---------

Co-authored-by: Zongheng Yang <[email protected]>

sky/serve/constants.py

@@ -12,8 +12,8 @@
 # Signal file path for controller to handle signals.
 SIGNAL_FILE_PATH = '/tmp/sky_serve_controller_signal_{}'
 
-# Time to wait in seconds for service to initialize.
-INITIALIZATION_TIMEOUT_SECONDS = 60
+# Time to wait in seconds for service to register on the controller.
+SERVICE_REGISTER_TIMEOUT_SECONDS = 60
 
 # The time interval in seconds for load balancer to sync with controller. Every
 # time the load balancer syncs with controller, it will update all available

sky/serve/core.py

@@ -177,11 +177,11 @@ def up(
         # TODO(tian): Cache endpoint locally to speedup. Endpoint won't
         # change after the first time, so there is no consistency issue.
         with rich_utils.safe_status(
-                '[cyan]Waiting for the service to initialize[/]'):
+                '[cyan]Waiting for the service to register[/]'):
             # This function will check the controller job id in the database
             # and return the endpoint if the job id matches. Otherwise it will
             # return None.
-            code = serve_utils.ServeCodeGen.wait_service_initialization(
+            code = serve_utils.ServeCodeGen.wait_service_registration(
                 service_name, controller_job_id)
             backend = backend_utils.get_backend_from_handle(controller_handle)
             assert isinstance(backend, backends.CloudVmRayBackend)

sky/serve/serve_utils.py

@@ -432,7 +432,7 @@ def terminate_services(service_names: Optional[List[str]], purge: bool) -> str:
     return '\n'.join(messages)
 
 
-def wait_service_initialization(service_name: str, job_id: int) -> str:
+def wait_service_registration(service_name: str, job_id: int) -> str:
     """Util function to call at the end of `sky.serve.up()`.
 
     This function will:
@@ -465,10 +465,19 @@ def wait_service_initialization(service_name: str, job_id: int) -> str:
                 raise RuntimeError('Max number of services reached. '
                                    'To spin up more services, please '
                                    'tear down some existing services.')
-        if time.time() - start_time > constants.INITIALIZATION_TIMEOUT_SECONDS:
+        elapsed = time.time() - start_time
+        if elapsed > constants.SERVICE_REGISTER_TIMEOUT_SECONDS:
+            # Print the controller log to help user debug.
+            controller_log_path = (
+                generate_remote_controller_log_file_name(service_name))
+            with open(os.path.expanduser(controller_log_path),
+                      'r',
+                      encoding='utf-8') as f:
+                log_content = f.read()
             with ux_utils.print_exception_no_traceback():
-                raise ValueError(
-                    f'Initialization of service {service_name!r} timeout.')
+                raise ValueError(f'Failed to register service {service_name!r} '
+                                 'on the SkyServe controller. '
+                                 f'Reason:\n{log_content}')
         time.sleep(1)
 
 
@@ -836,9 +845,9 @@ def terminate_services(cls, service_names: Optional[List[str]],
         return cls._build(code)
 
     @classmethod
-    def wait_service_initialization(cls, service_name: str, job_id: int) -> str:
+    def wait_service_registration(cls, service_name: str, job_id: int) -> str:
         code = [
-            'msg = serve_utils.wait_service_initialization('
+            'msg = serve_utils.wait_service_registration('
             f'{service_name!r}, {job_id})', 'print(msg, end="", flush=True)'
         ]
         return cls._build(code)