skypilot-nightly 1.0.0.dev20250228__py3-none-any.whl → 1.0.0.dev2025022801__py3-none-any.whl

sky/__init__.py

@@ -5,7 +5,7 @@ from typing import Optional
 import urllib.request
 
 # Replaced with the current commit when building the wheels.
-_SKYPILOT_COMMIT_SHA = '0f178b2af47ec2b185bc685cce6965c675613cc7'
+_SKYPILOT_COMMIT_SHA = '2619dc8f63f271852a6b00f91381b27428ff1780'
 
 
 def _get_git_commit():
@@ -35,7 +35,7 @@ def _get_git_commit():
 
 
 __commit__ = _get_git_commit()
-__version__ = '1.0.0.dev20250228'
+__version__ = '1.0.0.dev2025022801'
 __root_dir__ = os.path.dirname(os.path.abspath(__file__))
 
 

sky/cli.py

@@ -5504,10 +5504,19 @@ def api():
               required=False,
               help=('The host to deploy the SkyPilot API server. To allow '
                     'remote access, set this to 0.0.0.0'))
+@click.option('--foreground',
+              is_flag=True,
+              default=False,
+              required=False,
+              help='Run the SkyPilot API server in the foreground and output '
+              'its logs to stdout/stderr. Allowing external systems '
+              'to manage the process lifecycle and collect logs directly. '
+              'This is useful when the API server is managed by systems '
+              'like systemd and Kubernetes.')
 @usage_lib.entrypoint
-def api_start(deploy: bool, host: Optional[str]):
+def api_start(deploy: bool, host: Optional[str], foreground: bool):
     """Starts the SkyPilot API server locally."""
-    sdk.api_start(deploy=deploy, host=host)
+    sdk.api_start(deploy=deploy, host=host, foreground=foreground)
 
 
 @api.command('stop', cls=_DocumentedCodeCommand)

sky/client/cli.py

@@ -5504,10 +5504,19 @@ def api():
               required=False,
               help=('The host to deploy the SkyPilot API server. To allow '
                     'remote access, set this to 0.0.0.0'))
+@click.option('--foreground',
+              is_flag=True,
+              default=False,
+              required=False,
+              help='Run the SkyPilot API server in the foreground and output '
+              'its logs to stdout/stderr. Allowing external systems '
+              'to manage the process lifecycle and collect logs directly. '
+              'This is useful when the API server is managed by systems '
+              'like systemd and Kubernetes.')
 @usage_lib.entrypoint
-def api_start(deploy: bool, host: Optional[str]):
+def api_start(deploy: bool, host: Optional[str], foreground: bool):
     """Starts the SkyPilot API server locally."""
-    sdk.api_start(deploy=deploy, host=host)
+    sdk.api_start(deploy=deploy, host=host, foreground=foreground)
 
 
 @api.command('stop', cls=_DocumentedCodeCommand)

sky/client/sdk.py

@@ -1611,6 +1611,7 @@ def api_start(
     *,
     deploy: bool = False,
     host: str = '127.0.0.1',
+    foreground: bool = False,
 ) -> None:
     """Starts the API server.
 
@@ -1622,7 +1623,8 @@ def api_start(
             resources of the machine.
         host: The host to deploy the API server. It will be set to 0.0.0.0
             if deploy is True, to allow remote access.
-
+        foreground: Whether to run the API server in the foreground (run in
+            the current process).
     Returns:
         None
     """
@@ -1641,7 +1643,10 @@ def api_start(
                              'from the config file and/or unset the '
                              'SKYPILOT_API_SERVER_ENDPOINT environment '
                              'variable.')
-    server_common.check_server_healthy_or_start_fn(deploy, host)
+    server_common.check_server_healthy_or_start_fn(deploy, host, foreground)
+    if foreground:
+        # Explain why current process exited
+        logger.info('API server is already running:')
     logger.info(f'{ux_utils.INDENT_SYMBOL}SkyPilot API server: '
                 f'{server_common.get_server_url(host)}\n'
                 f'{ux_utils.INDENT_LAST_SYMBOL}'

sky/server/common.py

@@ -145,57 +145,6 @@ def get_api_server_status(endpoint: Optional[str] = None) -> ApiServerInfo:
     return ApiServerInfo(status=ApiServerStatus.UNHEALTHY, api_version=None)
 
 
-def start_api_server_in_background(deploy: bool = False,
-                                   host: str = '127.0.0.1'):
-    if not is_api_server_local():
-        raise RuntimeError(
-            f'Cannot start API server: {get_server_url()} is not a local URL')
-
-    # Check available memory before starting the server.
-    avail_mem_size_gb: float = common_utils.get_mem_size_gb()
-    if avail_mem_size_gb <= server_constants.MIN_AVAIL_MEM_GB:
-        logger.warning(
-            f'{colorama.Fore.YELLOW}Your SkyPilot API server machine only has '
-            f'{avail_mem_size_gb:.1f}GB memory available. '
-            f'At least {server_constants.MIN_AVAIL_MEM_GB}GB is recommended to '
-            f'support higher load with better performance.'
-            f'{colorama.Style.RESET_ALL}')
-    log_path = os.path.expanduser(constants.API_SERVER_LOGS)
-    os.makedirs(os.path.dirname(log_path), exist_ok=True)
-
-    api_server_cmd = API_SERVER_CMD
-    if deploy:
-        api_server_cmd += ' --deploy'
-    if host is not None:
-        api_server_cmd += f' --host {host}'
-    cmd = f'{sys.executable} {api_server_cmd} > {log_path} 2>&1 < /dev/null'
-
-    # Start the API server process in the background and don't wait for it.
-    # If this is called from a CLI invocation, we need start_new_session=True so
-    # that SIGINT on the CLI will not also kill the API server.
-    subprocess.Popen(cmd, shell=True, start_new_session=True)
-
-    # Wait for the server to start until timeout.
-    # Conservative upper time bound for starting the server based on profiling.
-    timeout_sec = 12
-    start_time = time.time()
-    while True:
-        api_server_info = get_api_server_status()
-        assert api_server_info.status != ApiServerStatus.VERSION_MISMATCH, (
-            f'API server version mismatch when starting the server. '
-            f'Server version: {api_server_info.api_version} '
-            f'Client version: {server_constants.API_VERSION}')
-        if api_server_info.status == ApiServerStatus.HEALTHY:
-            break
-        elif time.time() - start_time >= timeout_sec:
-            with ux_utils.print_exception_no_traceback():
-                raise RuntimeError(
-                    'Failed to start SkyPilot API server at '
-                    f'{get_server_url(host)}'
-                    f'\nView logs at: {constants.API_SERVER_LOGS}')
-        time.sleep(0.5)
-
-
 def handle_request_error(response: requests.Response) -> None:
     if response.status_code != 200:
         with ux_utils.print_exception_no_traceback():
@@ -218,10 +167,10 @@ def get_request_id(response: requests.Response) -> RequestId:
     return request_id
 
 
-def _start_api_server(deploy: bool = False, host: str = '127.0.0.1'):
+def _start_api_server(deploy: bool = False,
+                      host: str = '127.0.0.1',
+                      foreground: bool = False):
     """Starts a SkyPilot API server locally."""
-    # Lock to prevent multiple processes from starting the server at the
-    # same time, causing issues with database initialization.
     server_url = get_server_url(host)
     assert server_url in AVAILABLE_LOCAL_API_SERVER_URLS, (
         f'server url {server_url} is not a local url')
@@ -230,7 +179,60 @@ def _start_api_server(deploy: bool = False, host: str = '127.0.0.1'):
                     f'SkyPilot API server at {server_url}. '
                     'Starting a local server.'
                     f'{colorama.Style.RESET_ALL}')
-        start_api_server_in_background(deploy=deploy, host=host)
+        if not is_api_server_local():
+            raise RuntimeError(f'Cannot start API server: {get_server_url()} '
+                               'is not a local URL')
+
+        # Check available memory before starting the server.
+        avail_mem_size_gb: float = common_utils.get_mem_size_gb()
+        if avail_mem_size_gb <= server_constants.MIN_AVAIL_MEM_GB:
+            logger.warning(
+                f'{colorama.Fore.YELLOW}Your SkyPilot API server machine only '
+                f'has {avail_mem_size_gb:.1f}GB memory available. '
+                f'At least {server_constants.MIN_AVAIL_MEM_GB}GB is '
+                'recommended to support higher load with better performance.'
+                f'{colorama.Style.RESET_ALL}')
+
+        args = [sys.executable, *API_SERVER_CMD.split()]
+        if deploy:
+            args += ['--deploy']
+        if host is not None:
+            args += [f'--host={host}']
+
+        if foreground:
+            # Replaces the current process with the API server
+            os.execvp(args[0], args)
+
+        log_path = os.path.expanduser(constants.API_SERVER_LOGS)
+        os.makedirs(os.path.dirname(log_path), exist_ok=True)
+        cmd = f'{" ".join(args)} > {log_path} 2>&1 < /dev/null'
+
+        # Start the API server process in the background and don't wait for it.
+        # If this is called from a CLI invocation, we need
+        # start_new_session=True so that SIGINT on the CLI will not also kill
+        # the API server.
+        subprocess.Popen(cmd, shell=True, start_new_session=True)
+
+        # Wait for the server to start until timeout.
+        # Conservative upper time bound for starting the server based on
+        # profiling.
+        timeout_sec = 12
+        start_time = time.time()
+        while True:
+            api_server_info = get_api_server_status()
+            assert api_server_info.status != ApiServerStatus.VERSION_MISMATCH, (
+                f'API server version mismatch when starting the server. '
+                f'Server version: {api_server_info.api_version} '
+                f'Client version: {server_constants.API_VERSION}')
+            if api_server_info.status == ApiServerStatus.HEALTHY:
+                break
+            elif time.time() - start_time >= timeout_sec:
+                with ux_utils.print_exception_no_traceback():
+                    raise RuntimeError(
+                        'Failed to start SkyPilot API server at '
+                        f'{get_server_url(host)}'
+                        f'\nView logs at: {constants.API_SERVER_LOGS}')
+            time.sleep(0.5)
         logger.info(ux_utils.finishing_message('SkyPilot API server started.'))
 
 
@@ -260,7 +262,8 @@ def check_server_healthy(endpoint: Optional[str] = None,) -> None:
 
 
 def check_server_healthy_or_start_fn(deploy: bool = False,
-                                     host: str = '127.0.0.1'):
+                                     host: str = '127.0.0.1',
+                                     foreground: bool = False):
     try:
         check_server_healthy()
     except exceptions.ApiServerConnectionError as exc:
@@ -268,13 +271,15 @@ def check_server_healthy_or_start_fn(deploy: bool = False,
         if not is_api_server_local():
             with ux_utils.print_exception_no_traceback():
                 raise exceptions.ApiServerConnectionError(endpoint) from exc
+        # Lock to prevent multiple processes from starting the server at the
+        # same time, causing issues with database initialization.
         with filelock.FileLock(
                 os.path.expanduser(constants.API_SERVER_CREATION_LOCK_PATH)):
             # Check again if server is already running. Other processes may
             # have started the server while we were waiting for the lock.
             api_server_info = get_api_server_status(endpoint)
             if api_server_info.status == ApiServerStatus.UNHEALTHY:
-                _start_api_server(deploy, host)
+                _start_api_server(deploy, host, foreground)
 
 
 def check_server_healthy_or_start(func):

sky/server/requests/executor.py

@@ -371,7 +371,8 @@ def request_worker(worker: RequestWorker, max_parallel_size: int) -> None:
             # TODO(aylei): crash the API server or recreate the worker process
             # to avoid broken state.
             logger.error(f'[{worker}] Worker process interrupted')
-            raise
+            with ux_utils.print_exception_no_traceback():
+                raise
         except (Exception, SystemExit) as e:  # pylint: disable=broad-except
             # Catch any other exceptions to avoid crashing the worker process.
             logger.error(
@@ -408,6 +409,7 @@ def start(deploy: bool) -> List[multiprocessing.Process]:
         f'long requests and will allow at max '
         f'{max_parallel_for_short} short requests in parallel.')
 
+    sub_procs = []
     # Setup the queues.
     if queue_backend == QueueBackend.MULTIPROCESSING:
         logger.info('Creating shared request queues')
@@ -424,27 +426,26 @@ def start(deploy: bool) -> List[multiprocessing.Process]:
         queue_server = multiprocessing.Process(
             target=mp_queue.start_queue_manager, args=(queue_names, port))
         queue_server.start()
-
+        sub_procs.append(queue_server)
         mp_queue.wait_for_queues_to_be_ready(queue_names, port=port)
 
     logger.info('Request queues created')
 
-    worker_procs = []
     for worker_id in range(max_parallel_for_long):
         worker = RequestWorker(id=worker_id,
                                schedule_type=api_requests.ScheduleType.LONG)
         worker_proc = multiprocessing.Process(target=request_worker,
                                               args=(worker, 1))
         worker_proc.start()
-        worker_procs.append(worker_proc)
+        sub_procs.append(worker_proc)
 
     # Start a worker for short requests.
     worker = RequestWorker(id=1, schedule_type=api_requests.ScheduleType.SHORT)
     worker_proc = multiprocessing.Process(target=request_worker,
                                           args=(worker, max_parallel_for_short))
     worker_proc.start()
-    worker_procs.append(worker_proc)
-    return worker_procs
+    sub_procs.append(worker_proc)
+    return sub_procs
 
 
 @annotations.lru_cache(scope='global', maxsize=1)

sky/server/server.py

@@ -1085,9 +1085,9 @@ if __name__ == '__main__':
     if cmd_args.deploy:
         num_workers = os.cpu_count()
 
-    workers = []
+    sub_procs = []
     try:
-        workers = executor.start(cmd_args.deploy)
+        sub_procs = executor.start(cmd_args.deploy)
         logger.info('Starting SkyPilot API server')
         # We don't support reload for now, since it may cause leakage of request
         # workers or interrupt running requests.
@@ -1101,5 +1101,6 @@ if __name__ == '__main__':
         raise
     finally:
         logger.info('Shutting down SkyPilot API server...')
-        for worker in workers:
-            worker.terminate()
+        for sub_proc in sub_procs:
+            sub_proc.terminate()
+            sub_proc.join()

{skypilot_nightly-1.0.0.dev20250228.dist-info → skypilot_nightly-1.0.0.dev2025022801.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: skypilot-nightly
-Version: 1.0.0.dev20250228
+Version: 1.0.0.dev2025022801
 Summary: SkyPilot: An intercloud broker for the clouds
 Author: SkyPilot Team
 License: Apache 2.0

{skypilot_nightly-1.0.0.dev20250228.dist-info → skypilot_nightly-1.0.0.dev2025022801.dist-info}/RECORD

@@ -1,8 +1,8 @@
-sky/__init__.py,sha256=jLsjhG2RasaqPtD_RYBd_LfTFzhUYIB8j6WkrjZVbKY,6428
+sky/__init__.py,sha256=ePkgbQbrblshpx26zEL8mRs6JSmXHiMsiSEZlh_wg3w,6430
 sky/admin_policy.py,sha256=hPo02f_A32gCqhUueF0QYy1fMSSKqRwYEg_9FxScN_s,3248
 sky/authentication.py,sha256=hCEqi77nprQEg3ktfRL51xiiw16zwZOmFEDB_Z7fWVU,22384
 sky/check.py,sha256=NDKx_Zm7YRxPjMv82wz3ESLnGIPljaACyqVdVNM0PzY,11258
-sky/cli.py,sha256=K28Bowflmjhir2e3hIPbZhsnfqvBsANSeBSnEXFoy10,219929
+sky/cli.py,sha256=pdnOEBu-YjiyZCYeQnAsjUk40ETiGdBKiKk2cZgNe8k,220432
 sky/cloud_stores.py,sha256=-95XIqi_ouo7hvoN5mQNP6bGm07MyF6Yk-YP4Txb5wg,24034
 sky/core.py,sha256=X83hdpPTiWyEJLamrguCd03PUjkRiGgqTFfEBEQkzWc,45471
 sky/dag.py,sha256=Yl7Ry26Vql5cv4YMz8g9kOUgtoCihJnw7c8NgZYakMY,3242
@@ -43,9 +43,9 @@ sky/benchmark/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 sky/benchmark/benchmark_state.py,sha256=X8CXmuU9KgsDRhKedhFgjeRMUFWtQsjFs1qECvPG2yg,8723
 sky/benchmark/benchmark_utils.py,sha256=o4RymqSceq5mLEZL0upQM6NVEzJJQzj9s9tTm49uUTc,26365
 sky/client/__init__.py,sha256=pz6xvVSd9X-gwqbsDL0E9QOojYqM0KAD0j-NCyCIF1k,38
-sky/client/cli.py,sha256=K28Bowflmjhir2e3hIPbZhsnfqvBsANSeBSnEXFoy10,219929
+sky/client/cli.py,sha256=pdnOEBu-YjiyZCYeQnAsjUk40ETiGdBKiKk2cZgNe8k,220432
 sky/client/common.py,sha256=axDic7WOG1e78SdFm5XIwdhX7YNvf3g4k7INrsW3X4s,14611
-sky/client/sdk.py,sha256=U4v8Khu1lf1oUoBuJUhIFnjsFhYM9x8XcKsnVRMtihI,66990
+sky/client/sdk.py,sha256=YPDsrHLBfSjAi2ufDmlYVOTivoDG_Qj1sWq22KdYgJI,67259
 sky/clouds/__init__.py,sha256=OW6mJ-9hpJSBORCgt2LippLQEYZHNfnBW1mooRNNvxo,1416
 sky/clouds/aws.py,sha256=J8tczaTDL239UowN9tUlhI92SeHw01wtFucSckvG63w,54112
 sky/clouds/azure.py,sha256=bawEw6wOLAVyrjxMD-4UjLCuMj1H5_jH8qggpfZYS54,31703
@@ -228,13 +228,13 @@ sky/serve/server/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,
 sky/serve/server/core.py,sha256=pRvFadEIH_WTUkTtSmuFoPBP4JFq8Obt68ifi9DWuog,36865
 sky/serve/server/server.py,sha256=gQGVU9nHYdGbaLhGjIUNIYn4xwKjRASRJkiiTL5AI1Y,3283
 sky/server/__init__.py,sha256=MPPBqFzXz6Jv5QSk6td_IcvnfXfNErDZVcizu4MLRow,27
-sky/server/common.py,sha256=0LphKrp89_sGI-xDakK2uEqI-zKuvbc4OTcuLCiKfmQ,17560
+sky/server/common.py,sha256=pEa-q3P5aOm6RMlit0pVzlDoJnZU_6zViO7aK_7htn0,17843
 sky/server/constants.py,sha256=89jKE3SIe1T3_7j6ECTy4pZnhZZD7fBwsWOCOkTban8,770
-sky/server/server.py,sha256=4ipJG67sBFWylNYdPD1FUhth36yX23XbcROXipRSZsw,42438
+sky/server/server.py,sha256=BPJCsfHy0mZHiwpO_15CgeE1a2VPv3JM5M2HLsy-q3M,42476
 sky/server/stream_utils.py,sha256=-3IX1YCgxAFfcvQIV0TCvOn1wbRLWovAx3ckCrsExWU,5651
 sky/server/html/log.html,sha256=TSGZktua9Ysl_ysg3w60rjxAxhH61AJnsYDHdtqrjmI,6929
 sky/server/requests/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-sky/server/requests/executor.py,sha256=TWX2jYkkstgRyRkWNE19Mgw4_CfzadebOW30iTGdK_Q,19693
+sky/server/requests/executor.py,sha256=Jk8RJoQlicDqaHhgVWMH3UiL-dJS7lGSGd05GPv-Lrc,19781
 sky/server/requests/payloads.py,sha256=QYgEz85jswXkEYxO1mkwPA8MWXD_pETs-g_JH_Tlm_w,16038
 sky/server/requests/requests.py,sha256=aMdjiK5kjSYP36pxdXFU6qgKOXcOmtViHbFm3V8Dvf8,19590
 sky/server/requests/queues/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -344,9 +344,9 @@ sky/utils/kubernetes/k8s_gpu_labeler_setup.yaml,sha256=VLKT2KKimZu1GDg_4AIlIt488
 sky/utils/kubernetes/kubernetes_deploy_utils.py,sha256=iAjfyPclOs8qlALACcfxLpRAO9CZ-h16leFqXZ6tNaY,10096
 sky/utils/kubernetes/rsync_helper.sh,sha256=h4YwrPFf9727CACnMJvF3EyK_0OeOYKKt4su_daKekw,1256
 sky/utils/kubernetes/ssh_jump_lifecycle_manager.py,sha256=Kq1MDygF2IxFmu9FXpCxqucXLmeUrvs6OtRij6XTQbo,6554
-skypilot_nightly-1.0.0.dev20250228.dist-info/LICENSE,sha256=emRJAvE7ngL6x0RhQvlns5wJzGI3NEQ_WMjNmd9TZc4,12170
-skypilot_nightly-1.0.0.dev20250228.dist-info/METADATA,sha256=IBTEc-5NtZ9KP63RfsWsMmXI6qzs7ZBSdi9bAN8QfAc,19236
-skypilot_nightly-1.0.0.dev20250228.dist-info/WHEEL,sha256=jB7zZ3N9hIM9adW7qlTAyycLYW9npaWKLRzaoVcLKcM,91
-skypilot_nightly-1.0.0.dev20250228.dist-info/entry_points.txt,sha256=StA6HYpuHj-Y61L2Ze-hK2IcLWgLZcML5gJu8cs6nU4,36
-skypilot_nightly-1.0.0.dev20250228.dist-info/top_level.txt,sha256=qA8QuiNNb6Y1OF-pCUtPEr6sLEwy2xJX06Bd_CrtrHY,4
-skypilot_nightly-1.0.0.dev20250228.dist-info/RECORD,,
+skypilot_nightly-1.0.0.dev2025022801.dist-info/LICENSE,sha256=emRJAvE7ngL6x0RhQvlns5wJzGI3NEQ_WMjNmd9TZc4,12170
+skypilot_nightly-1.0.0.dev2025022801.dist-info/METADATA,sha256=Kqe348n5KCVwF_hEOLzUTML-yW7aEwEeVUhO8YsDC_s,19238
+skypilot_nightly-1.0.0.dev2025022801.dist-info/WHEEL,sha256=jB7zZ3N9hIM9adW7qlTAyycLYW9npaWKLRzaoVcLKcM,91
+skypilot_nightly-1.0.0.dev2025022801.dist-info/entry_points.txt,sha256=StA6HYpuHj-Y61L2Ze-hK2IcLWgLZcML5gJu8cs6nU4,36
+skypilot_nightly-1.0.0.dev2025022801.dist-info/top_level.txt,sha256=qA8QuiNNb6Y1OF-pCUtPEr6sLEwy2xJX06Bd_CrtrHY,4
+skypilot_nightly-1.0.0.dev2025022801.dist-info/RECORD,,