skypilot-nightly 1.0.0.dev20241019__py3-none-any.whl → 1.0.0.dev20241020__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 = '9201def0ff1ac73681a82a26d46f56d0b027b03b'
+_SKYPILOT_COMMIT_SHA = 'c6ae536d8dfedc3bbcf427a81480382b9d5f4c29'
 
 
 def _get_git_commit():
@@ -35,7 +35,7 @@ def _get_git_commit():
 
 
 __commit__ = _get_git_commit()
-__version__ = '1.0.0.dev20241019'
+__version__ = '1.0.0.dev20241020'
 __root_dir__ = os.path.dirname(os.path.abspath(__file__))
 
 

sky/cli.py

@@ -4380,9 +4380,14 @@ def serve_status(all: bool, endpoint: bool, service_names: List[str]):
               default=False,
               required=False,
               help='Skip confirmation prompt.')
+@click.option('--replica-id',
+              default=None,
+              type=int,
+              help='Tear down a given replica')
 # pylint: disable=redefined-builtin
-def serve_down(service_names: List[str], all: bool, purge: bool, yes: bool):
-    """Teardown service(s).
+def serve_down(service_names: List[str], all: bool, purge: bool, yes: bool,
+               replica_id: Optional[int]):
+    """Teardown service(s) or a replica.
 
     SERVICE_NAMES is the name of the service (or glob pattern) to tear down. If
     both SERVICE_NAMES and ``--all`` are supplied, the latter takes precedence.
@@ -4408,6 +4413,12 @@ def serve_down(service_names: List[str], all: bool, purge: bool, yes: bool):
         \b
         # Forcefully tear down a service in failed status.
         sky serve down failed-service --purge
+        \b
+        # Tear down a specific replica
+        sky serve down my-service --replica-id 1
+        \b
+        # Forcefully tear down a specific replica, even in failed status.
+        sky serve down my-service --replica-id 1 --purge
     """
     if sum([len(service_names) > 0, all]) != 1:
         argument_str = f'SERVICE_NAMES={",".join(service_names)}' if len(
@@ -4417,22 +4428,45 @@ def serve_down(service_names: List[str], all: bool, purge: bool, yes: bool):
             'Can only specify one of SERVICE_NAMES or --all. '
             f'Provided {argument_str!r}.')
 
+    replica_id_is_defined = replica_id is not None
+    if replica_id_is_defined:
+        if len(service_names) != 1:
+            service_names_str = ', '.join(service_names)
+            raise click.UsageError(f'The --replica-id option can only be used '
+                                   f'with a single service name. Got: '
+                                   f'{service_names_str}.')
+        if all:
+            raise click.UsageError('The --replica-id option cannot be used '
+                                   'with the --all option.')
+
     backend_utils.is_controller_accessible(
         controller=controller_utils.Controllers.SKY_SERVE_CONTROLLER,
         stopped_message='All services should have been terminated.',
         exit_if_not_accessible=True)
 
     if not yes:
-        quoted_service_names = [f'{name!r}' for name in service_names]
-        service_identity_str = f'service(s) {", ".join(quoted_service_names)}'
-        if all:
-            service_identity_str = 'all services'
-        click.confirm(f'Terminating {service_identity_str}. Proceed?',
-                      default=True,
-                      abort=True,
-                      show_default=True)
-
-    serve_lib.down(service_names=service_names, all=all, purge=purge)
+        if replica_id_is_defined:
+            click.confirm(
+                f'Terminating replica ID {replica_id} in '
+                f'{service_names[0]!r}. Proceed?',
+                default=True,
+                abort=True,
+                show_default=True)
+        else:
+            quoted_service_names = [f'{name!r}' for name in service_names]
+            service_identity_str = (f'service(s) '
+                                    f'{", ".join(quoted_service_names)}')
+            if all:
+                service_identity_str = 'all services'
+            click.confirm(f'Terminating {service_identity_str}. Proceed?',
+                          default=True,
+                          abort=True,
+                          show_default=True)
+
+    if replica_id_is_defined:
+        serve_lib.terminate_replica(service_names[0], replica_id, purge)
+    else:
+        serve_lib.down(service_names=service_names, all=all, purge=purge)
 
 
 @serve.command('logs', cls=_DocumentedCodeCommand)

sky/serve/__init__.py

@@ -8,6 +8,7 @@ from sky.serve.constants import SKYSERVE_METADATA_DIR
 from sky.serve.core import down
 from sky.serve.core import status
 from sky.serve.core import tail_logs
+from sky.serve.core import terminate_replica
 from sky.serve.core import up
 from sky.serve.core import update
 from sky.serve.serve_state import ReplicaStatus
@@ -42,6 +43,7 @@ __all__ = [
     'SKY_SERVE_CONTROLLER_NAME',
     'SKYSERVE_METADATA_DIR',
     'status',
+    'terminate_replica',
     'tail_logs',
     'up',
     'update',

sky/serve/constants.py

@@ -92,4 +92,11 @@ REPLICA_ID_ENV_VAR = 'SKYPILOT_SERVE_REPLICA_ID'
 # change for the serve_utils.ServeCodeGen, we need to bump this version, so that
 # the user can be notified to update their SkyPilot serve version on the remote
 # cluster.
-SERVE_VERSION = 1
+# Changelog:
+# v1.0 - Introduce rolling update.
+# v2.0 - Added template-replica feature.
+SERVE_VERSION = 2
+
+TERMINATE_REPLICA_VERSION_MISMATCH_ERROR = (
+    'The version of service is outdated and does not support manually '
+    'terminating replicas. Please terminate the service and spin up again.')

sky/serve/controller.py

@@ -9,6 +9,7 @@ import time
 import traceback
 from typing import Any, Dict, List
 
+import colorama
 import fastapi
 from fastapi import responses
 import uvicorn
@@ -157,6 +158,75 @@ class SkyServeController:
                 return responses.JSONResponse(content={'message': 'Error'},
                                               status_code=500)
 
+        @self._app.post('/controller/terminate_replica')
+        async def terminate_replica(
+                request: fastapi.Request) -> fastapi.Response:
+            request_data = await request.json()
+            replica_id = request_data['replica_id']
+            assert isinstance(replica_id,
+                              int), 'Error: replica ID must be an integer.'
+            purge = request_data['purge']
+            assert isinstance(purge, bool), 'Error: purge must be a boolean.'
+            replica_info = serve_state.get_replica_info_from_id(
+                self._service_name, replica_id)
+            assert replica_info is not None, (f'Error: replica '
+                                              f'{replica_id} does not exist.')
+            replica_status = replica_info.status
+
+            if replica_status == serve_state.ReplicaStatus.SHUTTING_DOWN:
+                return responses.JSONResponse(
+                    status_code=409,
+                    content={
+                        'message':
+                            f'Replica {replica_id} of service '
+                            f'{self._service_name!r} is already in the process '
+                            f'of terminating. Skip terminating now.'
+                    })
+
+            if (replica_status in serve_state.ReplicaStatus.failed_statuses()
+                    and not purge):
+                return responses.JSONResponse(
+                    status_code=409,
+                    content={
+                        'message': f'{colorama.Fore.YELLOW}Replica '
+                                   f'{replica_id} of service '
+                                   f'{self._service_name!r} is in failed '
+                                   f'status ({replica_info.status}). '
+                                   f'Skipping its termination as it could '
+                                   f'lead to a resource leak. '
+                                   f'(Use `sky serve down '
+                                   f'{self._service_name!r} --replica-id '
+                                   f'{replica_id} --purge` to '
+                                   'forcefully terminate the replica.)'
+                                   f'{colorama.Style.RESET_ALL}'
+                    })
+
+            self._replica_manager.scale_down(replica_id, purge=purge)
+
+            action = 'terminated' if not purge else 'purged'
+            message = (f'{colorama.Fore.GREEN}Replica {replica_id} of service '
+                       f'{self._service_name!r} is scheduled to be '
+                       f'{action}.{colorama.Style.RESET_ALL}\n'
+                       f'Please use {ux_utils.BOLD}sky serve status '
+                       f'{self._service_name}{ux_utils.RESET_BOLD} '
+                       f'to check the latest status.')
+            return responses.JSONResponse(status_code=200,
+                                          content={'message': message})
+
+        @self._app.exception_handler(Exception)
+        async def validation_exception_handler(
+                request: fastapi.Request, exc: Exception) -> fastapi.Response:
+            with ux_utils.enable_traceback():
+                logger.error(f'Error in controller: {exc!r}')
+            return responses.JSONResponse(
+                status_code=500,
+                content={
+                    'message':
+                        (f'Failed method {request.method} at URL {request.url}.'
+                         f' Exception message is {exc!r}.')
+                },
+            )
+
         threading.Thread(target=self._run_autoscaler).start()
 
         logger.info('SkyServe Controller started on '

sky/serve/core.py

@@ -503,6 +503,53 @@ def down(
     sky_logging.print(stdout)
 
 
+@usage_lib.entrypoint
+def terminate_replica(service_name: str, replica_id: int, purge: bool) -> None:
+    """Tear down a specific replica for the given service.
+
+    Args:
+        service_name: Name of the service.
+        replica_id: ID of replica to terminate.
+        purge: Whether to terminate replicas in a failed status. These replicas
+          may lead to resource leaks, so we require the user to explicitly
+          specify this flag to make sure they are aware of this potential
+          resource leak.
+
+    Raises:
+        sky.exceptions.ClusterNotUpError: if the sky sere controller is not up.
+        RuntimeError: if failed to terminate the replica.
+    """
+    handle = backend_utils.is_controller_accessible(
+        controller=controller_utils.Controllers.SKY_SERVE_CONTROLLER,
+        stopped_message=
+        'No service is running now. Please spin up a service first.',
+        non_existent_message='No service is running now. '
+        'Please spin up a service first.',
+    )
+
+    backend = backend_utils.get_backend_from_handle(handle)
+    assert isinstance(backend, backends.CloudVmRayBackend)
+
+    code = serve_utils.ServeCodeGen.terminate_replica(service_name, replica_id,
+                                                      purge)
+    returncode, stdout, stderr = backend.run_on_head(handle,
+                                                     code,
+                                                     require_outputs=True,
+                                                     stream_logs=False,
+                                                     separate_stderr=True)
+
+    try:
+        subprocess_utils.handle_returncode(returncode,
+                                           code,
+                                           'Failed to terminate the replica',
+                                           stderr,
+                                           stream_logs=True)
+    except exceptions.CommandError as e:
+        raise RuntimeError(e.error_msg) from e
+
+    sky_logging.print(stdout)
+
+
 @usage_lib.entrypoint
 def status(
     service_names: Optional[Union[str,

sky/serve/replica_managers.py

@@ -247,6 +247,8 @@ class ReplicaStatusProperty:
     is_scale_down: bool = False
     # The replica's spot instance was preempted.
     preempted: bool = False
+    # Whether the replica is purged.
+    purged: bool = False
 
     def remove_terminated_replica(self) -> bool:
         """Whether to remove the replica record from the replica table.
@@ -307,6 +309,8 @@ class ReplicaStatusProperty:
             return False
         if self.preempted:
             return False
+        if self.purged:
+            return False
         return True
 
     def to_replica_status(self) -> serve_state.ReplicaStatus:
@@ -590,7 +594,7 @@ class ReplicaManager:
         """
         raise NotImplementedError
 
-    def scale_down(self, replica_id: int) -> None:
+    def scale_down(self, replica_id: int, purge: bool = False) -> None:
         """Scale down replica with replica_id."""
         raise NotImplementedError
 
@@ -679,7 +683,8 @@ class SkyPilotReplicaManager(ReplicaManager):
                            replica_id: int,
                            sync_down_logs: bool,
                            replica_drain_delay_seconds: int,
-                           is_scale_down: bool = False) -> None:
+                           is_scale_down: bool = False,
+                           purge: bool = False) -> None:
 
         if replica_id in self._launch_process_pool:
             info = serve_state.get_replica_info_from_id(self._service_name,
@@ -763,16 +768,18 @@ class SkyPilotReplicaManager(ReplicaManager):
         )
         info.status_property.sky_down_status = ProcessStatus.RUNNING
         info.status_property.is_scale_down = is_scale_down
+        info.status_property.purged = purge
         serve_state.add_or_update_replica(self._service_name, replica_id, info)
         p.start()
         self._down_process_pool[replica_id] = p
 
-    def scale_down(self, replica_id: int) -> None:
+    def scale_down(self, replica_id: int, purge: bool = False) -> None:
         self._terminate_replica(
             replica_id,
             sync_down_logs=False,
             replica_drain_delay_seconds=_DEFAULT_DRAIN_SECONDS,
-            is_scale_down=True)
+            is_scale_down=True,
+            purge=purge)
 
     def _handle_preemption(self, info: ReplicaInfo) -> bool:
         """Handle preemption of the replica if any error happened.
@@ -911,6 +918,8 @@ class SkyPilotReplicaManager(ReplicaManager):
                 # since user should fixed the error before update.
                 elif info.version != self.latest_version:
                     removal_reason = 'for version outdated'
+                elif info.status_property.purged:
+                    removal_reason = 'for purge'
                 else:
                     logger.info(f'Termination of replica {replica_id} '
                                 'finished. Replica info is kept since some '

sky/serve/serve_utils.py

@@ -313,6 +313,36 @@ def update_service_encoded(service_name: str, version: int, mode: str) -> str:
     return common_utils.encode_payload(service_msg)
 
 
+def terminate_replica(service_name: str, replica_id: int, purge: bool) -> str:
+    service_status = _get_service_status(service_name)
+    if service_status is None:
+        with ux_utils.print_exception_no_traceback():
+            raise ValueError(f'Service {service_name!r} does not exist.')
+    replica_info = serve_state.get_replica_info_from_id(service_name,
+                                                        replica_id)
+    if replica_info is None:
+        with ux_utils.print_exception_no_traceback():
+            raise ValueError(
+                f'Replica {replica_id} for service {service_name} does not '
+                'exist.')
+
+    controller_port = service_status['controller_port']
+    resp = requests.post(
+        _CONTROLLER_URL.format(CONTROLLER_PORT=controller_port) +
+        '/controller/terminate_replica',
+        json={
+            'replica_id': replica_id,
+            'purge': purge,
+        })
+
+    message: str = resp.json()['message']
+    if resp.status_code != 200:
+        with ux_utils.print_exception_no_traceback():
+            raise ValueError(f'Failed to terminate replica {replica_id} '
+                             f'in {service_name}. Reason:\n{message}')
+    return message
+
+
 def _get_service_status(
         service_name: str,
         with_replica_info: bool = True) -> Optional[Dict[str, Any]]:
@@ -735,7 +765,7 @@ def _get_replicas(service_record: Dict[str, Any]) -> str:
 
 
 def get_endpoint(service_record: Dict[str, Any]) -> str:
-    # Don't use backend_utils.is_controller_up since it is too slow.
+    # Don't use backend_utils.is_controller_accessible since it is too slow.
     handle = global_user_state.get_handle_from_cluster_name(
         SKY_SERVE_CONTROLLER_NAME)
     assert isinstance(handle, backends.CloudVmRayResourceHandle)
@@ -915,6 +945,18 @@ class ServeCodeGen:
         ]
         return cls._build(code)
 
+    @classmethod
+    def terminate_replica(cls, service_name: str, replica_id: int,
+                          purge: bool) -> str:
+        code = [
+            f'(lambda: print(serve_utils.terminate_replica({service_name!r}, '
+            f'{replica_id}, {purge}), end="", flush=True) '
+            'if getattr(constants, "SERVE_VERSION", 0) >= 2 else '
+            f'exec("raise RuntimeError('
+            f'{constants.TERMINATE_REPLICA_VERSION_MISMATCH_ERROR!r})"))()'
+        ]
+        return cls._build(code)
+
     @classmethod
     def wait_service_registration(cls, service_name: str, job_id: int) -> str:
         code = [

{skypilot_nightly-1.0.0.dev20241019.dist-info → skypilot_nightly-1.0.0.dev20241020.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: skypilot-nightly
-Version: 1.0.0.dev20241019
+Version: 1.0.0.dev20241020
 Summary: SkyPilot: An intercloud broker for the clouds
 Author: SkyPilot Team
 License: Apache 2.0

{skypilot_nightly-1.0.0.dev20241019.dist-info → skypilot_nightly-1.0.0.dev20241020.dist-info}/RECORD

@@ -1,8 +1,8 @@
-sky/__init__.py,sha256=LKUWX7CbWSSP_7RKe0BbpUtwfW9AyCM0C6t-6csG3dI,5854
+sky/__init__.py,sha256=njbGTeVXmuel8rNQYbPE9POlsdZUizsa3jEcGfwJklE,5854
 sky/admin_policy.py,sha256=hPo02f_A32gCqhUueF0QYy1fMSSKqRwYEg_9FxScN_s,3248
 sky/authentication.py,sha256=pAdCT60OxxiXI9KXDyP2lQ9u9vMc6aMtq5Xi2h_hbdw,20984
 sky/check.py,sha256=jLMIIJrseaZj1_o5WkbaD9XdyXIlCaT6pyAaIFdhdmA,9079
-sky/cli.py,sha256=PJR6W92twf89j17OWLQJ9RawdazJcGslfW2L_fLB2PM,208545
+sky/cli.py,sha256=XcQeVtH5J7xcARGagYicmTUfd2145jN2nvnZaZXtZlI,209981
 sky/cloud_stores.py,sha256=RjFgmRhUh1Kk__f6g3KxzLp9s7dA0pFK4W1AukEuUaw,21153
 sky/core.py,sha256=DW9OGE2kS2CmsvQ1grrpRnNFS3woMGWSHu5GE99e-I4,38190
 sky/dag.py,sha256=WLFWr5hfrwjd31uYlNvI-zWUk7tLaT_gzJn4LzbVtkE,2780
@@ -171,16 +171,16 @@ sky/provision/vsphere/common/service_manager_factory.py,sha256=YkvfHiRXFK_Nb406z
 sky/provision/vsphere/common/ssl_helper.py,sha256=TYzN9K0i_Mk_17PKGyGPgvOGfoizysuuIeYapcy_tWE,795
 sky/provision/vsphere/common/vapiconnect.py,sha256=R2I1ZWBA19d11fZ_FrIzQT8E1aLl1HU4Rdcj8Z5r3NE,2932
 sky/provision/vsphere/common/vim_utils.py,sha256=EMWLS8ILpdx6XwUZ9I53y0B_1yFrRrlr4jjIMT84hAc,17877
-sky/serve/__init__.py,sha256=Qg_XPOtQsUxiN-Q3njHZRfzoMcQ_KKU1QthkiTbESDw,1661
+sky/serve/__init__.py,sha256=gFZt7W3UPMi4qvYe2xgkHg1VxbR1WGavKyWLBUD3mpg,1731
 sky/serve/autoscalers.py,sha256=khY1oZ22PRaUQNsLCoNKH178X_NiJw0LSLOKr7_LNgY,30275
-sky/serve/constants.py,sha256=OansIC7a0Pwat-Y5SF43T9phad_EvyjKO3peZgKFEHk,4367
-sky/serve/controller.py,sha256=gfE_gB7wxE1VxvnYqw_-KcMGc6X2kufl-NLR7sWdzdY,8172
-sky/serve/core.py,sha256=yebcgmafGwKppXA1vyJdnbWdOg5BSlh87pKL9gkzHPE,29066
+sky/serve/constants.py,sha256=7MflfgTHO9gDSux93U4BmNeEMWXxZB4q7I54KUwgp-s,4651
+sky/serve/controller.py,sha256=R5iIEGEEFtbm_6MvSGelYZP-vSmW0cSFuy64OexUc4g,11719
+sky/serve/core.py,sha256=pz62ERWyHcg2p-rtzVjBZaWmKrK6Hx213YPoa_J5Tlo,31097
 sky/serve/load_balancer.py,sha256=aUfDsgUT_fYrchCwJCeunMPXmAkwJAY58BEu-IN2FaA,11571
 sky/serve/load_balancing_policies.py,sha256=ExdwH_pxPYpJ6CkoTQCOPSa4lzwbq1LFFMKzmIu8ryk,2331
-sky/serve/replica_managers.py,sha256=dO962WZ_6YWRDpyNemY7SzC7fZHlNfoL4kUS3MaKwDo,57405
+sky/serve/replica_managers.py,sha256=1xYDK9Te5wFEF5hUK0gyNIUib0MY-HScLHUBDlTSl-k,57774
 sky/serve/serve_state.py,sha256=5BZSKKKxQRk-0mku17Ch4Veu4qOhaFvaOJY3zrZCkLw,19315
-sky/serve/serve_utils.py,sha256=im_1cJoJmufFxkBVnhK4nI6XlHvEXersQyIivNruJJc,38009
+sky/serve/serve_utils.py,sha256=egGb4HB4yMyFISqZgMWnoHH8AfuLGt3xq4raU8V0qds,39755
 sky/serve/service.py,sha256=fkfJvNJ2BO6rfV0TblZG-QkOXaCyZlpkwbGgrsTzf2w,11872
 sky/serve/service_spec.py,sha256=iRhW95SERvb4NWtV10uCuhgvW31HuSAmZZ55OX0WK8s,15309
 sky/setup_files/MANIFEST.in,sha256=CXz8lIJMgWlH9TvYgzIL3vPFtSDoQq-UMfD9K62rtH4,590
@@ -274,9 +274,9 @@ sky/utils/kubernetes/k8s_gpu_labeler_job.yaml,sha256=KPqp23B-zQ2SZK03jdHeF9fLTog
 sky/utils/kubernetes/k8s_gpu_labeler_setup.yaml,sha256=VLKT2KKimZu1GDg_4AIlIt488oMQvhRZWwsj9vBbPUg,3812
 sky/utils/kubernetes/rsync_helper.sh,sha256=aRMa_0JRHtXFOPtEg4rFAwR1t57wvvAoGZhn3H3BtGk,1059
 sky/utils/kubernetes/ssh_jump_lifecycle_manager.py,sha256=RFLJ3k7MR5UN4SKHykQ0lV9SgXumoULpKYIAt1vh-HU,6560
-skypilot_nightly-1.0.0.dev20241019.dist-info/LICENSE,sha256=emRJAvE7ngL6x0RhQvlns5wJzGI3NEQ_WMjNmd9TZc4,12170
-skypilot_nightly-1.0.0.dev20241019.dist-info/METADATA,sha256=AE2fCPLtATmQ_7yujx2cd9zAwVowsMapi0bTc7-Gk6A,19540
-skypilot_nightly-1.0.0.dev20241019.dist-info/WHEEL,sha256=OVMc5UfuAQiSplgO0_WdW7vXVGAt9Hdd6qtN4HotdyA,91
-skypilot_nightly-1.0.0.dev20241019.dist-info/entry_points.txt,sha256=StA6HYpuHj-Y61L2Ze-hK2IcLWgLZcML5gJu8cs6nU4,36
-skypilot_nightly-1.0.0.dev20241019.dist-info/top_level.txt,sha256=qA8QuiNNb6Y1OF-pCUtPEr6sLEwy2xJX06Bd_CrtrHY,4
-skypilot_nightly-1.0.0.dev20241019.dist-info/RECORD,,
+skypilot_nightly-1.0.0.dev20241020.dist-info/LICENSE,sha256=emRJAvE7ngL6x0RhQvlns5wJzGI3NEQ_WMjNmd9TZc4,12170
+skypilot_nightly-1.0.0.dev20241020.dist-info/METADATA,sha256=L8MmlJIr14EcjFiMMWqkaWvJsVZnm_SmhgzceSuhdRs,19540
+skypilot_nightly-1.0.0.dev20241020.dist-info/WHEEL,sha256=OVMc5UfuAQiSplgO0_WdW7vXVGAt9Hdd6qtN4HotdyA,91
+skypilot_nightly-1.0.0.dev20241020.dist-info/entry_points.txt,sha256=StA6HYpuHj-Y61L2Ze-hK2IcLWgLZcML5gJu8cs6nU4,36
+skypilot_nightly-1.0.0.dev20241020.dist-info/top_level.txt,sha256=qA8QuiNNb6Y1OF-pCUtPEr6sLEwy2xJX06Bd_CrtrHY,4
+skypilot_nightly-1.0.0.dev20241020.dist-info/RECORD,,