skypilot-nightly 1.0.0.dev20250922__py3-none-any.whl → 1.0.0.dev20250925__py3-none-any.whl

sky/provision/vsphere/instance.py

@@ -30,9 +30,10 @@ HEAD_NODE_VALUE = '1'
 WORKER_NODE_VALUE = '0'
 
 
-def run_instances(region: str, cluster_name: str,
+def run_instances(region: str, cluster_name: str, cluster_name_on_cloud: str,
                   config: common.ProvisionConfig) -> common.ProvisionRecord:
     """See sky/provision/__init__.py"""
+    del cluster_name  # unused
     logger.info('New provision of Vsphere: run_instances().')
 
     resumed_instance_ids: List[str] = []
@@ -40,7 +41,7 @@ def run_instances(region: str, cluster_name: str,
     vc_object = _get_vc_object(region)
     vc_object.connect()
 
-    exist_instances = _get_filtered_instance(vc_object, cluster_name,
+    exist_instances = _get_filtered_instance(vc_object, cluster_name_on_cloud,
                                              config.provider_config)
     head_instance_id = _get_head_instance_id(exist_instances)
 
@@ -89,8 +90,8 @@ def run_instances(region: str, cluster_name: str,
             config, region, vc_object)
         # TODO: update logic for multi-node creation
         for _ in range(to_start_num):
-            created_instance_uuid = _create_instances(cluster_name, config,
-                                                      region, vc_object,
+            created_instance_uuid = _create_instances(cluster_name_on_cloud,
+                                                      config, region, vc_object,
                                                       vsphere_cluster_name)
             created_instance_ids.append(created_instance_uuid)
         if head_instance_id is None:
@@ -104,7 +105,7 @@ def run_instances(region: str, cluster_name: str,
         provider_name='vsphere',
         region=region,
         zone=vsphere_cluster_name,
-        cluster_name=cluster_name,
+        cluster_name=cluster_name_on_cloud,
         head_instance_id=head_instance_id,
         resumed_instance_ids=resumed_instance_ids,
         created_instance_ids=created_instance_ids,

sky/schemas/api/responses.py

@@ -86,7 +86,7 @@ class StatusResponse(ResponseBaseModel):
     # backends.ResourceHandle, so we use Any here.
     # This is an internally facing field anyway, so it's less
     # of a problem that it's not typed.
-    handle: Any
+    handle: Optional[Any] = None
     last_use: str
     status: status_lib.ClusterStatus
     autostop: int
@@ -118,6 +118,7 @@ class StatusResponse(ResponseBaseModel):
     cpus: Optional[str] = None
     memory: Optional[str] = None
     accelerators: Optional[str] = None
+    cluster_name_on_cloud: Optional[str] = None
 
 
 class UploadStatus(enum.Enum):

sky/serve/autoscalers.py

@@ -411,6 +411,8 @@ class _AutoscalerWithHysteresis(Autoscaler):
         # `_set_target_num_replicas_with_hysteresis` to have the replicas
         # quickly scale after each update.
         self.target_num_replicas = self._calculate_target_num_replicas()
+        logger.debug(f'Target number of replicas: {self.target_num_replicas}'
+                     'after update_version.')
         # Cleanup hysteresis counters.
         self.upscale_counter = 0
         self.downscale_counter = 0

sky/serve/client/impl.py

@@ -105,7 +105,8 @@ def update(
 
 
 def apply(
-    task: Union['sky.Task', 'sky.Dag'],
+    task: Optional[Union['sky.Task', 'sky.Dag']],
+    workers: Optional[int],
     service_name: str,
     mode: 'serve_utils.UpdateMode',
     pool: bool = False,
@@ -117,35 +118,60 @@ def apply(
     # Avoid circular import.
     from sky.client import sdk  # pylint: disable=import-outside-toplevel
 
-    dag = dag_utils.convert_entrypoint_to_dag(task)
-    with admin_policy_utils.apply_and_use_config_in_current_request(
-            dag, at_client_side=True) as dag:
-        sdk.validate(dag)
-        request_id = sdk.optimize(dag)
-        sdk.stream_and_get(request_id)
-        if _need_confirmation:
-            noun = 'pool' if pool else 'service'
-            prompt = f'Applying config to {noun} {service_name!r}. Proceed?'
-            if prompt is not None:
-                click.confirm(prompt,
-                              default=True,
-                              abort=True,
-                              show_default=True)
-
-        dag = client_common.upload_mounts_to_api_server(dag)
-        dag_str = dag_utils.dump_chain_dag_to_yaml_str(dag)
+    noun = 'pool' if pool else 'service'
+    # There are two cases here. If task is None, we should be trying to
+    # update the number of workers in the pool. If task is not None, we should
+    # be trying to apply a new config to the pool. The two code paths
+    # are slightly different with us needing to craft the dag and validate
+    # it if we have a task. In the future we could move this logic to the
+    # server side and simplify this code, for the time being we keep it here.
+    if task is None:
+        if workers is None:
+            raise ValueError(f'Cannot create a new {noun} without specifying '
+                             f'task or workers. Please provide either a task '
+                             f'or specify the number of workers.')
 
         body = payloads.JobsPoolApplyBody(
-            task=dag_str,
+            workers=workers,
             pool_name=service_name,
             mode=mode,
         )
+
         response = server_common.make_authenticated_request(
             'POST',
             '/jobs/pool_apply',
             json=json.loads(body.model_dump_json()),
             timeout=(5, None))
         return server_common.get_request_id(response)
+    else:
+        dag = dag_utils.convert_entrypoint_to_dag(task)
+        with admin_policy_utils.apply_and_use_config_in_current_request(
+                dag, at_client_side=True) as dag:
+            sdk.validate(dag)
+            request_id = sdk.optimize(dag)
+            sdk.stream_and_get(request_id)
+            if _need_confirmation:
+                prompt = f'Applying config to {noun} {service_name!r}. Proceed?'
+                if prompt is not None:
+                    click.confirm(prompt,
+                                  default=True,
+                                  abort=True,
+                                  show_default=True)
+
+            dag = client_common.upload_mounts_to_api_server(dag)
+            dag_str = dag_utils.dump_chain_dag_to_yaml_str(dag)
+
+            body = payloads.JobsPoolApplyBody(
+                task=dag_str,
+                pool_name=service_name,
+                mode=mode,
+            )
+            response = server_common.make_authenticated_request(
+                'POST',
+                '/jobs/pool_apply',
+                json=json.loads(body.model_dump_json()),
+                timeout=(5, None))
+            return server_common.get_request_id(response)
 
 
 def down(

sky/serve/replica_managers.py

@@ -422,11 +422,12 @@ class ReplicaInfo:
                 based on the cluster name.
         """
         if cluster_record is None:
-            cluster_record = global_user_state.get_cluster_from_name(
+            handle = global_user_state.get_handle_from_cluster_name(
                 self.cluster_name)
-        if cluster_record is None:
+        else:
+            handle = cluster_record['handle']
+        if handle is None:
             return None
-        handle = cluster_record['handle']
         assert isinstance(handle, backends.CloudVmRayResourceHandle)
         return handle
 
@@ -443,6 +444,12 @@ class ReplicaInfo:
         handle = self.handle()
         if handle is None:
             return None
+        if self.replica_port == '-':
+            # This is a pool replica so there is no endpoint and it's filled
+            # with this dummy value. We return None here so that we can
+            # get the active ready replicas and perform autoscaling. Otherwise,
+            # would error out when trying to get the endpoint.
+            return None
         replica_port_int = int(self.replica_port)
         try:
             endpoint_dict = backend_utils.get_endpoints(handle.cluster_name,
@@ -470,7 +477,7 @@ class ReplicaInfo:
                      with_handle: bool,
                      with_url: bool = True) -> Dict[str, Any]:
         cluster_record = global_user_state.get_cluster_from_name(
-            self.cluster_name)
+            self.cluster_name, include_user_info=False, summary_response=True)
         info_dict = {
             'replica_id': self.replica_id,
             'name': self.cluster_name,
@@ -956,7 +963,7 @@ class SkyPilotReplicaManager(ReplicaManager):
         # provision) or the cluster is preempted and cleaned up by the status
         # refresh. In this case, we skip spawning a new down process to save
         # controller resources.
-        if global_user_state.get_cluster_from_name(info.cluster_name) is None:
+        if not global_user_state.cluster_with_name_exists(info.cluster_name):
             self._handle_sky_down_finish(info, exitcode=0)
             return
 

sky/serve/serve_utils.py

@@ -262,7 +262,7 @@ def _validate_consolidation_mode_config(current_is_consolidation_mode: bool,
     controller = controller_utils.get_controller_for_pool(pool).value
     if current_is_consolidation_mode:
         controller_cn = controller.cluster_name
-        if global_user_state.get_cluster_from_name(controller_cn) is not None:
+        if global_user_state.cluster_with_name_exists(controller_cn):
             with ux_utils.print_exception_no_traceback():
                 raise exceptions.InconsistentConsolidationModeError(
                     f'{colorama.Fore.RED}Consolidation mode for '
@@ -896,8 +896,8 @@ def _terminate_failed_services(
     # replicas, so we don't need to try again here.
     for replica_info in serve_state.get_replica_infos(service_name):
         # TODO(tian): Refresh latest status of the cluster.
-        if global_user_state.get_cluster_from_name(
-                replica_info.cluster_name) is not None:
+        if global_user_state.cluster_with_name_exists(
+                replica_info.cluster_name):
             remaining_replica_clusters.append(f'{replica_info.cluster_name!r}')
         serve_state.remove_replica(service_name, replica_info.replica_id)
 
@@ -1133,10 +1133,8 @@ def _process_line(line: str,
     # `✓ Cluster launched: new-http.  View logs at: *.log`
     # We should tail the detailed logs for user.
     def cluster_is_up() -> bool:
-        cluster_record = global_user_state.get_cluster_from_name(cluster_name)
-        if cluster_record is None:
-            return False
-        return cluster_record['status'] == status_lib.ClusterStatus.UP
+        status = global_user_state.get_status_from_cluster_name(cluster_name)
+        return status == status_lib.ClusterStatus.UP
 
     provision_api_log_prompt = re.match(_SKYPILOT_PROVISION_API_LOG_PATTERN,
                                         line)

sky/serve/server/core.py

@@ -46,20 +46,23 @@ def up(
 
 
 @usage_lib.entrypoint
-def update(
-        task: 'sky.Task',
-        service_name: str,
-        mode: serve_utils.UpdateMode = serve_utils.DEFAULT_UPDATE_MODE) -> None:
+def update(task: Optional['sky.Task'],
+           service_name: str,
+           mode: serve_utils.UpdateMode = serve_utils.DEFAULT_UPDATE_MODE,
+           workers: Optional[int] = None) -> None:
     """Updates an existing service.
 
     Please refer to the sky.cli.serve_update for the document.
 
     Args:
-        task: sky.Task to update.
+        task: sky.Task to update, or None if updating
+            the number of workers/replicas.
         service_name: Name of the service.
         mode: Update mode.
+        workers: Number of workers/replicas to set for the service when
+            task is None.
     """
-    return impl.update(task, service_name, mode, pool=False)
+    return impl.update(task, service_name, mode, pool=False, workers=workers)
 
 
 @usage_lib.entrypoint

sky/serve/server/impl.py

@@ -411,6 +411,9 @@ def up(
                 f'\n{ux_utils.INDENT_LAST_SYMBOL}To terminate the pool:\t'
                 f'{ux_utils.BOLD}sky jobs pool down {service_name}'
                 f'{ux_utils.RESET_BOLD}'
+                f'\n{ux_utils.INDENT_SYMBOL}To update the number of workers:\t'
+                f'{ux_utils.BOLD}sky jobs pool apply --pool {service_name} '
+                f'--workers 5{ux_utils.RESET_BOLD}'
                 '\n\n' + ux_utils.finishing_message('Successfully created pool '
                                                     f'{service_name!r}.'))
         else:
@@ -448,37 +451,15 @@ def up(
 
 
 def update(
-    task: 'task_lib.Task',
+    task: Optional['task_lib.Task'],
     service_name: str,
     mode: serve_utils.UpdateMode = serve_utils.DEFAULT_UPDATE_MODE,
     pool: bool = False,
+    workers: Optional[int] = None,
 ) -> None:
     """Updates an existing service or pool."""
     noun = 'pool' if pool else 'service'
     capnoun = noun.capitalize()
-    task.validate()
-    serve_utils.validate_service_task(task, pool=pool)
-
-    # Always apply the policy again here, even though it might have been applied
-    # in the CLI. This is to ensure that we apply the policy to the final DAG
-    # and get the mutated config.
-    # TODO(cblmemo,zhwu): If a user sets a new skypilot_config, the update
-    # will not apply the config.
-    dag, _ = admin_policy_utils.apply(task)
-    task = dag.tasks[0]
-    if pool:
-        if task.run is not None:
-            logger.warning(f'{colorama.Fore.YELLOW}The `run` section will be '
-                           f'ignored for pool.{colorama.Style.RESET_ALL}')
-        # Use dummy run script for cluster pool.
-        task.run = serve_constants.POOL_DUMMY_RUN_COMMAND
-
-    assert task.service is not None
-    if not pool and task.service.tls_credential is not None:
-        logger.warning('Updating TLS keyfile and certfile is not supported. '
-                       'Any updates to the keyfile and certfile will not take '
-                       'effect. To update TLS keyfile and certfile, please '
-                       'tear down the service and spin up a new one.')
 
     controller_type = controller_utils.get_controller_for_pool(pool)
     handle = backend_utils.is_controller_accessible(
@@ -505,6 +486,77 @@ def update(
                                f'To spin up a {noun}, use {ux_utils.BOLD}'
                                f'{cmd}{ux_utils.RESET_BOLD}')
 
+    # If task is None and workers is specified, load existing configuration
+    # and update replica count.
+    if task is None:
+        if workers is None:
+            with ux_utils.print_exception_no_traceback():
+                raise ValueError(
+                    f'Cannot update {noun} without specifying '
+                    f'task or workers. Please provide either a task '
+                    f'or specify the number of workers.')
+
+        if not pool:
+            with ux_utils.print_exception_no_traceback():
+                raise ValueError(
+                    'Non-pool service, trying to update replicas to '
+                    f'{workers} is not supported. Ignoring the update.')
+
+        # Load the existing task configuration from the service's YAML file
+        latest_yaml_path = serve_utils.generate_task_yaml_file_name(
+            service_name, service_record['version'], expand_user=False)
+
+        logger.debug('Loading existing task configuration from '
+                     f'{latest_yaml_path} to create a new modified task.')
+
+        # Get the path locally.
+        with tempfile.NamedTemporaryFile(
+                prefix=f'service-task-{service_name}-',
+                mode='w',
+        ) as service_file:
+            try:
+                backend.download_file(handle, latest_yaml_path,
+                                      service_file.name)
+            except exceptions.CommandError as e:
+                raise RuntimeError(
+                    f'Failed to download the old task configuration from '
+                    f'{latest_yaml_path}: {e.error_msg}') from e
+
+            # Load the existing task configuration
+            existing_config = yaml_utils.read_yaml(service_file.name)
+            task = task_lib.Task.from_yaml_config(existing_config)
+
+        if task.service is None:
+            with ux_utils.print_exception_no_traceback():
+                raise RuntimeError('No service configuration found in '
+                                   f'existing {noun} {service_name!r}')
+        task.set_service(task.service.copy(min_replicas=workers))
+
+    task.validate()
+    serve_utils.validate_service_task(task, pool=pool)
+
+    # Now apply the policy and handle task-specific logic
+    # Always apply the policy again here, even though it might have been applied
+    # in the CLI. This is to ensure that we apply the policy to the final DAG
+    # and get the mutated config.
+    # TODO(cblmemo,zhwu): If a user sets a new skypilot_config, the update
+    # will not apply the config.
+    dag, _ = admin_policy_utils.apply(task)
+    task = dag.tasks[0]
+    if pool:
+        if task.run is not None:
+            logger.warning(f'{colorama.Fore.YELLOW}The `run` section will be '
+                           f'ignored for pool.{colorama.Style.RESET_ALL}')
+        # Use dummy run script for cluster pool.
+        task.run = serve_constants.POOL_DUMMY_RUN_COMMAND
+
+    assert task.service is not None
+    if not pool and task.service.tls_credential is not None:
+        logger.warning('Updating TLS keyfile and certfile is not supported. '
+                       'Any updates to the keyfile and certfile will not take '
+                       'effect. To update TLS keyfile and certfile, please '
+                       'tear down the service and spin up a new one.')
+
     prompt = None
     if (service_record['status'] == serve_state.ServiceStatus.CONTROLLER_FAILED
        ):
@@ -625,6 +677,7 @@ def update(
 
 def apply(
     task: 'task_lib.Task',
+    workers: Optional[int],
     service_name: str,
     mode: serve_utils.UpdateMode = serve_utils.DEFAULT_UPDATE_MODE,
     pool: bool = False,
@@ -640,7 +693,7 @@ def apply(
             service_record = _get_service_record(service_name, pool, handle,
                                                  backend)
             if service_record is not None:
-                return update(task, service_name, mode, pool)
+                return update(task, service_name, mode, pool, workers)
         except exceptions.ClusterNotUpError:
             pass
         up(task, service_name, pool)

sky/serve/server/server.py

@@ -98,7 +98,7 @@ async def tail_logs(
     request: fastapi.Request, log_body: payloads.ServeLogsBody,
     background_tasks: fastapi.BackgroundTasks
 ) -> fastapi.responses.StreamingResponse:
-    executor.schedule_request(
+    request_task = executor.prepare_request(
         request_id=request.state.request_id,
         request_name='serve.logs',
         request_body=log_body,
@@ -106,10 +106,9 @@ async def tail_logs(
         schedule_type=api_requests.ScheduleType.SHORT,
         request_cluster_name=common.SKY_SERVE_CONTROLLER_NAME,
     )
-
-    request_task = await api_requests.get_request_async(request.state.request_id
-                                                       )
-
+    task = executor.execute_request_in_coroutine(request_task)
+    # Cancel the coroutine after the request is done or client disconnects
+    background_tasks.add_task(task.cancel)
     return stream_utils.stream_response(
         request_id=request_task.request_id,
         logs_path=request_task.log_path,

sky/serve/service_spec.py

@@ -506,3 +506,36 @@ class SkyServiceSpec:
         if not hasattr(self, '_pool'):
             return False
         return bool(self._pool)
+
+    def copy(self, **override) -> 'SkyServiceSpec':
+        return SkyServiceSpec(
+            readiness_path=override.pop('readiness_path', self._readiness_path),
+            initial_delay_seconds=override.pop('initial_delay_seconds',
+                                               self._initial_delay_seconds),
+            readiness_timeout_seconds=override.pop(
+                'readiness_timeout_seconds', self._readiness_timeout_seconds),
+            min_replicas=override.pop('min_replicas', self._min_replicas),
+            max_replicas=override.pop('max_replicas', self._max_replicas),
+            num_overprovision=override.pop('num_overprovision',
+                                           self._num_overprovision),
+            ports=override.pop('ports', self._ports),
+            target_qps_per_replica=override.pop('target_qps_per_replica',
+                                                self._target_qps_per_replica),
+            post_data=override.pop('post_data', self._post_data),
+            tls_credential=override.pop('tls_credential', self._tls_credential),
+            readiness_headers=override.pop('readiness_headers',
+                                           self._readiness_headers),
+            dynamic_ondemand_fallback=override.pop(
+                'dynamic_ondemand_fallback', self._dynamic_ondemand_fallback),
+            base_ondemand_fallback_replicas=override.pop(
+                'base_ondemand_fallback_replicas',
+                self._base_ondemand_fallback_replicas),
+            spot_placer=override.pop('spot_placer', self._spot_placer),
+            upscale_delay_seconds=override.pop('upscale_delay_seconds',
+                                               self._upscale_delay_seconds),
+            downscale_delay_seconds=override.pop('downscale_delay_seconds',
+                                                 self._downscale_delay_seconds),
+            load_balancing_policy=override.pop('load_balancing_policy',
+                                               self._load_balancing_policy),
+            pool=override.pop('pool', self._pool),
+        )

sky/server/constants.py

@@ -10,7 +10,7 @@ from sky.skylet import constants
 # based on version info is needed.
 # For more details and code guidelines, refer to:
 # https://docs.skypilot.co/en/latest/developers/CONTRIBUTING.html#backward-compatibility-guidelines
-API_VERSION = 18
+API_VERSION = 20
 
 # The minimum peer API version that the code should still work with.
 # Notes (dev):

sky/server/daemons.py

@@ -8,7 +8,6 @@ from sky import sky_logging
 from sky import skypilot_config
 from sky.server import constants as server_constants
 from sky.utils import annotations
-from sky.utils import common
 from sky.utils import common_utils
 from sky.utils import env_options
 from sky.utils import subprocess_utils
@@ -94,13 +93,13 @@ class InternalRequestDaemon:
 def refresh_cluster_status_event():
     """Periodically refresh the cluster status."""
     # pylint: disable=import-outside-toplevel
-    from sky import core
+    from sky.backends import backend_utils
 
     logger.info('=== Refreshing cluster status ===')
     # This periodically refresh will hold the lock for the cluster being
     # refreshed, but it is OK because other operations will just wait for
     # the lock and get the just refreshed status without refreshing again.
-    core.status(refresh=common.StatusRefreshMode.FORCE, all_users=True)
+    backend_utils.refresh_cluster_records()
     logger.info('Status refreshed. Sleeping '
                 f'{server_constants.CLUSTER_REFRESH_DAEMON_INTERVAL_SECONDS}'
                 ' seconds for the next refresh...\n')

sky/server/requests/executor.py

@@ -502,7 +502,35 @@ def _record_memory_metrics(request_name: str, proc: psutil.Process,
         name=request_name).observe(max(peak_rss - rss_begin, 0))
 
 
-async def execute_request_coroutine(request: api_requests.Request):
+class CoroutineTask:
+    """Wrapper of a background task runs in coroutine"""
+
+    def __init__(self, task: asyncio.Task):
+        self.task = task
+
+    async def cancel(self):
+        try:
+            self.task.cancel()
+            await self.task
+        except asyncio.CancelledError:
+            pass
+
+
+def execute_request_in_coroutine(
+        request: api_requests.Request) -> CoroutineTask:
+    """Execute a request in current event loop.
+
+    Args:
+        request: The request to execute.
+
+    Returns:
+        A CoroutineTask handle to operate the background task.
+    """
+    task = asyncio.create_task(_execute_request_coroutine(request))
+    return CoroutineTask(task)
+
+
+async def _execute_request_coroutine(request: api_requests.Request):
     """Execute a request in current event loop.
 
     Similar to _request_execution_wrapper, but executed as coroutine in current
@@ -640,13 +668,35 @@ def schedule_request(request_id: str,
             The precondition is waited asynchronously and does not block the
             caller.
     """
-    prepare_request(request_id, request_name, request_body, func,
-                    request_cluster_name, schedule_type, is_skypilot_system)
+    request_task = prepare_request(request_id, request_name, request_body, func,
+                                   request_cluster_name, schedule_type,
+                                   is_skypilot_system)
+    schedule_prepared_request(request_task, ignore_return_value, precondition,
+                              retryable)
+
+
+def schedule_prepared_request(request_task: api_requests.Request,
+                              ignore_return_value: bool = False,
+                              precondition: Optional[
+                                  preconditions.Precondition] = None,
+                              retryable: bool = False) -> None:
+    """Enqueue a request to the request queue
+
+    Args:
+        request_task: The prepared request task to schedule.
+        ignore_return_value: If True, the return value of the function will be
+            ignored.
+        precondition: If a precondition is provided, the request will only be
+            scheduled for execution when the precondition is met (returns True).
+            The precondition is waited asynchronously and does not block the
+            caller.
+        retryable: Whether the request should be retried if it fails.
+    """
 
     def enqueue():
-        input_tuple = (request_id, ignore_return_value, retryable)
-        logger.info(f'Queuing request: {request_id}')
-        _get_queue(schedule_type).put(input_tuple)
+        input_tuple = (request_task.request_id, ignore_return_value, retryable)
+        logger.info(f'Queuing request: {request_task.request_id}')
+        _get_queue(request_task.schedule_type).put(input_tuple)
 
     if precondition is not None:
         # Wait async to avoid blocking caller.

sky/server/requests/payloads.py

@@ -316,6 +316,9 @@ class StatusBody(RequestBody):
     all_users: bool = True
     # TODO (kyuds): default to False post 0.10.5
     include_credentials: bool = True
+    # Only return fields that are needed for the
+    # dashboard / CLI summary response
+    summary_response: bool = False
 
 
 class StartBody(RequestBody):
@@ -475,6 +478,17 @@ class VolumeListBody(RequestBody):
     pass
 
 
+class VolumeValidateBody(RequestBody):
+    """The request body for the volume validate endpoint."""
+    name: Optional[str] = None
+    volume_type: Optional[str] = None
+    infra: Optional[str] = None
+    size: Optional[str] = None
+    labels: Optional[Dict[str, str]] = None
+    resource_name: Optional[str] = None
+    config: Optional[Dict[str, Any]] = None
+
+
 class EndpointsBody(RequestBody):
     """The request body for the endpoint."""
     cluster: str
@@ -669,9 +683,15 @@ class LocalUpBody(RequestBody):
     ssh_key: Optional[str] = None
     cleanup: bool = False
     context_name: Optional[str] = None
+    name: Optional[str] = None
     password: Optional[str] = None
 
 
+class LocalDownBody(RequestBody):
+    """The request body for the local down endpoint."""
+    name: Optional[str] = None
+
+
 class SSHUpBody(RequestBody):
     """The request body for the SSH up/down endpoints."""
     infra: Optional[str] = None
@@ -709,19 +729,22 @@ class JobsDownloadLogsBody(RequestBody):
 
 class JobsPoolApplyBody(RequestBody):
     """The request body for the jobs pool apply endpoint."""
-    task: str
+    task: Optional[str] = None
+    workers: Optional[int] = None
     pool_name: str
     mode: serve.UpdateMode
 
     def to_kwargs(self) -> Dict[str, Any]:
         kwargs = super().to_kwargs()
-        dag = common.process_mounts_in_task_on_api_server(self.task,
-                                                          self.env_vars,
-                                                          workdir_only=False)
-        assert len(
-            dag.tasks) == 1, ('Must only specify one task in the DAG for '
-                              'a pool.', dag)
-        kwargs['task'] = dag.tasks[0]
+        if self.task is not None:
+            dag = common.process_mounts_in_task_on_api_server(
+                self.task, self.env_vars, workdir_only=False)
+            assert len(
+                dag.tasks) == 1, ('Must only specify one task in the DAG for '
+                                  'a pool.', dag)
+            kwargs['task'] = dag.tasks[0]
+        else:
+            kwargs['task'] = None
         return kwargs