Introducing SkyServe

sky/__init__.py

@@ -49,7 +49,6 @@ def get_git_commit():
 from sky.core import download_logs
 from sky.core import job_status
 from sky.core import queue
-from sky.core import serve_down
 from sky.core import spot_cancel
 from sky.core import spot_queue
 from sky.core import spot_status
@@ -65,7 +64,6 @@ def get_git_commit():
 from sky.data import StoreType
 from sky.execution import exec  # pylint: disable=redefined-builtin
 from sky.execution import launch
-from sky.execution import serve_up
 from sky.execution import spot_launch
 from sky.optimizer import Optimizer
 from sky.optimizer import OptimizeTarget
@@ -116,7 +114,6 @@ def get_git_commit():
     'launch',
     'exec',
     'spot_launch',
-    'serve_up',
     # core APIs
     'status',
     'start',
@@ -137,6 +134,4 @@ def get_git_commit():
     # core APIs Storage Management
     'storage_ls',
     'storage_delete',
-    # core APIs Serve Management
-    'serve_down',
 ]

sky/backends/cloud_vm_ray_backend.py

@@ -4917,7 +4917,7 @@ def _execute_task_n_nodes(self, handle: CloudVmRayResourceHandle,
                 env_vars=task.envs,
                 task_name=task.name,
                 job_run_id=job_run_id,
-                ray_resources_dict=resources_dict,
+                ray_resources_dict=backend_utils.get_task_demands_dict(task),
                 log_dir=log_dir,
                 gang_scheduling_id=i,
                 use_sudo=use_sudo,

sky/cli.py

@@ -1692,7 +1692,7 @@ def _get_services(service_names: Optional[List[str]],
             if not service_names:
                 # Change empty list to None
                 service_names = None
-            service_records = core.serve_status(service_names)
+            service_records = sky.serve.status(service_names)
             num_services = len(service_records)
     except exceptions.ClusterNotUpError as e:
         controller_status = e.cluster_status
@@ -1721,9 +1721,9 @@ def _get_services(service_names: Optional[List[str]],
                     f'{service_num} service{plural} found. Please specify '
                     'an existing service to show its endpoint. Usage: '
                     'sky serve status --endpoint <service-name>')
-            msg = status_utils.get_endpoint(service_records[0])
+            msg = serve_lib.get_endpoint(service_records[0])
         else:
-            msg = status_utils.format_service_table(service_records, show_all)
+            msg = serve_lib.format_service_table(service_records, show_all)
             service_not_found_msg = ''
             if service_names is not None:
                 for service_name in service_names:
@@ -2863,7 +2863,7 @@ def _hint_or_raise_for_down_sky_serve_controller(controller_name: str):
         with rich_utils.safe_status(
                 '[bold cyan]Checking for running services[/]'):
             try:
-                services = core.serve_status()
+                services = sky.serve.status()
             except exceptions.ClusterNotUpError:
                 cluster_status = backend_utils.refresh_cluster_status_handle(
                     controller_name)
@@ -4243,7 +4243,6 @@ def serve_up(
 
         sky serve up service.yaml
     """
-    # TODO(tian): Update the example after we move the ports to resources.
     if service_name is None:
         service_name = serve_lib.generate_service_name()
 
@@ -4283,7 +4282,7 @@ def serve_up(
         if prompt is not None:
             click.confirm(prompt, default=True, abort=True, show_default=True)
 
-    sky.serve_up(task, service_name)
+    sky.serve.up(task, service_name)
 
 
 @serve.command('status', cls=_DocumentedCodeCommand)
@@ -4410,14 +4409,19 @@ def serve_status(all: bool, endpoint: bool, service_names: List[str]):
               default=False,
               is_flag=True,
               help='Tear down all services.')
+@click.option('--purge',
+              '-p',
+              default=False,
+              is_flag=True,
+              help='Tear down services in failed status.')
 @click.option('--yes',
               '-y',
               is_flag=True,
               default=False,
               required=False,
               help='Skip confirmation prompt.')
 # pylint: disable=redefined-builtin
-def serve_down(service_names: List[str], all: bool, yes: bool):
+def serve_down(service_names: List[str], all: bool, purge: bool, yes: bool):
     """Teardown service(s).
 
     SERVICE_NAMES is the name of the service (or glob pattern) to tear down. If
@@ -4441,7 +4445,9 @@ def serve_down(service_names: List[str], all: bool, yes: bool):
         \b
         # Tear down all existing services.
         sky serve down -a
-
+        \b
+        # Forcefully tear down a service in failed status.
+        sky serve down failed-service --purge
     """
     if sum([len(service_names) > 0, all]) != 1:
         argument_str = f'SERVICE_NAMES={",".join(service_names)}' if len(
@@ -4468,7 +4474,7 @@ def serve_down(service_names: List[str], all: bool, yes: bool):
                       abort=True,
                       show_default=True)
 
-    sky.serve_down(service_names=service_names, all=all)
+    sky.serve.down(service_names=service_names, all=all, purge=purge)
 
 
 @serve.command('logs', cls=_DocumentedCodeCommand)
@@ -4488,22 +4494,17 @@ def serve_down(service_names: List[str], all: bool, yes: bool):
               default=False,
               required=False,
               help='Show the load balancer logs of this service.')
-@click.option('--target',
-              default=None,
-              type=click.Choice(['controller', 'load-balancer', 'replica'],
-                                case_sensitive=False),
-              required=False,
-              help='Target to stream logs.')
 @click.argument('service_name', required=True, type=str)
 @click.argument('replica_id', required=False, type=int)
 @usage_lib.entrypoint
+# TODO(tian): Add default argument for this CLI if none of the flags are
+# specified.
 def serve_logs(
     service_name: str,
     follow: bool,
     controller: bool,
     load_balancer: bool,
     replica_id: Optional[int],
-    target: Optional[str],
 ):
     """Tail the log of a service.
 
@@ -4512,58 +4513,35 @@ def serve_logs(
     .. code-block:: bash
 
         # Tail the controller logs of a service
-        sky serve logs --controller [SERVICE_ID]
+        sky serve logs --controller [SERVICE_NAME]
         \b
         # Print the load balancer logs so far and exit
-        sky serve logs --load-balancer --no-follow [SERVICE_ID]
+        sky serve logs --load-balancer --no-follow [SERVICE_NAME]
         \b
         # Tail the logs of replica 1
-        sky serve logs [SERVICE_ID] 1
-        \b
-        # Specify target to stream logs by `--target` is also supported
-        sky serve logs --target controller [SERVICE_ID]
-        sky serve logs --target load-balancer [SERVICE_ID]
-        sky serve logs --target replica [SERVICE_ID] 1
-        \b
-        # If both --target and --controller/--load-balancer are specified,
-        # --controller/--load-balancer takes precedence.
-        # Tail the controller logs of a service:
-        sky serve logs --controller --target load-balancer [SERVICE_ID]
+        sky serve logs [SERVICE_NAME] 1
     """
     have_replica_id = replica_id is not None
     num_flags = (controller + load_balancer + have_replica_id)
     if num_flags > 1:
         raise click.UsageError('At most one of --controller, --load-balancer, '
                                '[REPLICA_ID] can be specified.')
-    if num_flags == 0 and target is None:
-        raise click.UsageError(
-            'One of --controller, --load-balancer, [REPLICA_ID] or --target '
-            'must be specified.')
+    if num_flags == 0:
+        raise click.UsageError('One of --controller, --load-balancer, '
+                               '[REPLICA_ID] must be specified.')
     if controller:
-        if target is not None:
-            click.secho(f'Overriding --target={target} with --controller.',
-                        fg='yellow')
         target_component = serve_lib.ServiceComponent.CONTROLLER
     elif load_balancer:
-        if target is not None:
-            click.secho(f'Overriding --target={target} with --load-balancer.',
-                        fg='yellow')
         target_component = serve_lib.ServiceComponent.LOAD_BALANCER
-    elif target is not None:
-        # Change load-balancer to load_balancer to match the enum.
-        target = target.replace('-', '_')
-        target_component = serve_lib.ServiceComponent(target)
-        if (target_component == serve_lib.ServiceComponent.REPLICA and
-                not have_replica_id):
-            raise click.UsageError(
-                'REPLICA_ID must be specified when using --target replica.')
     else:
+        # Already checked that num_flags == 1.
+        assert replica_id is not None
         target_component = serve_lib.ServiceComponent.REPLICA
     try:
-        core.serve_tail_logs(service_name,
-                             target=target_component,
-                             replica_id=replica_id,
-                             follow=follow)
+        sky.serve.tail_logs(service_name,
+                            target=target_component,
+                            replica_id=replica_id,
+                            follow=follow)
     except exceptions.ClusterNotUpError:
         # Hint messages already printed by the call above.
         sys.exit(1)