skypilot-nightly 1.0.0.dev20250925__py3-none-any.whl → 1.0.0.dev20250927__py3-none-any.whl

sky/jobs/server/core.py

@@ -19,7 +19,9 @@ from sky import provision as provision_lib
 from sky import sky_logging
 from sky import skypilot_config
 from sky import task as task_lib
+from sky.adaptors import common as adaptors_common
 from sky.backends import backend_utils
+from sky.backends import cloud_vm_ray_backend
 from sky.catalog import common as service_catalog_common
 from sky.data import storage as storage_lib
 from sky.jobs import constants as managed_job_constants
@@ -44,8 +46,15 @@ from sky.utils import ux_utils
 from sky.workspaces import core as workspaces_core
 
 if typing.TYPE_CHECKING:
+    from google.protobuf import json_format
+
     import sky
-    from sky.backends import cloud_vm_ray_backend
+    from sky.schemas.generated import managed_jobsv1_pb2
+else:
+    json_format = adaptors_common.LazyImport('google.protobuf.json_format')
+
+    managed_jobsv1_pb2 = adaptors_common.LazyImport(
+        'sky.schemas.generated.managed_jobsv1_pb2')
 
 logger = sky_logging.init_logger(__name__)
 
@@ -368,6 +377,8 @@ def launch(
                 'priority': priority,
                 'consolidation_mode_job_id': consolidation_mode_job_id,
                 'pool': pool,
+                'job_controller_indicator_file':
+                    managed_job_constants.JOB_CONTROLLER_INDICATOR_FILE,
                 **controller_utils.shared_controller_vars_to_fill(
                     controller,
                     remote_user_config_path=remote_user_config_path,
@@ -699,11 +710,13 @@ def queue_v2(
     assert isinstance(backend, backends.CloudVmRayBackend)
 
     user_hashes: Optional[List[Optional[str]]] = None
+    show_jobs_without_user_hash = False
     if not all_users:
         user_hashes = [common_utils.get_user_hash()]
         # For backwards compatibility, we show jobs that do not have a
         # user_hash. TODO(cooperc): Remove before 0.12.0.
         user_hashes.append(None)
+        show_jobs_without_user_hash = True
     elif user_match is not None:
         users = global_user_state.get_user_by_name_match(user_match)
         if not users:
@@ -711,6 +724,38 @@ def queue_v2(
         user_hashes = [user.id for user in users]
 
     accessible_workspaces = list(workspaces_core.get_workspaces().keys())
+
+    if handle.is_grpc_enabled_with_flag:
+        try:
+            request = managed_jobsv1_pb2.GetJobTableRequest(
+                skip_finished=skip_finished,
+                accessible_workspaces=accessible_workspaces,
+                job_ids=managed_jobsv1_pb2.JobIds(
+                    ids=job_ids) if job_ids is not None else None,
+                workspace_match=workspace_match,
+                name_match=name_match,
+                pool_match=pool_match,
+                page=page,
+                limit=limit,
+                # Remove None from user_hashes, as the gRPC server uses the
+                # show_jobs_without_user_hash flag instead.
+                user_hashes=managed_jobsv1_pb2.UserHashes(hashes=[
+                    user_hash for user_hash in user_hashes
+                    if user_hash is not None
+                ]) if user_hashes is not None else None,
+                statuses=managed_jobsv1_pb2.Statuses(
+                    statuses=statuses) if statuses is not None else None,
+                show_jobs_without_user_hash=show_jobs_without_user_hash,
+            )
+            response = backend_utils.invoke_skylet_with_retries(
+                lambda: cloud_vm_ray_backend.SkyletClient(
+                    handle.get_grpc_channel()).get_managed_job_table(request))
+            jobs = managed_job_utils.decode_managed_job_protos(response.jobs)
+            return jobs, response.total, dict(
+                response.status_counts), response.total_no_filter
+        except exceptions.SkyletMethodNotImplementedError:
+            pass
+
     code = managed_job_utils.ManagedJobCodeGen.get_job_table(
         skip_finished, accessible_workspaces, job_ids, workspace_match,
         name_match, pool_match, page, limit, user_hashes, statuses)
@@ -817,33 +862,60 @@ def cancel(name: Optional[str] = None,
                     'Can only specify one of JOB_IDS, name, pool, or all/'
                     f'all_users. Provided {" ".join(arguments)!r}.')
 
+        job_ids = None if (all_users or all) else job_ids
+
         backend = backend_utils.get_backend_from_handle(handle)
         assert isinstance(backend, backends.CloudVmRayBackend)
-        if all_users:
-            code = managed_job_utils.ManagedJobCodeGen.cancel_jobs_by_id(
-                None, all_users=True)
-        elif all:
-            code = managed_job_utils.ManagedJobCodeGen.cancel_jobs_by_id(None)
-        elif job_ids:
-            code = managed_job_utils.ManagedJobCodeGen.cancel_jobs_by_id(
-                job_ids)
-        elif name is not None:
-            code = managed_job_utils.ManagedJobCodeGen.cancel_job_by_name(name)
-        else:
-            assert pool is not None, (job_ids, name, pool, all)
-            code = managed_job_utils.ManagedJobCodeGen.cancel_jobs_by_pool(pool)
-        # The stderr is redirected to stdout
-        returncode, stdout, stderr = backend.run_on_head(handle,
-                                                         code,
-                                                         require_outputs=True,
-                                                         stream_logs=False)
-        try:
-            subprocess_utils.handle_returncode(returncode, code,
-                                               'Failed to cancel managed job',
-                                               stdout + stderr)
-        except exceptions.CommandError as e:
-            with ux_utils.print_exception_no_traceback():
-                raise RuntimeError(e.error_msg) from e
+
+        use_legacy = not handle.is_grpc_enabled_with_flag
+
+        if not use_legacy:
+            current_workspace = skypilot_config.get_active_workspace()
+            try:
+                request = managed_jobsv1_pb2.CancelJobsRequest(
+                    current_workspace=current_workspace)
+
+                if all_users or all or job_ids:
+                    request.all_users = all_users
+                    if all:
+                        request.user_hash = common_utils.get_user_hash()
+                    if job_ids is not None:
+                        request.job_ids.CopyFrom(
+                            managed_jobsv1_pb2.JobIds(ids=job_ids))
+                elif name is not None:
+                    request.job_name = name
+                else:
+                    assert pool is not None, (job_ids, name, pool, all)
+                    request.pool_name = pool
+
+                response = backend_utils.invoke_skylet_with_retries(
+                    lambda: cloud_vm_ray_backend.SkyletClient(
+                        handle.get_grpc_channel()).cancel_managed_jobs(request))
+                stdout = response.message
+            except exceptions.SkyletMethodNotImplementedError:
+                use_legacy = True
+
+        if use_legacy:
+            if all_users or all or job_ids:
+                code = managed_job_utils.ManagedJobCodeGen.cancel_jobs_by_id(
+                    job_ids, all_users=all_users)
+            elif name is not None:
+                code = managed_job_utils.ManagedJobCodeGen.cancel_job_by_name(
+                    name)
+            else:
+                assert pool is not None, (job_ids, name, pool, all)
+                code = managed_job_utils.ManagedJobCodeGen.cancel_jobs_by_pool(
+                    pool)
+            # The stderr is redirected to stdout
+            returncode, stdout, stderr = backend.run_on_head(
+                handle, code, require_outputs=True, stream_logs=False)
+            try:
+                subprocess_utils.handle_returncode(
+                    returncode, code, 'Failed to cancel managed job',
+                    stdout + stderr)
+            except exceptions.CommandError as e:
+                with ux_utils.print_exception_no_traceback():
+                    raise RuntimeError(e.error_msg) from e
 
         logger.info(stdout)
         if 'Multiple jobs found with name' in stdout:

sky/jobs/server/utils.py

@@ -1,13 +1,24 @@
 """Utility functions for managed jobs."""
+import typing
+
 from sky import backends
+from sky import exceptions
 from sky import sky_logging
+from sky.adaptors import common as adaptors_common
 from sky.backends import backend_utils
+from sky.backends import cloud_vm_ray_backend
 from sky.jobs import utils as managed_job_utils
 from sky.skylet import constants as skylet_constants
 from sky.utils import controller_utils
 
 logger = sky_logging.init_logger(__name__)
 
+if typing.TYPE_CHECKING:
+    from sky.schemas.generated import managed_jobsv1_pb2
+else:
+    managed_jobsv1_pb2 = adaptors_common.LazyImport(
+        'sky.schemas.generated.managed_jobsv1_pb2')
+
 
 def check_version_mismatch_and_non_terminal_jobs() -> None:
     """Check if controller has version mismatch and non-terminal jobs exist.
@@ -28,42 +39,64 @@ def check_version_mismatch_and_non_terminal_jobs() -> None:
     backend = backend_utils.get_backend_from_handle(handle)
     assert isinstance(backend, backends.CloudVmRayBackend)
 
-    # Get controller version and raw job table
-    code = managed_job_utils.ManagedJobCodeGen.get_version_and_job_table()
-
-    returncode, output, stderr = backend.run_on_head(handle,
-                                                     code,
-                                                     require_outputs=True,
-                                                     stream_logs=False,
-                                                     separate_stderr=True)
-
-    if returncode != 0:
-        logger.error(output + stderr)
-        raise ValueError('Failed to check controller version and jobs with '
-                         f'returncode: {returncode}.\n{output + stderr}')
-
-    # Parse the output to extract controller version (split only on first
-    # newline)
-    output_parts = output.strip().split('\n', 1)
-
-    # Extract controller version from first line
-    if len(output_parts) < 2 or not output_parts[0].startswith(
-            'controller_version:'):
-        raise ValueError(
-            f'Expected controller version in first line, got: {output}')
-
-    controller_version = output_parts[0].split(':', 1)[1]
-
-    # Rest is job table payload (preserving any newlines within it)
-    job_table_payload = output_parts[1]
+    use_legacy = not handle.is_grpc_enabled_with_flag
+
+    if not use_legacy:
+        try:
+            version_request = managed_jobsv1_pb2.GetVersionRequest()
+            version_response = backend_utils.invoke_skylet_with_retries(
+                lambda: cloud_vm_ray_backend.SkyletClient(
+                    handle.get_grpc_channel(
+                    )).get_managed_job_controller_version(version_request))
+            controller_version = version_response.controller_version
+
+            job_table_request = managed_jobsv1_pb2.GetJobTableRequest()
+            job_table_response = backend_utils.invoke_skylet_with_retries(
+                lambda: cloud_vm_ray_backend.SkyletClient(
+                    handle.get_grpc_channel()).get_managed_job_table(
+                        job_table_request))
+            jobs = managed_job_utils.decode_managed_job_protos(
+                job_table_response.jobs)
+        except exceptions.SkyletMethodNotImplementedError:
+            use_legacy = True
+
+    if use_legacy:
+        # Get controller version and raw job table
+        code = managed_job_utils.ManagedJobCodeGen.get_version_and_job_table()
+
+        returncode, output, stderr = backend.run_on_head(handle,
+                                                         code,
+                                                         require_outputs=True,
+                                                         stream_logs=False,
+                                                         separate_stderr=True)
+
+        if returncode != 0:
+            logger.error(output + stderr)
+            raise ValueError('Failed to check controller version and jobs with '
+                             f'returncode: {returncode}.\n{output + stderr}')
+
+        # Parse the output to extract controller version (split only on first
+        # newline)
+        output_parts = output.strip().split('\n', 1)
+
+        # Extract controller version from first line
+        if len(output_parts) < 2 or not output_parts[0].startswith(
+                'controller_version:'):
+            raise ValueError(
+                f'Expected controller version in first line, got: {output}')
+
+        controller_version = output_parts[0].split(':', 1)[1]
+
+        # Rest is job table payload (preserving any newlines within it)
+        job_table_payload = output_parts[1]
+
+        # Load and filter jobs locally using existing method
+        jobs, _, _, _, _ = managed_job_utils.load_managed_job_queue(
+            job_table_payload)
 
     # Process locally: check version match and filter non-terminal jobs
     version_matches = (controller_version == local_version or
                        int(controller_version) > 17)
-
-    # Load and filter jobs locally using existing method
-    jobs, _, _, _, _ = managed_job_utils.load_managed_job_queue(
-        job_table_payload)
     non_terminal_jobs = [job for job in jobs if not job['status'].is_terminal()]
     has_non_terminal_jobs = len(non_terminal_jobs) > 0
 

sky/jobs/state.py

@@ -25,6 +25,7 @@ from sqlalchemy.ext import declarative
 from sky import exceptions
 from sky import sky_logging
 from sky import skypilot_config
+from sky.adaptors import common as adaptors_common
 from sky.skylet import constants
 from sky.utils import common_utils
 from sky.utils import context_utils
@@ -34,6 +35,11 @@ from sky.utils.db import migration_utils
 if typing.TYPE_CHECKING:
     from sqlalchemy.engine import row
 
+    from sky.schemas.generated import managed_jobsv1_pb2
+else:
+    managed_jobsv1_pb2 = adaptors_common.LazyImport(
+        'sky.schemas.generated.managed_jobsv1_pb2')
+
 # Separate callback types for sync and async contexts
 SyncCallbackType = Callable[[str], None]
 AsyncCallbackType = Callable[[str], Awaitable[Any]]
@@ -448,6 +454,75 @@ class ManagedJobStatus(enum.Enum):
             cls.RECOVERING,
         ]
 
+    @classmethod
+    def from_protobuf(
+        cls, protobuf_value: 'managed_jobsv1_pb2.ManagedJobStatus'
+    ) -> Optional['ManagedJobStatus']:
+        """Convert protobuf ManagedJobStatus enum to Python enum value."""
+        protobuf_to_enum = {
+            managed_jobsv1_pb2.MANAGED_JOB_STATUS_UNSPECIFIED: None,
+            managed_jobsv1_pb2.MANAGED_JOB_STATUS_PENDING: cls.PENDING,
+            managed_jobsv1_pb2.MANAGED_JOB_STATUS_SUBMITTED:
+                cls.DEPRECATED_SUBMITTED,
+            managed_jobsv1_pb2.MANAGED_JOB_STATUS_STARTING: cls.STARTING,
+            managed_jobsv1_pb2.MANAGED_JOB_STATUS_RUNNING: cls.RUNNING,
+            managed_jobsv1_pb2.MANAGED_JOB_STATUS_SUCCEEDED: cls.SUCCEEDED,
+            managed_jobsv1_pb2.MANAGED_JOB_STATUS_FAILED: cls.FAILED,
+            managed_jobsv1_pb2.MANAGED_JOB_STATUS_FAILED_CONTROLLER:
+                cls.FAILED_CONTROLLER,
+            managed_jobsv1_pb2.MANAGED_JOB_STATUS_FAILED_SETUP:
+                cls.FAILED_SETUP,
+            managed_jobsv1_pb2.MANAGED_JOB_STATUS_CANCELLED: cls.CANCELLED,
+            managed_jobsv1_pb2.MANAGED_JOB_STATUS_RECOVERING: cls.RECOVERING,
+            managed_jobsv1_pb2.MANAGED_JOB_STATUS_CANCELLING: cls.CANCELLING,
+            managed_jobsv1_pb2.MANAGED_JOB_STATUS_FAILED_PRECHECKS:
+                cls.FAILED_PRECHECKS,
+            managed_jobsv1_pb2.MANAGED_JOB_STATUS_FAILED_NO_RESOURCE:
+                cls.FAILED_NO_RESOURCE,
+        }
+
+        if protobuf_value not in protobuf_to_enum:
+            raise ValueError(
+                f'Unknown protobuf ManagedJobStatus value: {protobuf_value}')
+
+        return protobuf_to_enum[protobuf_value]
+
+    def to_protobuf(self) -> 'managed_jobsv1_pb2.ManagedJobStatus':
+        """Convert this Python enum value to protobuf enum value."""
+        enum_to_protobuf = {
+            ManagedJobStatus.PENDING:
+                managed_jobsv1_pb2.MANAGED_JOB_STATUS_PENDING,
+            ManagedJobStatus.DEPRECATED_SUBMITTED:
+                managed_jobsv1_pb2.MANAGED_JOB_STATUS_SUBMITTED,
+            ManagedJobStatus.STARTING:
+                managed_jobsv1_pb2.MANAGED_JOB_STATUS_STARTING,
+            ManagedJobStatus.RUNNING:
+                managed_jobsv1_pb2.MANAGED_JOB_STATUS_RUNNING,
+            ManagedJobStatus.SUCCEEDED:
+                managed_jobsv1_pb2.MANAGED_JOB_STATUS_SUCCEEDED,
+            ManagedJobStatus.FAILED:
+                managed_jobsv1_pb2.MANAGED_JOB_STATUS_FAILED,
+            ManagedJobStatus.FAILED_CONTROLLER:
+                managed_jobsv1_pb2.MANAGED_JOB_STATUS_FAILED_CONTROLLER,
+            ManagedJobStatus.FAILED_SETUP:
+                managed_jobsv1_pb2.MANAGED_JOB_STATUS_FAILED_SETUP,
+            ManagedJobStatus.CANCELLED:
+                managed_jobsv1_pb2.MANAGED_JOB_STATUS_CANCELLED,
+            ManagedJobStatus.RECOVERING:
+                managed_jobsv1_pb2.MANAGED_JOB_STATUS_RECOVERING,
+            ManagedJobStatus.CANCELLING:
+                managed_jobsv1_pb2.MANAGED_JOB_STATUS_CANCELLING,
+            ManagedJobStatus.FAILED_PRECHECKS:
+                managed_jobsv1_pb2.MANAGED_JOB_STATUS_FAILED_PRECHECKS,
+            ManagedJobStatus.FAILED_NO_RESOURCE:
+                managed_jobsv1_pb2.MANAGED_JOB_STATUS_FAILED_NO_RESOURCE,
+        }
+
+        if self not in enum_to_protobuf:
+            raise ValueError(f'Unknown ManagedJobStatus value: {self}')
+
+        return enum_to_protobuf[self]
+
 
 _SPOT_STATUS_TO_COLOR = {
     ManagedJobStatus.PENDING: colorama.Fore.BLUE,
@@ -537,6 +612,60 @@ class ManagedJobScheduleState(enum.Enum):
     # The job is in a terminal state. (Not necessarily SUCCEEDED.)
     DONE = 'DONE'
 
+    @classmethod
+    def from_protobuf(
+        cls, protobuf_value: 'managed_jobsv1_pb2.ManagedJobScheduleState'
+    ) -> Optional['ManagedJobScheduleState']:
+        """Convert protobuf ManagedJobScheduleState enum to Python enum value.
+        """
+        protobuf_to_enum = {
+            managed_jobsv1_pb2.MANAGED_JOB_SCHEDULE_STATE_UNSPECIFIED: None,
+            managed_jobsv1_pb2.MANAGED_JOB_SCHEDULE_STATE_INVALID: cls.INVALID,
+            managed_jobsv1_pb2.MANAGED_JOB_SCHEDULE_STATE_INACTIVE:
+                cls.INACTIVE,
+            managed_jobsv1_pb2.MANAGED_JOB_SCHEDULE_STATE_WAITING: cls.WAITING,
+            managed_jobsv1_pb2.MANAGED_JOB_SCHEDULE_STATE_ALIVE_WAITING:
+                cls.ALIVE_WAITING,
+            managed_jobsv1_pb2.MANAGED_JOB_SCHEDULE_STATE_LAUNCHING:
+                cls.LAUNCHING,
+            managed_jobsv1_pb2.MANAGED_JOB_SCHEDULE_STATE_ALIVE_BACKOFF:
+                cls.ALIVE_BACKOFF,
+            managed_jobsv1_pb2.MANAGED_JOB_SCHEDULE_STATE_ALIVE: cls.ALIVE,
+            managed_jobsv1_pb2.MANAGED_JOB_SCHEDULE_STATE_DONE: cls.DONE,
+        }
+
+        if protobuf_value not in protobuf_to_enum:
+            raise ValueError('Unknown protobuf ManagedJobScheduleState value: '
+                             f'{protobuf_value}')
+
+        return protobuf_to_enum[protobuf_value]
+
+    def to_protobuf(self) -> 'managed_jobsv1_pb2.ManagedJobScheduleState':
+        """Convert this Python enum value to protobuf enum value."""
+        enum_to_protobuf = {
+            ManagedJobScheduleState.INVALID:
+                managed_jobsv1_pb2.MANAGED_JOB_SCHEDULE_STATE_INVALID,
+            ManagedJobScheduleState.INACTIVE:
+                managed_jobsv1_pb2.MANAGED_JOB_SCHEDULE_STATE_INACTIVE,
+            ManagedJobScheduleState.WAITING:
+                managed_jobsv1_pb2.MANAGED_JOB_SCHEDULE_STATE_WAITING,
+            ManagedJobScheduleState.ALIVE_WAITING:
+                managed_jobsv1_pb2.MANAGED_JOB_SCHEDULE_STATE_ALIVE_WAITING,
+            ManagedJobScheduleState.LAUNCHING:
+                managed_jobsv1_pb2.MANAGED_JOB_SCHEDULE_STATE_LAUNCHING,
+            ManagedJobScheduleState.ALIVE_BACKOFF:
+                managed_jobsv1_pb2.MANAGED_JOB_SCHEDULE_STATE_ALIVE_BACKOFF,
+            ManagedJobScheduleState.ALIVE:
+                managed_jobsv1_pb2.MANAGED_JOB_SCHEDULE_STATE_ALIVE,
+            ManagedJobScheduleState.DONE:
+                managed_jobsv1_pb2.MANAGED_JOB_SCHEDULE_STATE_DONE,
+        }
+
+        if self not in enum_to_protobuf:
+            raise ValueError(f'Unknown ManagedJobScheduleState value: {self}')
+
+        return enum_to_protobuf[self]
+
 
 # === Status transition functions ===
 @_init_db
@@ -792,8 +921,14 @@ def set_local_log_file(job_id: int, task_id: Optional[int],
 # ======== utility functions ========
 @_init_db
 def get_nonterminal_job_ids_by_name(name: Optional[str],
+                                    user_hash: Optional[str] = None,
                                     all_users: bool = False) -> List[int]:
-    """Get non-terminal job ids by name."""
+    """Get non-terminal job ids by name.
+
+    If name is None:
+    1. if all_users is False, get for the given user_hash
+    2. otherwise, get for all users
+    """
     assert _SQLALCHEMY_ENGINE is not None
 
     with orm.Session(_SQLALCHEMY_ENGINE) as session:
@@ -810,8 +945,15 @@ def get_nonterminal_job_ids_by_name(name: Optional[str],
             ])
         ]
         if name is None and not all_users:
-            where_conditions.append(
-                job_info_table.c.user_hash == common_utils.get_user_hash())
+            if user_hash is None:
+                # For backwards compatibility. With codegen, USER_ID_ENV_VAR
+                # was set to the correct value by the jobs controller, as
+                # part of ManagedJobCodeGen._build(). This is no longer the
+                # case for the Skylet gRPC server, which is why we need to
+                # pass it explicitly through the request body.
+                logger.debug('user_hash is None, using current user hash')
+                user_hash = common_utils.get_user_hash()
+            where_conditions.append(job_info_table.c.user_hash == user_hash)
         if name is not None:
             # We match the job name from `job_info` for the jobs submitted after
             # #1982, and from `spot` for the jobs submitted before #1982, whose

sky/jobs/utils.py

@@ -16,8 +16,8 @@ import textwrap
 import time
 import traceback
 import typing
-from typing import (Any, Deque, Dict, List, Literal, Optional, Set, TextIO,
-                    Tuple, Union)
+from typing import (Any, Deque, Dict, Iterable, List, Literal, Optional, Set,
+                    TextIO, Tuple, Union)
 
 import colorama
 import filelock
@@ -51,16 +51,23 @@ from sky.utils import subprocess_utils
 from sky.utils import ux_utils
 
 if typing.TYPE_CHECKING:
+    from google.protobuf import descriptor
+    from google.protobuf import json_format
     import grpc
     import psutil
 
     import sky
     from sky import dag as dag_lib
     from sky.schemas.generated import jobsv1_pb2
+    from sky.schemas.generated import managed_jobsv1_pb2
 else:
+    json_format = adaptors_common.LazyImport('google.protobuf.json_format')
+    descriptor = adaptors_common.LazyImport('google.protobuf.descriptor')
     psutil = adaptors_common.LazyImport('psutil')
     grpc = adaptors_common.LazyImport('grpc')
     jobsv1_pb2 = adaptors_common.LazyImport('sky.schemas.generated.jobsv1_pb2')
+    managed_jobsv1_pb2 = adaptors_common.LazyImport(
+        'sky.schemas.generated.managed_jobsv1_pb2')
 
 logger = sky_logging.init_logger(__name__)
 
@@ -169,7 +176,7 @@ def _validate_consolidation_mode_config(
         if all_jobs:
             nonterminal_jobs = (
                 managed_job_state.get_nonterminal_job_ids_by_name(
-                    None, all_users=True))
+                    None, None, all_users=True))
             if nonterminal_jobs:
                 with ux_utils.print_exception_no_traceback():
                     raise exceptions.InconsistentConsolidationModeError(
@@ -698,14 +705,15 @@ def generate_managed_job_cluster_name(task_name: str, job_id: int) -> str:
 
 def cancel_jobs_by_id(job_ids: Optional[List[int]],
                       all_users: bool = False,
-                      current_workspace: Optional[str] = None) -> str:
+                      current_workspace: Optional[str] = None,
+                      user_hash: Optional[str] = None) -> str:
     """Cancel jobs by id.
 
     If job_ids is None, cancel all jobs.
     """
     if job_ids is None:
         job_ids = managed_job_state.get_nonterminal_job_ids_by_name(
-            None, all_users)
+            None, user_hash, all_users)
     job_ids = list(set(job_ids))
     if not job_ids:
         return 'No job to cancel.'
@@ -1241,6 +1249,24 @@ def dump_managed_job_queue(
     user_hashes: Optional[List[Optional[str]]] = None,
     statuses: Optional[List[str]] = None,
 ) -> str:
+    return message_utils.encode_payload(
+        get_managed_job_queue(skip_finished, accessible_workspaces, job_ids,
+                              workspace_match, name_match, pool_match, page,
+                              limit, user_hashes, statuses))
+
+
+def get_managed_job_queue(
+    skip_finished: bool = False,
+    accessible_workspaces: Optional[List[str]] = None,
+    job_ids: Optional[List[int]] = None,
+    workspace_match: Optional[str] = None,
+    name_match: Optional[str] = None,
+    pool_match: Optional[str] = None,
+    page: Optional[int] = None,
+    limit: Optional[int] = None,
+    user_hashes: Optional[List[Optional[str]]] = None,
+    statuses: Optional[List[str]] = None,
+) -> Dict[str, Any]:
     # Make sure to get all jobs - some logic below (e.g. high priority job
     # detection) requires a full view of the jobs table.
     jobs = managed_job_state.get_managed_jobs()
@@ -1371,12 +1397,12 @@ def dump_managed_job_queue(
         else:
             job['details'] = None
 
-    return message_utils.encode_payload({
+    return {
         'jobs': jobs,
         'total': total,
         'total_no_filter': total_no_filter,
         'status_counts': status_counts
-    })
+    }
 
 
 def filter_jobs(
@@ -1824,6 +1850,58 @@ def format_job_table(
     return output
 
 
+def decode_managed_job_protos(
+    job_protos: Iterable['managed_jobsv1_pb2.ManagedJobInfo']
+) -> List[Dict[str, Any]]:
+    """Decode job protos to dicts. Similar to load_managed_job_queue."""
+    user_hash_to_user = global_user_state.get_users(
+        set(job.user_hash for job in job_protos if job.user_hash))
+
+    jobs = []
+    for job_proto in job_protos:
+        job_dict = _job_proto_to_dict(job_proto)
+        user_hash = job_dict.get('user_hash', None)
+        if user_hash is not None:
+            # Skip jobs that do not have user_hash info.
+            # TODO(cooperc): Remove check before 0.12.0.
+            user = user_hash_to_user.get(user_hash, None)
+            job_dict['user_name'] = user.name if user is not None else None
+        jobs.append(job_dict)
+    return jobs
+
+
+def _job_proto_to_dict(
+        job_proto: 'managed_jobsv1_pb2.ManagedJobInfo') -> Dict[str, Any]:
+    job_dict = json_format.MessageToDict(
+        job_proto,
+        always_print_fields_with_no_presence=True,
+        # Our API returns fields in snake_case.
+        preserving_proto_field_name=True,
+        use_integers_for_enums=True)
+    for field in job_proto.DESCRIPTOR.fields:
+        # Ensure optional fields are present with None values for
+        # backwards compatibility with older clients.
+        if field.has_presence and field.name not in job_dict:
+            job_dict[field.name] = None
+        # json_format.MessageToDict is meant for encoding to JSON,
+        # and Protobuf encodes int64 as decimal strings in JSON,
+        # so we need to convert them back to ints.
+        # https://protobuf.dev/programming-guides/json/#field-representation
+        if field.type == descriptor.FieldDescriptor.TYPE_INT64:
+            job_dict[field.name] = int(job_dict[field.name])
+    job_dict['status'] = managed_job_state.ManagedJobStatus.from_protobuf(
+        job_dict['status'])
+    # For backwards compatibility, convert schedule_state to a string,
+    # as we don't have the logic to handle it in our request
+    # encoder/decoder, unlike status.
+    schedule_state_enum = (
+        managed_job_state.ManagedJobScheduleState.from_protobuf(
+            job_dict['schedule_state']))
+    job_dict['schedule_state'] = (schedule_state_enum.value
+                                  if schedule_state_enum is not None else None)
+    return job_dict
+
+
 class ManagedJobCodeGen:
     """Code generator for managed job utility functions.
 

sky/provision/runpod/__init__.py

@@ -11,4 +11,6 @@ from sky.provision.runpod.instance import terminate_instances
 from sky.provision.runpod.instance import wait_instances
 from sky.provision.runpod.volume import apply_volume
 from sky.provision.runpod.volume import delete_volume
+from sky.provision.runpod.volume import get_all_volumes_usedby
 from sky.provision.runpod.volume import get_volume_usedby
+from sky.provision.runpod.volume import map_all_volumes_usedby