skypilot-nightly 1.0.0.dev20250612__py3-none-any.whl → 1.0.0.dev20250614__py3-none-any.whl

sky/jobs/state.py

@@ -161,8 +161,8 @@ def create_table(cursor, conn):
     conn.commit()
 
 
-# Module-level connection/cursor; thread-safe as the module is only imported
-# once.
+# Module-level connection/cursor; thread-safe as the db is initialized once
+# across all threads.
 def _get_db_path() -> str:
     """Workaround to collapse multi-step Path ops for type checker.
     Ensures _DB_PATH is str, avoiding Union[Path, str] inference.
@@ -173,8 +173,7 @@ def _get_db_path() -> str:
     return str(path)
 
 
-_DB_PATH = _get_db_path()
-_db_initialized = False
+_DB_PATH = None
 _db_init_lock = threading.Lock()
 
 
@@ -183,13 +182,13 @@ def _init_db(func):
 
     @functools.wraps(func)
     def wrapper(*args, **kwargs):
-        global _db_initialized
-        if _db_initialized:
+        global _DB_PATH
+        if _DB_PATH is not None:
             return func(*args, **kwargs)
         with _db_init_lock:
-            if not _db_initialized:
+            if _DB_PATH is None:
+                _DB_PATH = _get_db_path()
                 db_utils.SQLiteConn(_DB_PATH, create_table)
-                _db_initialized = True
         return func(*args, **kwargs)
 
     return wrapper
@@ -442,7 +441,7 @@ class ManagedJobScheduleState(enum.Enum):
 # === Status transition functions ===
 @_init_db
 def set_job_info(job_id: int, name: str, workspace: str, entrypoint: str):
-    assert _db_initialized
+    assert _DB_PATH is not None
     with db_utils.safe_cursor(_DB_PATH) as cursor:
         cursor.execute(
             """\
@@ -456,7 +455,7 @@ def set_job_info(job_id: int, name: str, workspace: str, entrypoint: str):
 @_init_db
 def set_pending(job_id: int, task_id: int, task_name: str, resources_str: str):
     """Set the task to pending state."""
-    assert _db_initialized
+    assert _DB_PATH is not None
     with db_utils.safe_cursor(_DB_PATH) as cursor:
         cursor.execute(
             """\
@@ -484,7 +483,7 @@ def set_starting(job_id: int, task_id: int, run_timestamp: str,
         specs: The specs of the managed task.
         callback_func: The callback function.
     """
-    assert _db_initialized
+    assert _DB_PATH is not None
     # Use the timestamp in the `run_timestamp` ('sky-2022-10...'), to make
     # the log directory and submission time align with each other, so as to
     # make it easier to find them based on one of the values.
@@ -524,7 +523,7 @@ def set_backoff_pending(job_id: int, task_id: int):
     This should only be used to transition from STARTING or RECOVERING back to
     PENDING.
     """
-    assert _db_initialized
+    assert _DB_PATH is not None
     with db_utils.safe_cursor(_DB_PATH) as cursor:
         cursor.execute(
             """\
@@ -552,7 +551,7 @@ def set_restarting(job_id: int, task_id: int, recovering: bool):
     after using set_backoff_pending to transition back to PENDING during
     launch retry backoff.
     """
-    assert _db_initialized
+    assert _DB_PATH is not None
     target_status = ManagedJobStatus.STARTING.value
     if recovering:
         target_status = ManagedJobStatus.RECOVERING.value
@@ -578,7 +577,7 @@ def set_restarting(job_id: int, task_id: int, recovering: bool):
 def set_started(job_id: int, task_id: int, start_time: float,
                 callback_func: CallbackType):
     """Set the task to started state."""
-    assert _db_initialized
+    assert _DB_PATH is not None
     logger.info('Job started.')
     with db_utils.safe_cursor(_DB_PATH) as cursor:
         cursor.execute(
@@ -610,7 +609,7 @@ def set_started(job_id: int, task_id: int, start_time: float,
 @_init_db
 def set_recovering(job_id: int, task_id: int, callback_func: CallbackType):
     """Set the task to recovering state, and update the job duration."""
-    assert _db_initialized
+    assert _DB_PATH is not None
     logger.info('=== Recovering... ===')
     with db_utils.safe_cursor(_DB_PATH) as cursor:
         cursor.execute(
@@ -634,7 +633,7 @@ def set_recovering(job_id: int, task_id: int, callback_func: CallbackType):
 def set_recovered(job_id: int, task_id: int, recovered_time: float,
                   callback_func: CallbackType):
     """Set the task to recovered."""
-    assert _db_initialized
+    assert _DB_PATH is not None
     with db_utils.safe_cursor(_DB_PATH) as cursor:
         cursor.execute(
             """\
@@ -658,7 +657,7 @@ def set_recovered(job_id: int, task_id: int, recovered_time: float,
 def set_succeeded(job_id: int, task_id: int, end_time: float,
                   callback_func: CallbackType):
     """Set the task to succeeded, if it is in a non-terminal state."""
-    assert _db_initialized
+    assert _DB_PATH is not None
     with db_utils.safe_cursor(_DB_PATH) as cursor:
         cursor.execute(
             """\
@@ -703,7 +702,7 @@ def set_failed(
         override_terminal: If True, override the current status even if end_at
             is already set.
     """
-    assert _db_initialized
+    assert _DB_PATH is not None
     assert failure_type.is_failed(), failure_type
     end_time = time.time() if end_time is None else end_time
 
@@ -761,7 +760,7 @@ def set_cancelling(job_id: int, callback_func: CallbackType):
     task_id is not needed, because we expect the job should be cancelled
     as a whole, and we should not cancel a single task.
     """
-    assert _db_initialized
+    assert _DB_PATH is not None
     with db_utils.safe_cursor(_DB_PATH) as cursor:
         rows = cursor.execute(
             """\
@@ -783,7 +782,7 @@ def set_cancelled(job_id: int, callback_func: CallbackType):
 
     The set_cancelling should be called before this function.
     """
-    assert _db_initialized
+    assert _DB_PATH is not None
     with db_utils.safe_cursor(_DB_PATH) as cursor:
         rows = cursor.execute(
             """\
@@ -804,7 +803,7 @@ def set_cancelled(job_id: int, callback_func: CallbackType):
 def set_local_log_file(job_id: int, task_id: Optional[int],
                        local_log_file: str):
     """Set the local log file for a job."""
-    assert _db_initialized
+    assert _DB_PATH is not None
     filter_str = 'spot_job_id=(?)'
     filter_args = [local_log_file, job_id]
 
@@ -822,7 +821,7 @@ def set_local_log_file(job_id: int, task_id: Optional[int],
 def get_nonterminal_job_ids_by_name(name: Optional[str],
                                     all_users: bool = False) -> List[int]:
     """Get non-terminal job ids by name."""
-    assert _db_initialized
+    assert _DB_PATH is not None
     statuses = ', '.join(['?'] * len(ManagedJobStatus.terminal_statuses()))
     field_values = [
         status.value for status in ManagedJobStatus.terminal_statuses()
@@ -866,7 +865,7 @@ def get_schedule_live_jobs(job_id: Optional[int]) -> List[Dict[str, Any]]:
     exception: the job may have just transitioned from WAITING to LAUNCHING, but
     the controller process has not yet started.
     """
-    assert _db_initialized
+    assert _DB_PATH is not None
     job_filter = '' if job_id is None else 'AND spot_job_id=(?)'
     job_value = (job_id,) if job_id is not None else ()
 
@@ -909,7 +908,7 @@ def get_jobs_to_check_status(job_id: Optional[int] = None) -> List[int]:
     - Jobs have schedule_state DONE but are in a non-terminal status
     - Legacy jobs (that is, no schedule state) that are in non-terminal status
     """
-    assert _db_initialized
+    assert _DB_PATH is not None
     job_filter = '' if job_id is None else 'AND spot.spot_job_id=(?)'
     job_value = () if job_id is None else (job_id,)
 
@@ -958,7 +957,7 @@ def get_jobs_to_check_status(job_id: Optional[int] = None) -> List[int]:
 @_init_db
 def get_all_job_ids_by_name(name: Optional[str]) -> List[int]:
     """Get all job ids by name."""
-    assert _db_initialized
+    assert _DB_PATH is not None
     name_filter = ''
     field_values = []
     if name is not None:
@@ -987,7 +986,7 @@ def get_all_job_ids_by_name(name: Optional[str]) -> List[int]:
 @_init_db
 def _get_all_task_ids_statuses(
         job_id: int) -> List[Tuple[int, ManagedJobStatus]]:
-    assert _db_initialized
+    assert _DB_PATH is not None
     with db_utils.safe_cursor(_DB_PATH) as cursor:
         id_statuses = cursor.execute(
             """\
@@ -1035,7 +1034,7 @@ def get_failure_reason(job_id: int) -> Optional[str]:
 
     If the job has multiple tasks, we return the first failure reason.
     """
-    assert _db_initialized
+    assert _DB_PATH is not None
     with db_utils.safe_cursor(_DB_PATH) as cursor:
         reason = cursor.execute(
             """\
@@ -1051,7 +1050,7 @@ def get_failure_reason(job_id: int) -> Optional[str]:
 @_init_db
 def get_managed_jobs(job_id: Optional[int] = None) -> List[Dict[str, Any]]:
     """Get managed jobs from the database."""
-    assert _db_initialized
+    assert _DB_PATH is not None
     job_filter = '' if job_id is None else f'WHERE spot.spot_job_id={job_id}'
 
     # Join spot and job_info tables to get the job name for each task.
@@ -1097,7 +1096,7 @@ def get_managed_jobs(job_id: Optional[int] = None) -> List[Dict[str, Any]]:
 @_init_db
 def get_task_name(job_id: int, task_id: int) -> str:
     """Get the task name of a job."""
-    assert _db_initialized
+    assert _DB_PATH is not None
     with db_utils.safe_cursor(_DB_PATH) as cursor:
         task_name = cursor.execute(
             """\
@@ -1110,7 +1109,7 @@ def get_task_name(job_id: int, task_id: int) -> str:
 @_init_db
 def get_latest_job_id() -> Optional[int]:
     """Get the latest job id."""
-    assert _db_initialized
+    assert _DB_PATH is not None
     with db_utils.safe_cursor(_DB_PATH) as cursor:
         rows = cursor.execute("""\
             SELECT spot_job_id FROM spot
@@ -1123,7 +1122,7 @@ def get_latest_job_id() -> Optional[int]:
 
 @_init_db
 def get_task_specs(job_id: int, task_id: int) -> Dict[str, Any]:
-    assert _db_initialized
+    assert _DB_PATH is not None
     with db_utils.safe_cursor(_DB_PATH) as cursor:
         task_specs = cursor.execute(
             """\
@@ -1136,7 +1135,7 @@ def get_task_specs(job_id: int, task_id: int) -> Dict[str, Any]:
 @_init_db
 def get_local_log_file(job_id: int, task_id: Optional[int]) -> Optional[str]:
     """Get the local log directory for a job."""
-    assert _db_initialized
+    assert _DB_PATH is not None
     filter_str = 'spot_job_id=(?)'
     filter_args = [job_id]
     if task_id is not None:
@@ -1159,7 +1158,7 @@ def scheduler_set_waiting(job_id: int, dag_yaml_path: str,
                           original_user_yaml_path: str, env_file_path: str,
                           user_hash: str, priority: int) -> None:
     """Do not call without holding the scheduler lock."""
-    assert _db_initialized
+    assert _DB_PATH is not None
     with db_utils.safe_cursor(_DB_PATH) as cursor:
         updated_count = cursor.execute(
             'UPDATE job_info SET '
@@ -1177,7 +1176,7 @@ def scheduler_set_waiting(job_id: int, dag_yaml_path: str,
 def scheduler_set_launching(job_id: int,
                             current_state: ManagedJobScheduleState) -> None:
     """Do not call without holding the scheduler lock."""
-    assert _db_initialized
+    assert _DB_PATH is not None
     with db_utils.safe_cursor(_DB_PATH) as cursor:
         updated_count = cursor.execute(
             'UPDATE job_info SET '
@@ -1191,7 +1190,7 @@ def scheduler_set_launching(job_id: int,
 @_init_db
 def scheduler_set_alive(job_id: int) -> None:
     """Do not call without holding the scheduler lock."""
-    assert _db_initialized
+    assert _DB_PATH is not None
     with db_utils.safe_cursor(_DB_PATH) as cursor:
         updated_count = cursor.execute(
             'UPDATE job_info SET '
@@ -1205,7 +1204,7 @@ def scheduler_set_alive(job_id: int) -> None:
 @_init_db
 def scheduler_set_alive_backoff(job_id: int) -> None:
     """Do not call without holding the scheduler lock."""
-    assert _db_initialized
+    assert _DB_PATH is not None
     with db_utils.safe_cursor(_DB_PATH) as cursor:
         updated_count = cursor.execute(
             'UPDATE job_info SET '
@@ -1219,7 +1218,7 @@ def scheduler_set_alive_backoff(job_id: int) -> None:
 @_init_db
 def scheduler_set_alive_waiting(job_id: int) -> None:
     """Do not call without holding the scheduler lock."""
-    assert _db_initialized
+    assert _DB_PATH is not None
     with db_utils.safe_cursor(_DB_PATH) as cursor:
         updated_count = cursor.execute(
             'UPDATE job_info SET '
@@ -1234,7 +1233,7 @@ def scheduler_set_alive_waiting(job_id: int) -> None:
 @_init_db
 def scheduler_set_done(job_id: int, idempotent: bool = False) -> None:
     """Do not call without holding the scheduler lock."""
-    assert _db_initialized
+    assert _DB_PATH is not None
     with db_utils.safe_cursor(_DB_PATH) as cursor:
         updated_count = cursor.execute(
             'UPDATE job_info SET '
@@ -1248,7 +1247,7 @@ def scheduler_set_done(job_id: int, idempotent: bool = False) -> None:
 
 @_init_db
 def set_job_controller_pid(job_id: int, pid: int):
-    assert _db_initialized
+    assert _DB_PATH is not None
     with db_utils.safe_cursor(_DB_PATH) as cursor:
         updated_count = cursor.execute(
             'UPDATE job_info SET '
@@ -1259,7 +1258,7 @@ def set_job_controller_pid(job_id: int, pid: int):
 
 @_init_db
 def get_job_schedule_state(job_id: int) -> ManagedJobScheduleState:
-    assert _db_initialized
+    assert _DB_PATH is not None
     with db_utils.safe_cursor(_DB_PATH) as cursor:
         state = cursor.execute(
             'SELECT schedule_state FROM job_info WHERE spot_job_id = (?)',
@@ -1269,7 +1268,7 @@ def get_job_schedule_state(job_id: int) -> ManagedJobScheduleState:
 
 @_init_db
 def get_num_launching_jobs() -> int:
-    assert _db_initialized
+    assert _DB_PATH is not None
     with db_utils.safe_cursor(_DB_PATH) as cursor:
         return cursor.execute(
             'SELECT COUNT(*) '
@@ -1280,7 +1279,7 @@ def get_num_launching_jobs() -> int:
 
 @_init_db
 def get_num_alive_jobs() -> int:
-    assert _db_initialized
+    assert _DB_PATH is not None
     with db_utils.safe_cursor(_DB_PATH) as cursor:
         return cursor.execute(
             'SELECT COUNT(*) '
@@ -1303,7 +1302,7 @@ def get_waiting_job() -> Optional[Dict[str, Any]]:
     Backwards compatibility note: jobs submitted before #4485 will have no
     schedule_state and will be ignored by this SQL query.
     """
-    assert _db_initialized
+    assert _DB_PATH is not None
     with db_utils.safe_cursor(_DB_PATH) as cursor:
         # Get the highest-priority WAITING or ALIVE_WAITING job whose priority
         # is greater than or equal to the highest priority LAUNCHING or
@@ -1338,7 +1337,7 @@ def get_waiting_job() -> Optional[Dict[str, Any]]:
 @_init_db
 def get_workspace(job_id: int) -> str:
     """Get the workspace of a job."""
-    assert _db_initialized
+    assert _DB_PATH is not None
     with db_utils.safe_cursor(_DB_PATH) as cursor:
         workspace = cursor.execute(
             'SELECT workspace FROM job_info WHERE spot_job_id = (?)',

sky/provision/__init__.py

@@ -18,6 +18,7 @@ from sky.provision import common
 from sky.provision import cudo
 from sky.provision import fluidstack
 from sky.provision import gcp
+from sky.provision import hyperbolic
 from sky.provision import kubernetes
 from sky.provision import lambda_cloud
 from sky.provision import nebius

sky/provision/common.py

@@ -238,7 +238,7 @@ class Endpoint:
 
 @dataclasses.dataclass
 class SocketEndpoint(Endpoint):
-    """Socket endpoint accesible via a host and a port."""
+    """Socket endpoint accessible via a host and a port."""
     port: Optional[int]
     host: str = ''
 

sky/provision/gcp/config.py

@@ -274,7 +274,7 @@ def _is_permission_satisfied(service_account, crm, iam, required_permissions,
     # For example, `roles/iam.serviceAccountUser` can be granted at the
     # skypilot-v1 service account level, which can be checked with
     # service_account_policy = iam.projects().serviceAccounts().getIamPolicy(
-    #    resource=f'projects/{project_id}/serviceAcccounts/{email}').execute()
+    #    resource=f'projects/{project_id}/serviceAccounts/{email}').execute()
     # We now skip the check for `iam.serviceAccounts.actAs` permission for
     # simplicity as it can be granted at the service account level.
     def check_permissions(policy, required_permissions):

sky/provision/hyperbolic/__init__.py

@@ -0,0 +1,11 @@
+"""Hyperbolic provisioner for SkyPilot."""
+
+from sky.provision.hyperbolic.config import bootstrap_instances
+from sky.provision.hyperbolic.instance import cleanup_ports
+from sky.provision.hyperbolic.instance import get_cluster_info
+from sky.provision.hyperbolic.instance import open_ports
+from sky.provision.hyperbolic.instance import query_instances
+from sky.provision.hyperbolic.instance import run_instances
+from sky.provision.hyperbolic.instance import stop_instances
+from sky.provision.hyperbolic.instance import terminate_instances
+from sky.provision.hyperbolic.instance import wait_instances

sky/provision/hyperbolic/config.py

@@ -0,0 +1,10 @@
+"""Hyperbolic Cloud configuration bootstrapping"""
+
+from sky.provision import common
+
+
+def bootstrap_instances(
+        region: str, cluster_name: str,
+        config: common.ProvisionConfig) -> common.ProvisionConfig:
+    del region, cluster_name  # unused
+    return config