snowflake.ml.jobs.MLJob¶

class snowflake.ml.jobs.MLJob(id: str, service_spec: Optional[dict[str, Any]] = None, session: Optional[Session] = None)¶

Bases: Generic[T], SerializableSessionMixin

Methods

cancel() → None¶: Cancel the running job. :raises RuntimeError: If cancellation fails. # noqa: DAR401

distributed_result(timeout: float = - 1) → DistributedResult¶

Block until completion, then return the structured per-instance summary of a distributed run.

This is the result API for distributed jobs (e.g. submitted with parallel=True), which have no single head result. On full success it returns a DistributedResult (success, per-instance exit_codes, failed_instance, instance 0’s return_value). On any instance failure it raises DistributedJobError, which carries that same DistributedResult as .result and the earliest-failing instance’s reconstructed exception as its cause. result() remains the single-head API.

Parameters:

timeout – The maximum time to wait in seconds. Negative values are treated as no timeout.

Returns:

The reduced per-instance result, on full success. # noqa: DAR401

Return type:

DistributedResult

Raises:

NotImplementedError – If the job was not launched with a distributed backend (e.g. parallel=True); single-head jobs have no per-instance result — use result().
DistributedJobError – If any instance did not exit 0. Carries the DistributedResult (.result) with per-instance exit_codes / failed_instance. # noqa: DAR402
RuntimeError – If the job’s per-instance records could not be retrieved.
TimeoutError – If the job does not complete within the specified timeout. # noqa: DAR402

get_logs(limit: int = - 1, instance_id: Optional[int] = None, *, as_list: Literal[True], verbose: bool = constants.DEFAULT_VERBOSE_LOG) → list[str]¶

get_logs(limit: int = - 1, instance_id: Optional[int] = None, *, as_list: Literal[False] = False, verbose: bool = constants.DEFAULT_VERBOSE_LOG) → str

Return the job’s execution logs.

Parameters:

limit – The maximum number of lines to return. Negative values are treated as no limit.
instance_id – Optional instance ID to get logs from a specific instance. If not provided, returns logs from the head node.
as_list – If True, returns logs as a list of lines. Otherwise, returns logs as a single string.
verbose – Whether to return the full log or just the user log.

Returns:

The job’s execution logs.

get_ray_dashboard_url() → Optional[str]¶

Get the Ray dashboard URL for the job.

Returns:

The Ray dashboard URL if the job is running and has a Ray dashboard endpoint,: None otherwise.

Return type:

Optional[str]

result(timeout: float = - 1) → T¶

Block until completion. Returns job execution result.

Parameters:

timeout – The maximum time to wait in seconds. Negative values are treated as no timeout.

Returns:

The deserialized job result. # noqa: DAR401

Return type:

Raises:

RuntimeError – If the job failed or if the job doesn’t have a result to retrieve.
TimeoutError – If the job does not complete within the specified timeout. # noqa: DAR402

show_logs(limit: int = - 1, instance_id: Optional[int] = None, verbose: bool = False) → None¶

Display the job’s execution logs.

Parameters:

limit – The maximum number of lines to display. Negative values are treated as no limit.
instance_id – Optional instance ID to get logs from a specific instance. If not provided, displays logs from the head node.
verbose – Whether to return the full log or just the user log.

wait(timeout: float = - 1) → Literal['PENDING', 'RUNNING', 'FAILED', 'DONE', 'CANCELLING', 'CANCELLED', 'INTERNAL_ERROR', 'DELETED']¶

Block until completion. Returns completion status.

Parameters:: timeout – The maximum time to wait in seconds. Negative values are treated as no timeout.
Returns:: The job’s completion status.
Raises:: TimeoutError – If the job does not complete within the specified timeout.

Attributes

id¶: Get the unique job ID

min_instances¶

name¶

session¶

status¶: Get the job’s execution status.

target_instances¶