nvflare.app_opt.job_launcher.slurm.manager module

Slurm submission, monitoring, and cancellation.

class SlurmJobHandle(manager: SlurmJobManager, nvflare_job_id: str, job_name: str, job_dir: str, job_id: str, pending_timeout: float, poll_interval: float)[source]

Bases: JobHandleSpec

Current-parent state and lifecycle driver for one Slurm submission.

poll()[source]

To get the return code of the job run.

Returns: return_code

terminate()[source]

To terminate the job run.

Returns: None

wait()[source]

To wait until the job run complete.

Returns: returns until the job run complete.

class SlurmJobManager(config: ~nvflare.app_opt.job_launcher.slurm.config.SlurmConfig, logger: ~logging.Logger, adapter: ~nvflare.app_opt.job_launcher.slurm.scheduler_client._SlurmCliAdapter | None = None, monotonic_clock: ~typing.Callable[[], float] = <built-in function monotonic>)[source]

Bases: object

Own scheduler access and live job handles for one NVFlare parent.

initialize() None[source]
launch(plan: LaunchPlan) SlurmJobHandle[source]
shutdown() None[source]