Automated scheduling

By default, torc leaves scheduling of compute nodes to the user. If you know that an initially-blocked job will need a specific compute node (or nodes), you can tell torc to schedule it for you when all other conditions are met.

Define the schedule_compute_nodes object in the JobModel. For example,

from torc.openapi_client import ComputeNodeScheduleParams, JobModel, SlurmSchedulerModel

short_scheduler = api.add_slurm_scheduler(
    workflow.key,
    SlurmSchedulerModel(
        name="short",
        account="my_account",
        nodes=1,
        walltime="04:00:00",
    ),
)
job = JobModel(
    name="job1",
    command=f"python job.py",
    resource_requirements=medium.id,
    schedule_compute_nodes=ComputeNodeScheduleParams(
        num_jobs=1,
        scheduler=short_scheduler.id,
    )
),

When that job reaches the ready status, torc will send the schedule command with those parameters.

Note

If one new compute node allocation can satisfy multiple jobs that will be ready at about the same time, you can set these fields for only one job. Setting it for multiple jobs may result in extra, unnecessary allocations.