Skip to content

DragonLauncher sub-launchers #593

Open
@al-rigazzi

Description

@al-rigazzi

Description

The steps belonging to the DragonLauncher attributes _pbs_launcher and _slurm_launcher should be managed in a better way.

Justification

Currently, when a DragonBatchStep is launched as a batch job using SbatchSettings or QsubSettings, a SlurmLauncher or PBSLauncher object is used to control it. Such launchers are stored as attributes of the DragonLauncher, and we call them sub-launchers. As SmartSim tracks jobs only by the ID assigned by the WLM to them, the DragonLauncher has no way of understanding whether a job id represents a Slurm or a PBS batch job. For that reason, we prefix the job id with the name of the sub-launcher. This workaround is not elegant and we should find a better way of handling such batch jobs.

Implementation Strategy

Instead of step id (integers or strings), SmartSim could use objects to track steps. For example, a _get_managed_step_update call, instead of using a list of ids, would use a list of StepID objects (these could be simple dataclass objects), and no id prefixing would be needed by the Dragon launcher, which could deduce the sub-launcher to call from the object type.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    area: Dragonarea: launcherIssues related to any of the launchers within SmartSimtype: featureIssues that include feature request or feature idea

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions