Background Jobs

What is a Job?

A job is work that should not block the interface. Typical jobs include model training, remote archive requests, large imports and dataset exports.

The platform JobManager runs jobs through a shared thread pool and returns a JobHandle.

Job Status

The JobManager records jobs as queued, running, finished, error or cancelled. Runtime diagnostics can show active jobs and a bounded history of recently completed jobs.

Plugins may expose their own progress UI, but the core job snapshot does not define a generic percentage-progress field.

Cancelling Work

Cancellation is cooperative. JobHandle.cancel() requests cancellation, sets the job’s cancellation token and also tries to cancel the underlying future before it starts.

A running function must check cancel_token.cancelled() at suitable points and return promptly when cancellation is requested.

Caution

Closing a panel does not make arbitrary Python code stop instantly. A panel that owns job handles must request cancellation during disposal where that work should not outlive the panel.

Duplicate Work

A plugin can submit a stable job key. If another submission uses the same key while that job is still active, the JobManager reuses the existing job and can attach the new completion callbacks instead of starting duplicate work.

Stale Results

A completed job may refer to an older record, dataset or request generation. Panels should compare the result with their current request before replacing visible state.