Skip to main content

Configuring job-level timeouts

By default, Dagster jobs have no automatic timeout. A stuck or runaway run can execute indefinitely until it's terminated manually. Job-level timeouts let you set a maximum runtime per job so the system can terminate runs that exceed it.

This is especially useful for:

  • Scheduled jobs that should not overlap with subsequent ticks
  • Jobs that depend on external systems prone to hangs
  • Long-running jobs where you want to bound the worst case

Configuration​

Use the dagster/max_runtime tag on the job (or run) to set the timeout in seconds. The mechanism is the same in Dagster Open Source and Dagster+, but how you enable it differs slightly between each.

Dagster Open Source​

  1. Add the dagster/max_runtime tag (in seconds) to your job definition.

  2. Enable run monitoring in your dagster.yaml:

    run_monitoring:
      enabled: true

  3. Restart your Dagster instance so the daemon picks up the configuration change.

Dagster+​

  1. Add the dagster/max_runtime tag (in seconds) to your job definition. Tags can also be applied at run launch time from the Launchpad or via the Dagster+ deployment settings.

No agent or instance restart is required.

Choosing a timeout value​

Pick a timeout based on observed runtime, with headroom for normal variance:

  • Watch a few weeks of completion times for the job.
  • Set the timeout above the typical p99 runtime, not just the average.
  • Re-evaluate after meaningful changes to data volume, dependencies, or upstream systems.