Configuring autostop for jobs

Jobs can be configured to stop the node they are running on after completion, or after a period of inactivitity (note that this is still in beta).

This can help reduce cost from idle-running instances, or when jobs run over a longer period of time.

Example: Configuring autostop

Autostop is configured under the job attribute in wisp-spec.yml:

job:
  autostop:
    enabled: True
    timeout_minutes: 10

With this configuration, the job will automatically notify Wisp when it's completed by success or failure, and Wisp will stop the instance. You will no longer be billed when instances are stopped.

Additionally, Wisp will monitor CPU and GPU utilization of the instance, and shut it down if it has been idle for longer than 10 minutes.

To resume your work, and submit a new job to the instance, you need to start it again.