Job states and lifetime
The states
Section titled “The states”| State | Meaning |
|---|---|
| booked | Accepted and assigned a slot. It has a node and a start time. |
| queued | Waiting for capacity. No slot yet. |
| running | The container is up. Tokens are being spent. |
| finished | Ran to completion, or the runtime expired. |
| failed | Stopped because something went wrong - usually the image or the command. |
| canceled | You or an administrator stopped it. |
finished, failed and canceled are terminal. Nothing moves out of them; a
new run means a new job, which is what Clone and Requeue do.
The normal path
Section titled “The normal path”A job you submit with Start ASAP, onto a queue with a free node, goes
booked → running → finished. The list refreshes itself, so you can watch it.
If the queue is busy it waits at queued until something frees up. The
Available in line in the submission summary told you to expect this before
you committed.
Where a job’s time goes
Section titled “Where a job’s time goes”The runtime is a ceiling, not a target. The job is billed for the minutes it actually holds resources, so cancelling a job you have finished with stops the spend - it does not waste the rest.
When the runtime expires, the container is stopped and removed.
Extending
Section titled “Extending”While a job is running you can extend it:
- Extend all on the job row adds the same amount of time to every service.
- Extend in a service’s detail panel extends that service.
The default offered is 60 minutes. Extending is only possible on a running job
- once it has expired there is nothing to extend.
Cancelling
Section titled “Cancelling”Cancel job on the row, with a confirmation. The action then disables itself so it cannot be run twice. Cancelling is the correct way to stop paying for something you are done with.
Selecting several jobs offers a bulk cancel, which skips any that are already in a terminal state and tells you how many it will actually affect.
Restarting and requeueing
Section titled “Restarting and requeueing”Restart, from the detail panel of a running or failed job, starts its services again. You can restart a single service instead of the whole job.
Restart is not offered on a finished or cancelled job. For those, Clone (one job, opens the form) or Requeue (a selection, resubmits directly).
Editing a job in place
Section titled “Editing a job in place”A job that is booked, queued, running or failed can be edited from its detail panel: job name, image, command and ports. Saving with an empty image is refused.
Anything you did not touch - registry, mounts - is kept. Editing is not offered on a finished or cancelled job, because the backend will not accept it.