aneil2
July 27, 2023, 12:07am
1
The issue, users submitting multiple jobs and then leaving them idle and never connecting. The jobs continue to consume resources until the user selected time limit - which could be days - expires…
Enable a configurable countdown timer on the connection link, so that unless the link is clicked within a given grace period the session and job will be cancelled.
Hi, thanks for the comment and sorry for the delay. We’re mostly looking at the Get Help section so this section doesn’t get a lot of attention.
I’ve filed this ticket on github for the same. Don’t know when/if we’ll get around to it, but we’ll track it there.
opened 08:24PM - 25 Jun 24 UTC
community request
component/dashboard
Cleaning out discourse - this has come up a few times:
https://discourse.openo… ndemand.org/t/detect-idle-sessions/741
https://discourse.openondemand.org/t/timeout-count-down-on-connection-link-to-permit-idle-job-timeout-before-the-user-requested-timelimit/2836
At the time of writing, this would be very very difficult. Maybe if we have something proxying for us on the compute node may help, but actually detecting on the web server what's happening on the compute node is an arduous task. Possibly solved through files on the NFS?
Even if we do have a mechanism to communicate this data, the web node may not be able to catch this because itself is being shut down because _it's_ idle. So there's a window where we have to catch this after the job starts, but before the PUN itself is shutdown due to inactivity. I guess that best effort is better than nothing?