When plugin gets configured with idleMinutes == 0, not only the agents get deleted sooner then they are launched turning the setup practically useless as reported in
. It even seem to cause the provisioning for serviced labels to stuck almost completely. Here is what I observed is happening: JENKINS-47953
- Agents was created and launching yet disappearing instantly.
- The plugin was logging "No such container" 3 times in a row, roughly for every botched node.
- Both stopped to happen after changing idleMinutes from 0 to 10.
- I presume the
kicked in deleting the container while provisioning was in progress. JENKINS-47953
- Eventually, and presumably because of this, all provisioning stopped with multiple pending launches that never completes that are not done nor cancelled and yet they do not have a running thread in stacktrace. This is causing the plannedCapacity > demand so nothing else is provisioned.
- I admit a do not quite understand how did the futures get in such state but this too stopped occurring right after fixing idleMinutes and cancelling dangling futures.
Having said that, I suggest to make the 0 timeout unsupported and use some sane defaults even when configured explicitly (this can either happen during migration or manually by user not quite aware of this surprising consequences). An alternative would be ensuring the slave will only be disposed after launched/used.