Job containers should be killed when no inner job for some time

[bzizou]

We could implement an optional feature into Leon, to automatically kill job containers that did not have any inner job for a defined timeout.

Containers may be used by users that want to schedule subjobs (for example complementary depopulated jobs, or LHC like "pilot jobs") with the side effect that the end of a container is not dependent of the the end of the inner jobs, wasting resources then...

[npf]

Could be implemented in the meta_sched quite easily I think (same place where oarwalltime is handled): get all running jobs in the container, if latest end time > N minutes in the past, kill container.

My suggestion would be to offer that with a job type container=autokill:N, where N is the number of minutes of grace before the container is to be killed.