[Condor-devel] KILLING_TIMEOUT in 7.6.0

[Dan Bradley <dan@xxxxxxxxxxxx>]

Hi all,

I'm wondering if the change in behavior of preemption in 7.5 was really 
intended to be as broad as it is.  Now, the KILL expression is useless 
by default, because when a job is preempted, it is hard-killed as soon 
as KILLING_TIMEOUT expires.  The default KILLING_TIMEOUT is just 30s.  
We've received complaints in CHTC from users who have self-checkpointing 
jobs that cannot save state in this amount of time.  I will crank the 
knob higher, but I consider this a workaround, not a resolution.

This new behavior came from the following:

https://condor-wiki.cs.wisc.edu/index.cgi/tktview?tn=1198

Prior to this, KILLING_TIMEOUT was the timeout applied to the 
Preempting/Killing activity.  Now, it effectively applies to 
Preempting/Vacating as well.

I see no release notes warning admins of this important change.

--Dan