[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Condor-users] Jobs being preempted with default rank settings



On Fri, Jul 18, 2008 at 4:41 PM, David Dynerman <dynerman@xxxxxxxxxxx> wrote:
> Matt,
>
> Setting a retirement time of 10 seconds did indeed seem to fix this
> problem. It seems like a bug though, since the retirement time should
> only apply post-claim, not as part of the pre-claim negotiation.

The retirement simply masks the problem (albeit effectively) since the
job seems to get renegotiated to another machine afterwards (the the
brief time in the retiring state doesn't affect the original job)

We haven't played with the POST setting. it would be nice to verify
this so it could be filed as a definite repeatable bug, I simply don't
have the time to set up a test farm (And I can't do it to the
production one).

To be honest unless you really do want to have very low latency on
certain high priority jobs a little retirement is a good thing. If you
submit something and it turns the whole farm into "Retiring" when that
wasn't your intention you have a bit of time to fix this. If
everything goes to "Vacating" it's too late...