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

Re: [Condor-users] Limit Number of Parallel Jobs



Sassy Natan <sassyn@xxxxxxxxx> wrote:
> MY QUESTION IS when a job become preemption? does RANK never impact
> the status of the job?

When a job is idle, the condor_negotiator will consider taking a
computer from a different running job if the idle job has a better
user priority, or the computer more highly RANKs the idle job.
The flow of logic is summarized here:
http://www.cs.wisc.edu/condor/manual/v7.6/3_4User_Priorities.html#SECTION00445000000000000000

> Say I have all machines configure with a RANK for a specific user X,
> and I send 1000 job to the pool (which have only 100 slots) from user
> Y, once one of the jobs own by user Y will finished a job from user X
> will start up? (based on the fact the have the same priority)?
>
> What I would like to have is when user X submit a job, one of the Y
> job get checkpointed and job X start immediate.

Condor should evict user Y's jobs (including giving them a chance
to checkpoint) and replace them with user X's jobs.  Thus, you
should get the exact behavior you are looking for.

Unfortunately there is a bug in Condor from roughly 7.3.2 through
7.4.4 that means that RANK expressions aren't quite working how
they should.  We believe we've fixed it for the upcoming 7.4.5
release.  For more details on the bug, you can see
https://condor-wiki.cs.wisc.edu/index.cgi/tktview?tn=2275

-- 
Alan De Smet                              Condor Project Research
adesmet@xxxxxxxxxxx                http://www.cs.wisc.edu/condor/