Am 12.04.2011 um 14:00 schrieb Dave Love:
> Reuti <[email protected]> writes:
>
>> Am 11.04.2011 um 18:48 schrieb Dave Love:
>>
>>> Doubtless this has come up before, but I can't find a post about it.
>>>
>>> I wanted to limit job runtimes by ACL with RQS, rather than defining a
>>> separate queue for longer jobs for particular users (not that I want
>>> jobs with longer run times...). I find it doesn't work sensibly for
>>> parallel jobs because h_rt is apparently multiplied by the slot count
>>> before being applied (experimentally -- I haven't looked for the
>>> relevant code).
>>
>> When do you think it is multiplied?
>>
>> - when compared to RQS during scheduling?
>> - for the enforced h_rt?
>
> It stops scheduling. (RQS doesn't enforce it in the sense of setrlimit,
> but there's a default specified in our case.)
>
> Specifically, with a rule like this:
>
> limit users {!@longusers} to h_rt=432000
Can you try without the {}, as it's not a consumable it shouldn't be necssary.
-- Reuti
> serial jobs with -l h_rt=432000 will run for all users, but parallel
> ones won't for users not in the ACL (due to that limit according to
> qstat). Two slots with half the h_rt will run.
_______________________________________________
users mailing list
[email protected]
https://gridengine.org/mailman/listinfo/users
