Hi Jeff...
Not sure if I really understood your point.
If the same user is always mapped to the same unix account,
independently of the CE, the batch system should enforce the limit based
on that unix account information. In this sense, CREAM should be
agnostic to this process. If that limit is really enforced, CREAM should
receive an error from BLAH at qsub time.
Did I missed something?
Cheers
Goncalo
On 03/20/2014 03:23 PM, Jeff Templon wrote:
> Hi *
>
> I wonder if the collective experience here has a solution for the
> following problem.
>
> We have some user groups that are fond of submitting (without looking)
> thousands of jobs. While Torque can handle many thousands, at some
> point it becomes a problem. We solved this by turning on e.g. this:
>
> set queue medium max_user_queuable = 3000
>
> in Torque. Now comes the problem : CREAM appends "@machine" to the
> user name, and even though torque sees that it's the same user
>
> euser = biome053
>
> it seems to use the field "Job_Owner" for determining whether
> max_user_queuable has been reached:
>
> Job_Owner = [log in to unmask]
>
> so if biome053 can submit to three CEs, then they can submit 9000 jobs
> instead. Not all VOs can submit to all CEs ... how have others solved
> this? Is it possible to have CREAM leave off the machine name when
> contstructing the Job_Owner field?
>
> Thanks
>
> J "templon@*.nikhef.nl" T
> l
|