In a big cluster with hundreds are users, not all of them would submit
their job to specific queues.
And creating hundreds are pool is not very easy. Having something like
userMaxMaps would limit users from consuming all resources.
We could create small set of queues with more weight, but JobTracker would
still waste time in preemption of user tasks who has consumed all resources.
It would have been nice to have such a config.
2012/10/18 Sandy Ryza <[EMAIL PROTECTED]>
> You're correct that there's no way to put a hard limit on the number of
> maps or reduces for a given user, and a user can potentially consume all of
> the cluster resources. However, if there are multiple users contending for
> resources, the scheduler makes an effort to schedule tasks equally, so it
> would be unlikely for a single user to get that much of the cluster.
> Can I ask what you need a userMaxMaps/Reducers for?
> On Thu, Oct 18, 2012 at 4:41 PM, lohit <[EMAIL PROTECTED]> wrote:
> > I am trying to understand FairScheduler configs I am trying to see if
> > is a way to achieve the below.
> > I see that if there are no pools configured (or only few pools are
> > configured ) and a user submits a job, it would end up in his own pool,
> > right?
> > Now, I see there are some limits you can set globally for such users, for
> > example userMaxJobsDefault.
> > Is there a way to set userMaxMaps or userMaxReducers? It looks like if I
> > have few pools configured and a user who submits a job without specify a
> > pool will be given his own pool. He can potentially consume 100% of
> > Map/Reduce slots. Is my understand correct?
> > --
> > Have a Nice Day!
> > Lohit
Have a Nice Day!