Hi,
Is there "a submit plugin that sets the QOS to <account><partition>"?
I've setup QoS on partitions as described in the emails below. Currently
I require users to select the correct QoS. I'm interested to learn of a
plugin and/or alternative methods to achieve the same results.
Thank you for any recommendations,
Kevin Abbey
On 04/24/2015 11:52 AM, Jared Casper wrote:
Re: [slurm-dev] Re: Account limits per partition
Thanks Markus!
That's kind of what I had in mind by defining the cross product of
<account>x<partition> as QOS. Good to know it works and very helpful
to have it spelled out like that.
Looks like that's the solution we'll have to use. Thanks again!
Jared
On Fri, Apr 24, 2015 at 5:47 AM, "Dr. Markus Stöhr"
<[email protected] <mailto:[email protected]>> wrote:
Dear Jared,
we are using QOSs to restrict the access of accounts in certain
partitions. The layout is basically:
-define QOS:
sacctmgr -i add qos name=qos_fast_a grpnodes=5 maxwall=3-0:0:0
description='qos a with 5 nodes, 3 days runtime of the jobs and
priority 1000' priority=1000
sacctmgr -i add qos name=qos_fast_b grpnodes=2 maxwall=1-0:0:0
description='qos a with 2 nodes, 1 day runtime of the jobs and
priority 100' priority=100
....
-assign QOS to accounts:
sacctmgr -i add account test_account_a parent=root share=1
qos=normal,qos_fast_a,qos_slow_a defaultqos=qos_fast_a
sacctmgr -i add account test_account_b parent=root share=1
qos=normal,qos_fast_b,qos_slow_b defaultqos=qos_fast_b
-assign QOS to partitions in slurm.conf:
PartitionName=fast .... AllowedQos=qos_fast_a,qos_fast_b
PartitionName=slow .... AllowedQOS=qos_slow_a,qos_slow_b
i hope, this what you are intending.
best regards,
Markus
On 04/23/2015 01:50 AM, Jared Casper wrote:
Hi all,
I basically want to do what this person wanted to do a couple
years ago:
https://groups.google.com/forum/#!topic/slurm-devel/KXx1zKeZL9k
<https://groups.google.com/forum/#%21topic/slurm-devel/KXx1zKeZL9k>
and am
wondering if anything has changed since then? To recap the
request in
that thread to save you the click, I want to set a limit for a
specific
account and specific partition without specifying the user.
So, for
example, I want to limit the account X to using only 2 nodes
in the
partition "super-fast", but allow that account to use 10 nodes
in the
partition "old-and-slow". I also want to allow account Y to
use 5
nodes in "super-fast" and 20 nodes in "old-and-slow".
Looks like associations are still ordered
cluster/account/user/partition
so its not possible to set a limit that applies to a
cluster/account/partition tuple across all users. Have any other
capabilities been added that would help me do what I want
directly? I'm
guessing that hierarchy is fairly engrained in the code and
not easily
worked around? It seems to me like having partition after
user severely
limits the usefulness of being able to specify the partition
since it
can only be used if I want to set limits on a per-user basis.
Am I
missing something?
I'm also not sure I fully understand the solution of using QOS's
proposed in that old thread. It looks like to do what I want
I'd need
to create QOSs that are the cross product of <accounts> and
<partitions>, not just per account or per partition, and do a
submit
plugin that sets the QOS to <account><partition>, then set
limits on all
of those QOSs. Basically using QOSs only instead of associations.
Seems like this is using QOSs for something they weren't
intended for
and not an ideal solution, but might work. Is my
understanding correct?
I haven't looked at using an external scheduler like Maui.
Can anybody
save me the research time and let me know if this is the type
of thing I
might be able to do using Maui? Am I just bumping into the
limits of
what the built-in scheduler can handle and need to bite the
bullet and
move to an external scheduler?
Thanks for any help!
Jared
--
Kevin Abbey
Systems Administrator
Center for Computational and Integrative Biology (CCIB)
http://ccib.camden.rutgers.edu/
Rutgers University - Science Building
315 Penn St.
Camden, NJ 08102
Telephone: (856) 225-6770
Fax:(856) 225-6312
Email: [email protected]
