Issue with Vnode and Queue association

Users/Site Administrators
1

I’m currently facing an issue where jobs submitted to specific queues (e.g., workq) are being scheduled on Vnodes that shouldn’t be associated with those queues, despite my attempt to restrict them using the queue_list resource.

I have configured pbs with the following configurations:

sched_config:

[foo]# cat ./sched_priv/sched_config | grep queue_list
resources: "ncpus, mem, arch, host, vnode, aoe, eoe, ngpus, queue_list, scratch, nodetype"

resource:

[foo]# qmgr -c 'p r queue_list'
create resource queue_list
set resource queue_list type = string_array
set resource queue_list flag = h

queue parameters:

[foo]# qmgr -c 'p q @def' | grep queue_list
set queue workq resources_default.queue_list = workq
set queue debug resources_default.queue_list = debug
set queue longq resources_default.queue_list = longq
set queue prioq resources_default.queue_list = prioq

set queue workq default_chunk.queue_list = workq
set queue debug default_chunk.queue_list = debug
set queue longq default_chunk.queue_list = longq
set queue prioq default_chunk.queue_list = prioq

vnode parameters:

[foo]# qmgr -c 'p n @def' | grep queue_list
set node cnode01 resources_available.queue_list = workq
set node cnode01 resources_available.queue_list += debug
set node cnode01 resources_available.queue_list += longq
set node cnode01[0] resources_available.queue_list = workq
set node cnode01[0] resources_available.queue_list += debug
set node cnode01[0] resources_available.queue_list += longq
set node cnode01[1] resources_available.queue_list = workq
set node cnode01[1] resources_available.queue_list += debug
set node cnode01[1] resources_available.queue_list += longq
set node cnode02 resources_available.queue_list = workq
set node cnode02 resources_available.queue_list += debug
set node cnode02 resources_available.queue_list += longq
set node cnode02[0] resources_available.queue_list = workq
set node cnode02[0] resources_available.queue_list += debug
set node cnode02[0] resources_available.queue_list += longq
set node cnode02[1] resources_available.queue_list = workq
set node cnode02[1] resources_available.queue_list += debug
set node cnode02[1] resources_available.queue_list += longq
set node anode01 resources_available.queue_list = prioq
set node anode01[0] resources_available.queue_list = prioq
set node anode01[1] resources_available.queue_list = prioq

I followed section 4.9.2.2 of the official documentation (“Example of Associating Multiple Vnodes with Multiple Queues”) to implement this setup.

Despite the configuration above, jobs submitted to workq are still being scheduled on anode01 (which is only associated with prioq). It seems that the queue_list restrictions are being ignored by the scheduler.

I would really appreciate any insight into what might be going wrong.
Is there something I’m missing or misconfiguring?

Thank you in advance for your support and guidance!

2

You might have missed these steps. Please check.

source /etc/pbs.conf
cd $PBS_HOME/sched_priv/
edit sched_config file , find the **resources:** like below and add queue_list to the end as below

resources: "ncpus, mem, arch, host, vnode, aoe, eoe, ngpus, queue_list"

save and exit the sched_config

* kill -HUP <PID of the pbs_sched>

and try again

3

Hi @adarsh

Thanks for your reply.

The sched_config has been updated to include queue_list (as shown in the first command).
I’ve already restarted and reloaded the scheduler several times, but jobs are still being scheduled on anode01.

My main concern is this:
Are jobs that were submitted before modifying resources_default.queue_list and default_chunk.queue_list affected by the changes?
Is it possible that these jobs ignore the updated configuration?

4

Please unset these and try again.