Re: [HTCondor-users] Condor-CE job routing to same slurm cluster but different partition

[Jaime Frey <jfrey@xxxxxxxxxxx>]

Remove the â_â from âbatch_queueâ in your route description.

Background: Most attribute names in a submit description file correspond to a similar-yet-different name in the ClassAd job description in the queue (viewable via condor_q -long). Often the difference is just the presence of underscores. The job router route operates on the ClassAd description of a submitted job, so BatchQueue (the names are case-insensitive) is the attribute name to use in this case.

 - Jaime

On Feb 10, 2026, at 11:38âAM, Kamrul Shaker <Kamrul.Shaker@xxxxxxxxxxxx> wrote:

Thanksâ.

Route is now working, but it's not able to route to the actual GPU partition in Slurm cluster.

Would that need to be configured from the route or blah configuration?

JOB_ROUTER_ROUTE_Slurm_GPU_Cluster @=jrt

  REQUIREMENTS (RequestGPUs >= 1) || (CUDACapability != UNDEFINED) || (TargetHasGPUs =?= true)

  TargetUniverse = 9

  GridResource = "batch slurm"

  Set batch_queue  = "gpu"

@jrt

Is there any configuration syntax to mention the slurm partition name from the route?

Best Regards,

Kamrul Shaker

Cloud and High-Throughput Computing Specialist â IT4 (AESES)
Deanâs Office, Faculty of Science
University of Manitoba
ð Kamrul.Shaker@xxxxxxxxxxxx| ð +1 431 335 3508
ð Winnipeg, MB, Canada
ð www.umanitoba.ca

 

From: HTCondor-users <htcondor-users-bounces@xxxxxxxxxxx> on behalf of Jaime Frey via HTCondor-users <htcondor-users@xxxxxxxxxxx>
Date: Tuesday, February 10, 2026 at 10:19âAM
To: HTCondor-Users Mail List <htcondor-users@xxxxxxxxxxx>
Cc: Jaime Frey <jfrey@xxxxxxxxxxx>
Subject: Re: [HTCondor-users] Condor-CE job routing to same slurm cluster but different partition

Caution! This message was sent from outside the University of Manitoba.

You need to remove the â=â after REQUIREMENTS in the Slurm_GPU_Cluster route definition.

 - Jaime

On Feb 6, 2026, at 4:02âPM, Kamrul Shaker <Kamrul.Shaker@xxxxxxxxxxxx> wrote:

Hi

Please find my below router configuration :

# /etc/condor-ce/config.d/02-ce-slurm.conf

# --- GPU route: match only GPU-marked jobs, force Slurm 'gpu' partition

JOB_ROUTER_ROUTE_Slurm_GPU_Cluster @=jrt

  # Choose ONE condition that reflects how your submitters mark GPU jobs:

  #   (a) RequestGPUs integer (most common with CE v5+ routing)

  #   (b) CUDACapability (capability code set by some submitters)

  #   (c) A VO/site boolean you add upstream (TargetHasGPUs)

  REQUIREMENTS = (RequestGPUs >= 1) || (CUDACapability != UNDEFINED) || (TargetHasGPUs =?= true)

  TargetUniverse = 9

  GridResource = "batch slurm"

  # Force Slurm partition (the attribute the 'batch slurm' adaptor consumes)

  set_batch_queue = "gpu"

  # Optional: also enforce gpu-related resources for Slurm wrappers

  # set_request_gpus = 1

@jrt

# --- General Slurm route: everything else

JOB_ROUTER_ROUTE_Slurm_Cluster @=jrt

  TargetUniverse = 9

  GridResource = "batch slurm"

  # Optional: pick a default partition if desired, e.g., 'skylake'

  # set_batch_queue = "skylake"

@jrt

# Route order matters: the first matching route wins.

JOB_ROUTER_ROUTE_NAMES = Slurm_GPU_Cluster Slurm_Cluster

I want to route job to specific gpu partition if job submit file requests GPU or otherwise job will be routed to the other route.

How can I achieve that? I have tried above router configuration, but itâs not working. All the jobs are landing on Slurm_GPU_Cluster route.

Best Regards,

Kamrul Shaker

Cloud and High-Throughput Computing Specialist â IT4 (AESES)
Deanâs Office, Faculty of Science
University of Manitoba
ð Kamrul.Shaker@xxxxxxxxxxxx| ð +1 431 335 3508
ð Winnipeg, MB, Canada
ð www.umanitoba.ca

 

_______________________________________________
HTCondor-users mailing list
To unsubscribe, send a message to htcondor-users-request@xxxxxxxxxxx with a
subject: Unsubscribe

The archives can be found at: https://www-auth.cs.wisc.edu/lists/htcondor-users/