Prevent user to change specific parameters

When submitting a job, users can override the default parameters configured for the queue (click here to see a list of all parameters supported by SOCA).

For security, compliance or cost reasons, you may want prevent users to override these default parameters by configuring restricted_parameters on your queue_mapping.yml

Prevent user to choose a different instance type¶

Considering /opt/soca/<CLUSTER_ID>/cluster_manager/settings/queue_mapping.yml

queue_type:
  compute:
    queues: ["normal", "low"]
    instance_type: "c5.large"
    restricted_parameters: ["instance_type"]
    ...

In this example, a job will be rejected if a user try to specify the instance_type parameter when using the normal or low queues. In this particular case, any job sent to the normal or low queue will be forced to use c5.large instance, which is the default instance type configured by HPC admins.

qsub -q normal -l instance_type=m5.24xlarge -- /bin/echo test
qsub: instance_type is a restricted parameter and can't be configure by the user. Contact your HPC admin and update /opt/soca/<CLUSTER_ID>/cluster_manager/settings/queue_mapping.yml

Need to approve more than one instance type/family?

Read the documentation if you want to limit users to a list of multiple instance types

Prevent user to provision additional storage¶

Considering /opt/soca/<CLUSTER_ID>/cluster_manager/settings/queue_mapping.yml

queue_type:
  compute:
    queues: ["normal", "low"]
    scratch_size: "200"
    restricted_parameters: ["scratch_size"]
    ...

In this example, a job will be rejected if a user try to specify the scratch_size parameter when using the normal or low queues. In this particular case, any job sent to the normal or low queue will be forced to use a 200 GB EBS disk as /scratch partition. Users are no longer able to provision more storage than what's allocated to them.

qsub -q normal -l scratch_size=550 -- /bin/echo test
qsub: scratch_size is a restricted parameter and can't be configure by the user. Contact your HPC admin and update /opt/soca/<CLUSTER_ID>/cluster_manager/settings/queue_mapping.yml

Combine multiple restrictions¶

Considering /opt/soca/<CLUSTER_ID>/cluster_manager/settings/queue_mapping.yml

queue_type:
  compute:
    queues: ["normal", "low"]
    scratch_size: "200"
    restricted_parameters:["instance_type", "fsx_lustre_bucket", "scratch_size"]
    ...

In this example, a job will be rejected if a user try to change either instance_type, fsx_lustre_bucket or scratch_size parameters.

qsub -q normal -l fsx_lustre_bucket=mybucket -- /bin/echo test
qsub: fsx_lustre_bucket is a restricted parameter and can't be configure by the user. Contact your HPC admin and update /opt/soca/<CLUSTER_ID>/cluster_manager/settings/queue_mapping.yml

Check the logs¶

Scheduler hooks are located on /var/spool/pbs/server_logs/

Code¶

The hook file can be found under /opt/soca/cluster_hooks/$SOCA_CONFIGURATION/queuejob/check_queue_restricted_parameters.py on your Scale-Out Computing on AWS cluster)

Disable the hook¶

You can disable the hook by running the following command on the scheduler host (as root):

user@host: qmgr -c "delete hook check_queue_restricted_parameters event=queuejob"