During my testing I discovered that if the system is using 
AccountingStorageEnforce=QOS and a job is pending with state 
PartitionNodeLimits, that job remains in that state even if the MaxNodes 
limit for the QOS is increased so the job
can run and the command 'sacctmgr modify qos name=lowpri set 
flags=partitionmaxnodes' is issued.  The problem appears to be in 
src/slurmctld/job_scheduler.c build_job_queue() function. The test of 
job_ptr->part_ptr_list fails, so job_limits_check() is never called to 
re-evaluate the job.  The attached patch resolves this problem.
Best Regards,
Bill


Attachment: partitionmaxnodes.patch
Description: Binary data

Reply via email to