The following list are items that were resolved in Moab v9.1.0 release:
- Moab was not reporting the correct task count for jobs that allocate ALLPROCS
- The systemd file generated with the --prefix option was not using the specified path
- systemd was terminating the active Moab server instead of the local Moab server
- Scheduler Analysis Block Reasons section displayed duplicate messages
- Task count was incorrect when requesting generic resource scheduling (GRES) on the global node.
- Moab block messages were not always informative
- If jobs with a custom name were canceled, subsequent jobs with a similar custom name failed
- Jobs were rejected when requesting NUMA resources and NODEACCESSPOLICY=SINGLEJOB
- msub did not allow you to specify a working directory, as with qsub -w <path>
- The class of GPU jobs that request more than one task was not being mapped correctly
- DPROC was not being calculated correctly on GPU jobs
- mcredctl was not reporting variables in credentials
- Jobs place on hold or batchhold were not being correctly recognized as being on hold
- Requeuing a job that was restarting caused it to be canceled
- Requesting a node's name as a feature prevents other jobs from getting that node. Jobs that request node names as features will now be placed on hold
- Issues reported for groups, users, and accounts with spaces in the name. Enabled SCHEDCFG[] FLAGS=AllowCredentialsWithSpaces to let Moab ignore POSIX standards and allow groups, users, and accounts with spaces in the name
- Jobs with the "AFTER" dependency were not auto-cancelling
- Job ID collisions were occurring in a grid environment
- Fixed issue with multi-req jobs and "EnforceReservedNodes"
- In a Master/Slave grid configuration, the Master Moab did not honor the TTL attribute of jobs running on the slave node
- mdiag -C gave an error when specifying an SRCFG GRES resource.
