[slurm-users] Slurm version 22.05.5 is now available
tim at schedmd.com
Thu Oct 13 20:24:38 UTC 2022
We are pleased to announce the availability of Slurm version 22.05.5.
This fixes a number of moderate severity issues, alongside one
unfortunate problem with the upgrade process for running jobs with the
slurmstepd when using RPM-based installations. Please see Jason Booth's
email the slurm-users mailing list for further details, and ways to
mitigate this problem:
Slurm can be downloaded from https://www.schedmd.com/downloads.php .
Chief Technology Officer, SchedMD LLC
Commercial Slurm Development and Support
> * Changes in Slurm 22.05.5
> -- Fix node becoming IDLE while in an invalid registration state.
> -- When a job is completing avoid potential dereference.
> -- Avoid setting preempt_time for a job erroneously.
> -- Fix situation where we don't requeue correctly when a job is finishing.
> -- job_container/tmpfs - Avoid leaking namespace file descriptor.
> -- common/slurm_opt - fix memory leak in client commands or slurmrestd when the
> --chdir option is set after option reset.
> -- openapi/dbv0.0.38 - gracefully handle unknown associations assigned to jobs.
> -- openapi/dbv0.0.38 - query all associations to avoid errors while dumping
> -- Load hash plugin at slurmstepd launch time to prevent issues loading the
> plugin at step completion if the Slurm installation is upgraded.
> -- Fix gcc 12.2.1 compile errors.
> -- Fix future magnetic reservations preventing heterogeneous jobs from
> starting. > -- Prevent incorrect error message from being generated for
> using the 'scontrol top' command.
> -- slurmrestd - correct issue where larger requests could result in a single
> byte getting removed from inside of the POST request.
> -- Fix regression in task count calculation for --ntasks-per-gpu with multiple
> -- Update nvml plugin to match the unique id format for MIG devices in new
> Nvidia drivers.
> -- Fix segfault on backup slurmdbd if no QoS is present in DB.
> -- Fix clang 11 compile errors.
> -- Fix task distribution calculations across sockets with
> -- Fix task distribution calculations with --ntasks-per-gpu specified without
> an explicit --ntasks value.
> -- Fix job arrays not showing correct features.
> -- Fix job having wrong features used when using preferred features.
> -- Fix task/cray_aries error finishing an interactive step, avoiding correct
> -- Correctly set max_nodes when --ntasks=1.
> -- Fix configure script on FreeBSD.
More information about the slurm-users