We are pleased to announce the availability of Slurm version 22.05.1.
This includes one significant fix to an regression introduced in 22.05.0
issue that can lead to over-subscription of licenses. For sites running
22.05.0 the new "bf_licenses" option to SchedulerParameters will resolve
this issue, otherwise upgrading to this new maintenance release is
strongly encouraged.
Slurm can be downloaded from https://www.schedmd.com/downloads.php .
- Tim
--
Tim Wickberg
Chief Technology Officer, SchedMD LLC
Commercial Slurm Development and Support
* Changes in Slurm 22.05.1
==========================
-- Flush the list of Include config files on SIGHUP.
-- Fix and update Slurm completion script.
-- jobacct_gather/cgroup - Add VMem support both for cgroup v1 and v2.
-- Allow subset of node state transitions when node is in INVAL state.
-- Remove INVAL state from cloud node after being powered down.
-- When showing reason UID in scontrol show node, use the authenticated UID
instead of the login UID.
-- Fix calculation of reservation's NodeCnt when using dynamic nodes.
-- Add SBATCH_{ERROR,INPUT,OUTPUT} input environment variables for --error,
--input and --output options respectively.
-- Prevent oversubscription of licenses by the backfill scheduler when not
using the new "bf_licenses" option.
-- Jobs with multiple nodes in a heterogeneous cluster now have access to all
the memory on each node by using --mem=0. Previously the memory limit was
set by the node with the least amount of memory.
-- Don't limit the size of TaskProlog output (previously TaskProlog output was
limited to 4094 characters per line, which limited the size of exported
environment variables or logging to the task).
-- Fix usage of possibly uninitialized buffer in proctrack/cgroup.
-- Fix memleak in proctrack/cgroup proctrack_p_wait.
-- Fix cloud/remote het srun jobs.
-- Fix a segfault that may happen on gpu configured as no_consume.
