Changes in 26.05.2
- slurmctld - Prevent segfaulting from dereferencing stale QOS, association, or user pointers after updating the assoc_mgr.
- Fix nodes_alloc statistics overcount caused by shared nodes being counted multiple times across jobs.
- slurmscriptd - Fix support for profiling conmgr on SIGPROF.
- Add option to use the job id in the cgroup path instead of SLUID, thus allowing to restore the old behavior. Default is still SLUID.
- Fix undefined behavior that may affect future scheduling decisions for jobs or partitions that use --exclusive=topo.
- Fix memory leak in slurmstepd's PMIx startup.
- Fix pmix jobs failing on startup when requesting a large number of tasks.
- Disabled slurmctldparams=threadpool=enabled in slurm.conf.
- Make --test-only evaluate multi-partition jobs against the candidate partition being tested.
- Prevent slurmctld from crashing with a divide-by-zero when --test-only uses an empty partition before a usable one.
- slurmctld,slurmd - Avoid race condition that could cause a crash in the signal manager during shutdown.
- sacct,sacctmgr,scontrol,sdiag,sinfo,squeue,sshare - Include all command arguments in '.meta.command' field when using --json or --yaml options.
- Fix srun --relative placing steps on the wrong node.
- Fix slurmctld crashing when using job arrays that request mode 3 HRES.
- slurmctld - Fix regression from 23.02.0rc1 with failed update request for job --prefer, resulting incorrect reset of original options.
- slurmctld - Fix memory leak when using --prefer option.
- srun no longer reports a successful (0:0) completion when the step is aborted before any task could start.
- Fix error resuming a node that has not registered yet.
- Prevent delaying job cancellation if the slurmstepd was in the middle of launching a task when the job was cancelled.
- s2n/tls - Fix deadlock in the extern slurmstepd which may occur when doing X11 forwarding.
- data_parser/v0.0.45 - Fix printing empty list '[]' instead of empty strings for enumerated flag arrays in JSON/YAML.
- tls/s2n - Fix --async steps not running due to slurmstepd not being able to generate certificates with the certgen plugin.
- srun - Prevent deferring steps indefinitely that request --relative and -c without specifying the -n option.
- Fix topology/tree ROUTE debug logging the wrong switch name.
- Fix topology/tree fan-out for leaves with no common parent.
- sbatch and srun now emit "Slurm job queue full, sleeping and retrying" when the MaxJobCount limit is reached.
- Fix a memory leak of the TLS certificate when a task reattach fails.
- Fix memory leaks in the slurmstepd getpw/getgr/gethost request handlers.
- Fix a memory leak and a potential deadlock in slurmstepd stepmgr step completion handling.
- Allow FLEX+MAGNETIC reservations to pick up jobs that extend beyond the reservation's end time.