← Back to all sparks
S

Slurm

DEVOPS
Velocity5.0

Workload manager and job scheduler for HPC clusters

Slurm patches crashes and heterogeneous-job hangs on two branches while 26.11 is in RC.

hpcjob-schedulingcontainersmaintenancerelease-candidate
◆Current state
Slurm runs three branches. On November 10, 26.05.6 and 25.11.9 shipped fixes for a slurmstepd crash with stepmgr in array tasks, PMIx hangs in heterogeneous job steps, a slurmd memory leak and a heap corruption on failed RPC forwarding. The first 26.11 release candidate, out on November 6, turns on the slurmctld threadpool by default and moves OCI containers into a runtime plugin.
◆Where it's heading
Maintenance branches get regular crash and correctness fixes while 26.11 heads to release with new defaults. The 26.11 RC's container and threadpool changes are where behaviour will shift for sites; heterogeneous jobs and stepmgr remain areas with frequent fixes.
◆Prediction
A second 26.11 release candidate or 26.11.0 is the next likely release, followed by more point releases on 26.05 and 25.11.

◆Recent moves

  1. 3d ago

    v26.05.6: fixes heterogeneous-job hangs and a stepmgr crash

    A maintenance release fixing PMIx hangs and aborts in heterogeneous job steps, a stepmgr crash in array tasks, a slurmd memory leak and nodes shown idle while still completing. It is the current stable line's cleanup ahead of 26.11.

    View source ↗
  2. 3d ago

    v25.11.9: heap corruption and slurmrestd read fixes

    Backports crash fixes to the older 25.11 line, including heap corruption on failed RPC forwarding, an out-of-bounds read in slurmrestd and quote handling in runtime/oci. Routine branch maintenance.

    View source ↗
  3. 8d ago

    Slurm 26.11 RC1: threadpool on by default, OCI runtime plugin

    The first 26.11 release candidate changes defaults: the slurmctld threadpool is on, RPC replies are always asynchronous, and OCI containers move to a runtime/oci plugin selected with a renamed --runtime flag. It is the biggest structural change in the feed since the 26.05 line began, and sites will need to review renamed slurm.conf options before upgrading.

    View source ↗
  4. 10d ago

    wickberg-test: Testsuite - Cover a --stepmgr step cancel in test_101_18

    A testsuite commit, not a release: it checks that cancelling a --stepmgr job step records the owner, not root, in sacct. Hints at an accounting fix in the next point release.

    View source ↗
  5. 1mo ago

    v26.05.5

    Slurm 26.05.5 fixes qstat -f output, a slurmd double free and a regression placing powered-down nodes incorrectly. A routine maintenance release.

    View source ↗
  6. 2mo ago

    Slurm 26.05.4 fixes 7 CVEs, including privilege escalation

    v26.05.4 closes 7 CVEs on the current release branch — including a privilege escalation (operator → admin via accounting DB) and SQL injection via cluster names — alongside the usual bug fix sweep of memory leaks, crash paths, and heterogeneous job handling. For HPC clusters with multiple user roles, the privilege escalation fix is the most operationally critical.

    View source ↗