Event Date | Summary |
---|---|
We have implemented a workaround by defining the environment variable "SLURM_WHOLE=1" when OpenMPI modules are loaded. This restores previous behavior. |
Service | Incident status | Start Date | End Date |
---|---|---|---|
Cedar | Closed |
Scheduler problem - Problème d'ordonnanceur
Since the upgrade to Slurm 20.11 in the last maintenance, jobs started using "mpiexec" or "mpirun" no longer run properly when using multiple nodes. As a work around until this issue is fixed, the recommendation is to use "srun" when possible. / Depuis la mise à jour de Slurm à la version 20.11 lors de la maintenance, les tâches qui utilisent "mpiexec" ou "mpirun" ne s'exécutent plus correctement lorsqu'elles utilisent plus d'un noeud. Pour contourner le problème, nous recommandons d'utiliser "srun" si possible.
Updated by Maxime Boissonneault on