Description
Pinning CPUs on a Slurm-based system (as we do for colocated DBs) can result in cgroup issues, and prevent the model to be lauched. A quick fix is to set the custom_pinning to [], for colocated DBs.
How to reproduce
Run a standard co-located DB with pinned CPU on a Slurm-based system where Slurm does not assign CPU 0..N-1 to the srun-launched executable.
Expected behavior
The model should start on the requested CPUs.
System
- OS: RHEL
- Commit/version of library: current
develop
- Workload manager (e.g. PBS, Slurm): Slurm
Description
Pinning CPUs on a Slurm-based system (as we do for colocated DBs) can result in
cgroupissues, and prevent the model to be lauched. A quick fix is to set the custom_pinning to[], for colocated DBs.How to reproduce
Run a standard co-located DB with pinned CPU on a Slurm-based system where Slurm does not assign CPU
0..N-1to thesrun-launched executable.Expected behavior
The model should start on the requested CPUs.
System
develop