Home

slurmjobpacker

Places Slurm batch jobs on shared nodes so CPUs and memory run out together.

• sjpd: daemon on one controller; reads free CPUs and memory every second and keeps a placement table in /run/sjp.
• sjp.lua: called from your job_submit.lua; sjp.place() sets partitions (and the node, if the job can start now), or returns false and a reason.

Slurm still schedules. Installation: see the README.

• How it works
• Decision making
• Using sjp in job_submit.lua
• Configuration reference
• QOS limits and flex
• Logs and AdminComment
• Dry run and going live
• Visualizer
• Troubleshooting
• Limitations
• Upgrading