Issue

If one doesn't specify --nodes=1, then the job might get allocated slots across multiple machines, e.g.

$ sbatch --ntasks=65 slurm_envs.sh
Submitted batch job 9394

$ squeue -u "$USER"
             JOBID PARTITION     NAME     USER ST       TIME  NODES NODELIST(REASON) 
              9394       cbc slurm_en   henrik  R       0:07      2 c4-n[12-13] 

$ grep -E "^(SLURM|SBATCH)_" slurm-9394.out
SLURM_CPUS_ON_NODE=64
SLURM_GTIDS=0
SLURM_JOB_ACCOUNT=cbi
SLURM_JOB_CPUS_PER_NODE=64,1
SLURM_JOB_GID=509
SLURM_JOB_ID=9394
SLURM_JOBID=9394
SLURM_JOB_NAME=slurm_envs
SLURM_JOB_NODELIST=c4-n[12-13]
SLURM_JOB_NUM_NODES=2
SLURM_JOB_PARTITION=cbc
SLURM_JOB_QOS=max_cores
SLURM_JOB_UID=581
SLURM_JOB_USER=henrik
SLURM_LOCALID=0
SLURM_MEM_PER_NODE=2048
SLURM_NNODES=2
SLURM_NODE_ALIASES=(null)
SLURM_NODEID=0
SLURM_NODELIST=c4-n[12-13]
SLURM_NPROCS=65
SLURM_NTASKS=65
SLURM_PRIO_PROCESS=0
SLURM_PROCID=0
SLURM_SUBMIT_DIR=/c4/home/henrik/tests-slurm
SLURM_SUBMIT_HOST=c4-dev1
SLURM_TASK_PID=46543
SLURM_TASKS_PER_NODE=64,1
SLURM_TOPOLOGY_ADDR=c4-n12
SLURM_TOPOLOGY_ADDR_PATTERN=node

Action

[x] Update all examples to use an explicit --nodes=1
[x] Mention it in the text
[ ] Add a temporary warning panel (since we didn't mention --nodes=1 from the beginning

UCSF-CBI / c4

DOCS: Slurm --nodes=1 is required - it's not the default #42

Issue

Action