Cheatography
https://cheatography.com
Commands for interacting with slurm
This is a draft cheat sheet. It is a work in progress and is not finished yet.
Nodes
|
Your allocated compute nodes are NOT shared with anyone else. |
|
Number of compute nodes needed |
|
Specify the memory required per node. |
|
Specify the number of GPUs per node |
|
Prevent the job from being scheduled on the specified nodes. |
|
GPU
--gres=gpu:[type]:[count]
|
GPUs per node |
--gpus-per-node=[type]:[count]
|
Modern alternative to gres
|
--gpus-per-task=<count>
|
GPUs per task |
|
Specify total number of GPUs for the entire job |
|
Claim Resources
|
Request a interactive session |
sbatch <script-name.sh>
|
Submit a non-interactive job |
|
squeue
|
Limit info to only you |
|
Limit info to specific user |
|
Filter by state • PD
- Pending • R
- Running |
-j
/ --jobs=[job-id-list]
|
Limit info to specific jobs |
View information about jobs located in the Slurm scheduling queue
|
|
scancel
|
Cancel jobs owned by current user |
|
Cancel jobs under a specific charge account |
|
Cancel jobs with a matching job name |
|
Cancel all jobs owned by a specific user |
|
Cancel jobs in a specific state |
sacct
-j
/ --jobs=[job-id-list]
|
Check a specific job ID |
|
Hide job steps and show only main job allocation |
|
Check specified users. Defaults to running user |
|
Specify fields. Comma sep, no spaces. • JobID
• JobName
• State
• Elapsed
(HH:MM:SS) |
|