Interactive allocations with salloc

salloc reserves compute resources for an interactive session. After allocation, connect to the assigned node to run commands or development tools. Use sbatch for unattended batch work.

Syntax

In this original notation, 资源参数 means resource options:

salloc [资源参数]

Option	Meaning
-N	Number of nodes
--ntasks	Number of tasks
--cpus-per-task	CPUs per task
--mem	Memory per node, e.g. 16G
--gres=gpu:N	GPUs per node
--partition	Partition
--time	Wall-time limit, e.g. 02:00:00
--qos	Quality of service, subject to partition policy

CPU example

One node, eight CPUs, 16 GB RAM, and five minutes:

salloc -N 1 --cpus-per-task=8 --mem=16G -t 5:00 -p qcpu_18i

Example allocation output:

salloc: Pending job allocation 273377
salloc: job 273377 queued and waiting for resources
salloc: job 273377 has been allocated resources
salloc: Granted job allocation 273377
salloc: Waiting for resource configuration
salloc: Nodes bnode1 are ready for job

Connect to the assigned node:

ssh bnode1

GPU example

One node, one task, four CPUs, 16 GB RAM, two GPUs, and five minutes:

salloc -N 1 --ntasks=1 --cpus-per-task=4 --mem=16G --gres=gpu:2 -p qgpu_3090 -t 5:00

Example allocation output:

salloc: Pending job allocation 273390
salloc: job 273390 queued and waiting for resources
salloc: job 273390 has been allocated resources
salloc: Granted job allocation 273390
salloc: Waiting for resource configuration
salloc: Nodes gnode12 are ready for job

Connect to the assigned node:

ssh gnode12

Inspect allocations

List visible jobs; use squeue -u "$USER" for only your jobs:

squeue

Illustrative output (IDs, nodes, and elapsed times are examples, not the live state of the allocations above):

JOBID  PARTITION  NAME      USER  ST  TIME   NODES  NODELIST(REASON)
273377 qcpu_18i   interact  demo  R   13:41      1  bnode1
273390 qgpu_3090  interact  demo  R    4:25      1  gnode12

More detail:

squeue -l

Connect to a node

Use the hostname in your allocation, for example:

ssh bnode1

ssh gnode12

Without an active allocation from salloc, srun, or sbatch, direct SSH access to a compute node is normally denied:

Access denied by pam_slurm_adopt: you have no active jobs on this node

Workflow and release

Request resources from a login node, wait for allocation, inspect the assigned node, connect by SSH, and start your commands or tools. When finished, leave the node and exit the allocation shell:

exit

Or cancel it from a login node, replacing 作业ID with its numeric ID:

scancel 作业ID

See JupyterLab and VS Code for interactive development.

Demonstration

Additional tutorial