Migrating from Torque to Slurm

From Research Computing Center Wiki
Jump to navigation Jump to search

During the Oct.24-28 maintenance window, GACRC implemented the Slurm software for job scheduling and resource management on Sapelo2. Slurm replaced the Torque (PBS) resource manager and the Moab scheduling system that Sapelo2 used before then.

How is Slurm different from Torque?

Slurm is different from Torque in a number of ways, including the commands used to submit and monitor jobs, the syntax used to request resources, and the way environment variables behave.

Some specific ways in which Slurm is different from Torque include:

  • Slurm will not allow a job to be submitted whose requested resources exceed the set of resources the job owner has access to--whether or not those resources have been already allocated to other jobs at the moment. Torque will queue the job, but the job would never run.
  • What Torque calls queues, Slurm calls partitions.
  • Resources in Slurm are assigned per “task”/process.
  • In Slurm, environmental variables of the submitting process are passed to the job by default

How to Submit Jobs

To submit jobs in Slurm, replace qsub with one of the commands from the table below.

Info Torque Command Slurm Command
Submit a batch job to the queue qsub <job script> sbatch <job script>
Start an interactive job qsub -I <options> salloc <options>

where <job script> needs to be replaced by the name of your job submission script (e.g. sub.sh). Note that the syntax in the job submission script for Slurm is different from the Torque syntax.

Job Submission Options

As with Torque, job options and resource requests in Slurm can be set in the job submission script or as options to the job submission command. However, the syntax used to request resources is different and the table below summarizes some of the options that are frequently used.

Option Torque (qsub) Slurm (sbatch)
Script directive #PBS #SBATCH
Job name -N <name> --job-name=<name>
-J <name>
Queue -q <queue> --partition=<queue>
Wall time limit -l walltime=<hh:mm:ss> --time=<hh:mm:ss>
Node count -l nodes=<count> --nodes=<count>
-N <count>
Process count per node -l ppn=<count> --ntasks-per-node=<count>
core count (per process) --cpus-per-task=<cores>
Memory limit -l mem=<limit> --mem=<limit> (Memory per node in mega bytes – MB)
Minimum memory per processor -l pmem=<limit> --mem-per-cpu=<memory>
Request GPUs -l gpus=<count> --gres=gpu:<count>
Request specific nodes -l nodes=<node>[,node2[,...]]> -w, --nodelist=<node>[,node2[,...]]>
-F, --nodefile=<node file>
Request node feature -l nodes=<count>:ppn=<count>:<feature> --constraint=<feature>
Job array -t <array indices> --array <indexes>
-a <indexes>
Where <indexes> is replaced by a range (0-15), a list (0, 6, 16-32), or a step function (0-15:4)
Standard output file -o <file path> --output=<file path> (path must exist)
Standard error file -e <file path> --error=<file path> (path must exist)
Combine stdout/stderr to stdout -j oe --output=<combined out and err file path>
Copy environment -V --export=ALL (default)
--export=NONE to not export environment
Copy environment variable -v <variable[=value][,variable2=value2[,...]]> --export=<variable[=value][,variable2=value2[,...]]>
Job dependency -W depend=after:jobID[:jobID...]
-W depend=afterok:jobID[:jobID...]
-W depend=afternotok:jobID[:jobID...]
-W depend=afterany:jobID[:jobID...]
Request event notification -m <events> --mail-type=<events>
Note: multiple mail-type requests may be specified in a comma separated list:
Email address -M <email address> --mail-user=<email address>
Defer job until the specified time -a <date/time> --begin=<date/time>
Node exclusive job qsub -n --exclusive

Common Job Commands

How to Submit and Manage Jobs

Info Torque Command Slurm Command
Submit a job qsub <job script> sbatch <job script>
Delete a job qdel <job ID> scancel <job ID>
Hold a job qhold <job ID> scontrol hold <job ID>
Release a job qrls <job ID> scontrol release <job ID>
Start an interactive job qsub -I <options> salloc <options>
srun --pty <options>
Start an interactive job with X forwarding qsub -I -X <options> srun --x11 <options>

How to View Resources on the Cluster

Info Torque Command Slurm Command
Queue list / info qstat -q [queue] scontrol show partition [queue]
Node list pbsnodes -a
mdiag -n -v
scontrol show nodes
Node details pbsnodes <node> scontrol show node <node>
Cluster status qstat -B sinfo

How to Monitor Jobs

Info Torque Command Slurm Command
Job status (all) qstat
Job status (by job) qstat <job ID> squeue -j <job ID>
Job status (by user) qstat -u <user> squeue -u <user>
Job status (only own jobs) qstat_me squeue --me
squeue --me -l
Job status (detailed) qstat -f <job ID>
checkjob <job ID>
scontrol show job -dd <job ID>
Show expected start time showstart <job ID> squeue -j <job ID> --start
Monitor or review a job’s resource usage qstat -f <job ID> sacct -j <job ID> --format JobID,jobname,NTasks,nodelist,CPUTime,ReqMem,Elapsed
View job batch script scontrol write batch_script <job ID> [filename]

Valid Job States

Below are the job states you may encounter when monitoring your job(s) in Slurm.

Code State Meaning
CA Canceled Job was canceled
CD Completed Job completed
CF Configuring Job resources being configured
CG Completing Job is completing
F Failed Job terminated with non-zero exit code
NF Node Fail Job terminated due to failure of node(s)
PD Pending Job is waiting for compute node(s)
R Running Job is running on compute node(s)

Job Environment and Environment Variables

In Slurm, environment variables will get passed to your job by default.

If you have certain environment variables set that you think might interfere with your job you can either:

  • Log out then log back in and submit your job
  • Run sbatch with one of these options to override the default behavior:
     sbatch --export=None 
     sbatch --export MYPARAM=3 
     sbatch --export=ALL,MYPARAM=3

NOTE: We recommend that you submit sbatch Slurm jobs with the #SBATCH --export=NONE option to establish a clean environment, otherwise Slurm will propagate current environmental variables to the job. This could impact the behavior of the job, particularly for MPI jobs.

Like Torque, Slurm sets its own environment variables within your job. The table below summarizes some environment variables that are frequently used.

Info Torque Slurm Notes
Version $PBS_VERSION Can extract from sbatch --version
Batch or interactive $PBS_ENVIRONMENT
Submit directory $PBS_O_WORKDIR $SLURM_SUBMIT_DIR Slurm jobs start from the submit directory by default.
Node file $PBS_NODEFILE A filename and path that lists the nodes a job has been allocated.
Node list cat $PBS_NODEFILE $SLURM_JOB_NODELIST The Slurm variable has a different format to the Torque/PBS one. To get a list of nodes use:
scontrol show hostnames $SLURM_JOB_NODELIST
Job array index $PBS_ARRAYID
$SLURM_ARRAY_TASK_ID Only set when submitting a job array (with -a or –array)
Number of nodes allocated $PBS_NUM_NODES $SLURM_JOB_NUM_NODES
Number of processes $PBS_NP $SLURM_NTASKS
Number of processes per node $PBS_NUM_PPN $SLURM_TASKS_PER_NODE
List of allocated GPUs $PBS_GPUFILE
Requested tasks per node $SLURM_NTASKS_PER_NODE
Requested CPUs per task $SLURM_CPUS_PER_TASK
Scheduling priority $SLURM_PRIO_PROCESS

Slurm Documentation

Extensive documentation on Slurm is available at https://slurm.schedmd.com/documentation.html

Training material

To help users familiarize with Slurm and the test cluster environment, we have prepared some training videos that are available from the GACRC's Kaltura channel at https://kaltura.uga.edu/channel/GACRC/176125031 (login with MyID and password is required). Training sessions and slides are available at https://wiki.gacrc.uga.edu/wiki/Training

This page was adapted from https://arc-ts.umich.edu/migrating-from-torque-to-slurm/ and https://hpcc.usc.edu/support/documentation/pbs-to-slurm/