R batch jobs with PBS (retired)¶

Please note

CARC schedulers now run Slurm. See R on CARC systems and Example Slurm scripts instead.

The PBS script¶

In order to actually submit a job to the cluster you need a PBS script that lists all of the resources you are requesting, the software you want to use, and the commands to that software. So pretend we have an R script called "my_script.R" that you want to run on Wheeler, the PBS script would look something like this:

#!/bin/bash

#PBS -N my_r_job
#PBS -l walltime=48:00:00
#PBS -l nodes=1:ppn=8
#PBS -j oe
#PBS -V
#PBS -m abe
#PBS -M my_name@unm.edu

cd $PBS_O_WORKDIR

module load r-3.6.0-gcc-7.3.0-python2-7akol5t

Rscript my_script.R

This PBS script would then be submitted to the job queue with

yourusername@wheeler-sn$ qsub my_pbs_script.pbs

A PBS script is just a bash script that combines flags for the job scheduler (PBS), and bash commands. We can break it down by sections.

PBS directives¶

If you are unfamiliar with bash scripting the first line #!/bin/bash/ is called the "hashbang" or "shebang" and is directing your shell on what should be used to interpret the following code. This isn't actually necessary for a PBS script but is just common practice.

The next chunk of lines are the flags to qsub asking for specific resources. The # here are important because they represent comments to bash but are interpreted by qsub.

## The -N flag specifies the name of your job and will be what shows on the queue and the prefix for all 
## PBS specific output.
#PBS -N my_r_job

## The -l flag specifies the actual computational resources you want and can take many arguments. In this case we are 
## asking for 48 hours of walltime on the cluster with one node and all cpu's on that node, which for Wheeler is 8.
#PBS -l walltime=48:00:00
#PBS -l nodes=1:ppn=8

## The -j flag joins your standard out (stdout) and standard error (stderr) into one file. I do this just so there is 
## less output and clutter resulting from a job.
#PBS -j oe

## The -V flag exports user environmental variables from the head node to the compute node. Not always necessary.
#PBS -V

## The -m and -M flags specifiy how you want the scheduler to control mailing information about your job. The -m abe 
## is saying that you want emails for Abort, Begin, and End, and the -M flag is a comma separated list of who
## to send mail to. 
#PBS -m abe
#PBS -M my_name@unm.edu

At the very least you should always specify the walltime, nodes, and processors per node for each job. For an exhaustive list of qsub commands you can type man qsub on the head node of any CARC system.

Calling your code¶

The rest of the PBS script are usually bash commands (or whatever shell you prefer to use, which can be specified with the -S flag) that tells the compute node where your data is, loads the software you want, and then executes your job.

## There are many job specific variables that are created when PBS starts a job, including $PBS_O_WORKDIR. The
## following line moves to the directory where the qsub command was executed.
cd $PBS_O_WORKDIR

## This loads the R software environment.
module load r-3.6.0-gcc-7.3.0-python2-7akol5t

## If you are using a conda environment instead of an R module installed by CARC you would have the following
module load anaconda3
source activate my_r_env

## This line is where you are actually running your R script. The older way of using R CMD BATCH is deprecated
## and Rscript is the preferred way for launching an R batch job.
Rscript my_script.R

Migrated from UNM-CARC QuickBytes (last source update 2019-12-03). Spotted a problem? Open an issue or pull request.