I'm invoking a job with qsub myjob.pbs. In there, I have some logic to run my experiments, which includes running torchrun, a distributed utility for pytorch. In that command you can set the number of nodes and number of processes (+gpus) per node. Depending on the availability, I want to be able to invoke qsub with an arbitrary number of GPUs, so that both -l gpus= and torchrun --nproc_per_node= are set depending on the command line argument.
I tried, the following:
#!/bin/sh
#PBS -l "nodes=1:ppn=12:gpus=$1"
torchrun --standalone --nnodes=1 --nproc_per_node=$1 myscript.py
and invoked it like so:
qsub --pass "4" myjob.pbs
but I got the following error: ERROR: -l: gpus: expected valid integer, found '"$1"'. Is there a way to pass the number of GPUs to the script so that the PBS directives can read them?
The problem is that your shell sees PBS directives as comments, so it will not be able to expand arguments in this way. This means that the expansion of $1 will not be occur using:
#PBS -l "nodes=1:ppn=12:gpus=$1"
Instead, you can apply the -l gpus= argument on the command line and remove the directive from your PBS script. For example:
#!/bin/sh
#PBS -l ncpus=12
set -eu
torchrun \
--standalone \
--nnodes=1 \
--nproc_per_node="${nproc_per_node}" \
myscript.py
Then just use a simple wrapper, e.g. run_myjob.sh:
#!/bin/sh
set -eu
qsub \
-l gpus="$1" \
-v nproc_per_node="$1" \
myjob.pbs
Which should let you specify the number of gpus as a command-line argument:
sh run_myjob.sh 4
Related
I want to submit a work to our school's cluster, but it seems that something went error with the environment variable. Here is a test PBS file
#PBS -N test
#PBS -l nodes=1:ppn=1
#PBS -q ser
#PBS -V
#PBS -S /bin/bash
echo ${PBS_NODEFILE}
echo ${PBS_JOBID}
env
I can see the two variable in "env" but "echo" does not print this two variables, and I also can't use these variables in my script. For example "cat ${PBS_NODEFILE} > nodes" will output nothing. Why is this happen???
Well actually I found later that not only this two variables but all variables that is not passed from the current environment(by using -V), such as PBS_O_SHELL , cannot be used in the script. And the other environmental variables such as PATH can be printed normally
I can submit a job to PBS using both approaches of Non-interactive Batch Jobs and/or Interactive Batch Jobs. However, I need to use the pbs commands in a function. In other world I need a structure like this:
#!/bin/sh
pbs_setup () {
#PBS -l $1
#PBS -N $2
#PBS -q normal
#PBS -A $USER
#PBS -m ae
#PBS -M $USER"#gmail.com"
#PBS -q normal
#PBS -l nodes=1:ppn=8
#PBS
}
pbs_setup "walltime=6:00:00" "step3";
echo " "
echo "Job started
echo " "
echo "Job Ended
When I am submitting this job it is not working.
In fact my final goal is separating the commands of job from the main body of code. So when HPC will be changed I just edit a shell file which is included this function instead of editing all the shells. I appreciate if you give me some suggestions.
You could create your custom submission command that collects the job options and sends them as command line parameters to actual qsub call.
Here is a rather basic example of this. In real usage I would add more sophisticated parameter handling tailored to the type of jobs, and more consistent with qsub interface. Also handling interactive jobs needs additional work.
submit.sh
#!/bin/bash
walltime="${2:-06:00:00}"
name="${3:-step3}"
queue="normal"
acct="$USER"
mailevents="ae"
mailaddress="$USER#gmail.com"
resources="nodes=1:ppn=8"
if [ $# -lt 1 ] ; then
echo "Usage: submit.sh script [walltime [name]]" >
exit 1
fi
script="$1"
qsub -l "$walltime" -N "$name" -q "$queue" -A "$acct" \
-m "$mailevents" -M "$mailaddress" -l "$resources" "$script"
script.sh
#!/bin/bash
echo " "
echo "Job started"
echo " "
echo "Job Ended"
This is supposed to be used as
submit.sh script.sh 06:00:00 step3
The issue with that job script is that the #PBS lines need to be first non-comment lines in the script file.
In my attempt to do this same concept, I used the same type of function you have, but cat the results and the actual commands into another file. i.e. An overarching script creates the 'job' script. You can put the HPC requirements in a separate file, then source it from the creation script.
Edit in response to comment:
e.g.
To specify a path to start the job from:
#PBS - d init_path
"working directory path to be used for the job, PBS_O_INITDIR"
Or
#PBS -D root_path
"root directory to be used for the job, PBS_O_ROOTDIR."
Or
#PBS -w working_path
"If the -w option is not specified, the default working directory is the current directory. This option sets the environment variable PBS_O_WORKDIR."
So the default PBS_O_WORKDIR is the current directory you are IN when you call the script to submit the script to qsub.
Thus, if you set the specific options (d, D, w) for paths relative to the actual script running environment, you'll be able to use the paths you intend.
For specifics including default values of these and other options, you can check out the man page for your app. If using the Torque version of the PBS system, it's available at linux.die.net - qsub
Suppose that I have the following simple bash script which I want to submit to a batch server through SLURM:
#!/bin/bash
#SBATCH -o "outFile"$1".txt"
#SBATCH -e "errFile"$1".txt"
hostname
exit 0
In this script, I simply want to write the output of hostname on a textfile whose full name I control via the command-line, like so:
login-2:jobs$ sbatch -D `pwd` exampleJob.sh 1
Submitted batch job 203775
Unfortunately, it seems that my last command-line argument (1) is not parsed through sbatch, since the files created do not have the suffix I'm looking for and the string "$1" is interpreted literally:
login-2:jobs$ ls
errFile$1.txt exampleJob.sh outFile$1.txt
I've looked around places in SO and elsewhere, but I haven't had any luck. Essentially what I'm looking for is the equivalent of the -v switch of the qsub utility in Torque-enabled clusters.
Edit: As mentioned in the underlying comment thread, I solved my problem the hard way: instead of having one single script that would be submitted several times to the batch server, each with different command line arguments, I created a "master script" that simply echoed and redirected the same content onto different scripts, the content of each being changed by the command line parameter passed. Then I submitted all of those to my batch server through sbatch. However, this does not answer the original question, so I hesitate to add it as an answer to my question or mark this question solved.
I thought I'd offer some insight because I was also looking for the replacement to the -v option in qsub, which for sbatch can be accomplished using the --export option. I found a nice site here that shows a list of conversions from Torque to Slurm, and it made the transition much smoother.
You can specify the environment variable ahead of time in your bash script:
$ var_name='1'
$ sbatch -D `pwd` exampleJob.sh --export=var_name
Or define it directly within the sbatch command just like qsub allowed:
$ sbatch -D `pwd` exampleJob.sh --export=var_name='1'
Whether this works in the # preprocessors of exampleJob.sh is also another question, but I assume that it should give the same functionality found in Torque.
Using a wrapper is more convenient. I found this solution from this thread.
Basically the problem is that the SBATCH directives are seen as comments by the shell and therefore you can't use the passed arguments in them. Instead you can use a here document to feed in your bash script after the arguments are set accordingly.
In case of your question you can substitute the shell script file with this:
#!/bin/bash
sbatch <<EOT
#!/bin/bash
#SBATCH -o "outFile"$1".txt"
#SBATCH -e "errFile"$1".txt"
hostname
exit 0
EOT
And you run the shell script like this:
bash [script_name].sh [suffix]
And the outputs will be saved to outFile[suffix].txt and errFile[suffix].txt
If you pass your commands via the command line, you can actually bypass the issue of not being able to pass command line arguments in the batch script. So for instance, at the command line :
var1="my_error_file.txt"
var2="my_output_file.txt"
sbatch --error=$var1 --output=$var2 batch_script.sh
The lines starting with #SBATCH are not interpreted by bash but are replaced with code by sbatch.
The sbatch options do not support $1 vars (only %j and some others, replacing $1 by %1 will not work).
When you don't have different sbatch processes running in parallel, you could try
#!/bin/bash
touch outFile${1}.txt errFile${1}.txt
rm link_out.sbatch link_err.sbatch 2>/dev/null # remove links from previous runs
ln -s outFile${1}.txt link_out.sbatch
ln -s errFile${1}.txt link_err.sbatch
#SBATCH -o link_out.sbatch
#SBATCH -e link_err.sbatch
hostname
# I do not know about the background processing of sbatch, are the jobs still running
# at this point? When they are, you can not delete the temporary symlinks yet.
exit 0
Alternative:
As you said in a comment yourself, you could make a masterscript.
This script can contain lines like
cat exampleJob.sh.template | sed -e 's/File.txt/File'$1'.txt/' > exampleJob.sh
# I do not know, is the following needed with sbatch?
chmod +x exampleJob.sh
In your template the #SBATCH lines look like
#SBATCH -o "outFile.txt"
#SBATCH -e "errFile.txt"
This is an old question but I just stumbled into the same task and I think this solution is simpler:
Let's say I have the variable $OUT_PATH in the bash script launch_analysis.bash and I want to pass this variable to task_0_generate_features.sl which is my SLURM file to send the computation to a batch server. I would have the following in launch_analysis.bash:
`sbatch --export=OUT_PATH=$OUT_PATH task_0_generate_features.sl`
Which is directly accessible in task_0_generate_features.sl
In #Jason case we would have:
sbatch -D `pwd` --export=hostname=$hostname exampleJob.sh
Reference: Using Variables in SLURM Jobs
Something like this works for me and Torque
echo "$(pwd)/slurm.qsub 1" | qsub -S /bin/bash -N Slurm-TEST
slurm.qsub:
#!/bin/bash
hostname > outFile${1}.txt 2>errFile${1}.txt
exit 0
I'm trying to run a matlab script (ga_opt_main.m) in a cluster. I have to write a job submission file, which is essentially just a shell script. But I have never written a shell script and this is what I wrote
#!/bin/bash
#PBS -q *queuename*
#PBS -l nodes=1:ppn=20
#PBS -l walltime=02:00:00
#PBS -N ga_opt_main
module load matlab/R2011b
module list
unset DISPLAY
matlab -nodisplay -nodesktop -r *directory path/ga_opt_main.m*
MATLAB opens in the background but my job is not run. Instead I get an error file saying
bash: -c: line 0: syntax error in conditional expression
bash: -c: line 0: syntax error near `fraction'
Any ideas on why this occurs and how it can be avoided?
Thanks!
I've never used PBS before, but to run a MATLAB script from the shell, try the following:
matlab -nodesktop -nodisplay -r "addpath('/directory/path'); ga_opt_main; quit;"
where ga_opt_main.m is the name of the script file, and '/directory/path' is the directory where it resides. Note that you must have any other dependencies to this script on the MATLAB path as well.
There is also a convenient RUN function that does something similar:
matlab ... -r "run('/directory/path/ga_opt_main.m'); quit;"
###############################
#!/bin/sh
#PBS -l nodes=1
#PBS -l walltime=2:0:0
#PBS -j oe
#PBS -o localhost:/dev/null
#PBS -d /your/working/directory
cd $PBS_O_WORKDIR
matlab -nodisplay -nodesktop -nojvm -nosplash -r "your_matlab_function"
I like to add addpath(genpath('~/your/script's/home')); to the actual matlab script/function. Also, do not add the ".m" to your matlab filename.
Is there a way to directly pass parameters to a .pbs script before submitting a job? I need to loop over a list of files indicated by different numbers and apply a script to analyze each file.
The best I've been able to come up with is the following:
#!/bin/sh
for ((i= 1; i<= 10; i++))
do
export FILENUM=$i
qsub pass_test.pbs
done
where pass_test.pbs is the following script:
#!/bin/sh
#PBS -V
#PBS -S /bin/sh
#PBS -N pass_test
#PBS -l nodes=1:ppn=1,walltime=00:02:00
#PBS -M XXXXXX#XXX.edu
cd /scratch/XXXXXX/pass_test
./run_test $FILENUM
But this feels a bit wonky. Particularly, I want to avoid having to create an environment variable to handle this.
The qsub utility can read the script from the standard input, so by using a here document you can create scripts on the fly, dynamically:
#!/bin/sh
for i in `seq 1 10`
do
cat <<EOS | qsub -
#!/bin/sh
#PBS -V
#PBS -S /bin/sh
#PBS -N pass_test
#PBS -l nodes=1:ppn=1,walltime=00:02:00
#PBS -M XXXXXX#XXX.edu
cd /scratch/XXXXXX/pass_test
./run_test $i
EOS
done
Personally, I would use a more compact version:
#!/bin/sh
for i in `seq 1 10`
do
cat <<EOS | qsub -V -S /bin/sh -N pass_test -l nodes=1:ppn=1,walltime=00:02:00 -M XXXXXX#XXX.edu -
cd /scratch/XXXXXX/pass_test
./run_test $i
EOS
done
You can use the -F option, as described here:
-F
Specifies the arguments that will be passed to the job script when the script is launched. The accepted syntax is:
qsub -F "myarg1 myarg2 myarg3=myarg3value" myscript2.sh
Note: Quotation marks are required. qsub will fail with an error
message if the argument following -F is not a quoted value. The
pbs_mom server will pass the quoted value as arguments to the job
script when it launches the script.
See also this answer
If you just need to pass numbers and run a list of jobs with the same command except the input file number, it's better to use a job array instead of a for loop as job array would have less burden on the job scheduler.
To run, you specify the file number with PBS_ARRAYID like this in the pbs file:
./run_test ${PBS_ARRAYID}
And to invoke it, on command line, type:
qsub -t 1-10 pass_test.pbs
where you can specify what array id to use after -t option