3.5 Array jobs in Taito
In many cases the computational analysis job contains a number of similar independent subtasks. The user may have several datasets that will be analyzed in the same way or same simulation code is executed with a number of different parameters. These kind of tasks are often called as "embarrassingly parallel" jobs as the task can be in principle distributed to as many processors as there are subtasks to be run. In Taito this kind of tasks can be effectively run by using the array job function of the SLURM batch job system.
In SLURM, an array job is defined by using the option --array or -a . For example definition:
will launch not just one batch job, but 100 batch jobs where the subjob specific environment variable $SLURM_ARRAY_TASK_ID gets values form 1 to 100. This variable can then be utilized in the actual job launching commands so that each subtask gets processed. All the subjobs are launched to the batch jobs system at once and they will be executed using as many processors as there are available.
Note that in Taito the range of the $SLURM_ARRAY_TASK_ID variable is limited between 0 and 1000.
In addition to a defining a job range, you can also provide a list of job index values. For example definition:
would launch three jobs with $SLURM_ARRAY_TASK_ID values 4, 7 and 22.
You can also add a step size to the job range definition. For example following array job definition:
would run five jobs with $SLURM_ARRAY_TASK_ID values: 1, 21, 41, 61 and 81.
In some cases it may be reasonable to limit to number of simultaneously running processes. This is done with notation: %max_number_of_jobs. For example, in a case were you have 100 jobs but a license for only five simultaneous processes, you could ensure that you will not run out of license with following definition:
As a first array job example lets assume that we have 50 datasets (data_1.inp, data_2.inp … data_50.inp) that we would like to analyze using program my_prog, that uses syntax:
my_prog inputfile outputfile
Each of the subtasks requires less than 2 hours of computing time and less than 4 GB of memory. We can perform all 50 analysis tasks with following batch job script:
#!/bin/bash -l #SBATCH -J array_job #SBATCH -o array_job_out_%A_%a.txt #SBATCH -e array_job_err_%A_%a.txt #SBATCH -t 02:00:00 #SBATCH --mem-per-cpu=4000 #SBATCH --array=1-50 #SBATCH -n 1 #SBATCH -p serial # move to the directory where the data files locate cd data_dir # run the analysis command my_prog data_"$SLURM_ARRAY_TASK_ID".inp data_"$SLURM_ARRAY_TASK_ID".out
In the batch job script the line #SBATCH --array=1-50 defines that 50 subjobs will be submitted. Other #SBATCH lines refer to the individual subjobs. In this case one subjob uses one processor (-n 1) max. 4 GB of memory (--mem-per-cpu=4000) and can last max. 2 hours ( -t 02:00:00 ). However, the total wall clock time needed to process all the 50 tasks is not limited by any sense.
Note that adding --mail-user definition to the batch job script would make every subjob to send an e-mail when the subjob finishes. This is probably not a good idea in a case, where dozens of subjobs are to be processed.
In the job execution commands, the script utilizes $SLURM_ARRAY_TASK_ID variable in the definition of input and output files so that the first subjob will run command:
my_prog data_1.inp data_1.outsecond will run command:
my_prog data_2.inp data_2.outand so on.
The job can be now launched with command:
Typically not all jobs get into the execution at once. However after a while a large number of jobs may be running in the same time. When the batch job is finished the data_dir directory contains 50 output files.
If you give command squeue -l after submitting your array job, you can see that you have one job pending and possibly several jobs running in the batch job system. All these jobs have a jobid that contain two parts: jobid number of the array job and the sub-job number. Directing the output of each subjob into a separate file is recommended as the file system may fail if several dozens of processes try to write into same file at the same time. If the output files need to be merged into one file it can often be easily done after the array job has finished. For example in the case above we could collect the results into one file with command:
cat data_*.out > all_data.out
In the case of standard output and error files, defined in the SBATCH lines, you can use definitions %A and %a to give unique names to the output files of each sub-job. In the file names %A will be replaced by the ID of the array job and %a will be replaced by the $SLURM_ARRAY_TASK_ID.
In the example above, we were able to use $SLURM_ARRAY_TASK_ID to refer to the order numbers in the input files. If this type of approach is not possible a list of files or commands, created before the submission of the batch jobs, can be used. Let's assume that we have a similar task as defined above, but the file names don't contain numbers but are in format data_aa.inp, data_ab.inp, data_ac.inp… and so on. Now we need first to make a list of files to be analyzed, In this case we could collect the file names into file "namelist" with command:
ls data_*.inp > namelistIn example below we will use command :
sed –n "row_number"p inputfileto read a certain line form the name list file. In this case the actual command script could be following:
#!/bin/bash -l #SBATCH -J array_job #SBATCH -o array_job_out_%j.txt #SBATCH -e array_job_err_%j.txt #SBATCH -t 02:00:00 #SBATCH --mem-per-cpu=4000 #SBATCH --array=1-50 #SBATCH -n 1 #SBATCH -p serial # move to the directory where the data files locate cd data_dir # set input file to be processed name=$(sed -n "$SLURM_ARRAY_TASK_ID"p namelist) # run the analysis command my_prog $name $name.out
This example is otherwise similar to the first one, but it will read the name of the file to be analyzed form a file called namelist. This value is stored into variable $name, which will be used in the job execution command. As the row number to be read is defined by the $SLURM_ARRAY_TASK_ID, each data file listed in the file namelist will get processed in a different subjob. Note that as we now use the $name also in the output definition the output file name will be in format data_aa.inp.out, data_ab.inp.out, data_ac.inp.out… and so on.
3.5.4 Using array jobs in workflows with sbatch_commandlist
In Taito, you can use command sbatch_commandlist to execute a list of commands as an array job. This command takes as an input a text file. Each row in this file is executed as an independet sub-task of a array batch job, automatically generated by the sbatch_commandlist.
The sytax of this command is:
sbatch_commandlist -commands commandlist
Options -t and -mem can be used to modify the time and memory reservation of the subjobs (default 12 h, 8GB).
After submitting an array job, sabatch_commandlist monitors the progress of the job and finishes only when the array job has finished. Thus this command can be used in workflows (including batch job scripts), where only certain steps of the workflow can utilize array jobs based parallel computing.
As an example, lets assume we have a gzip compressed tar-archive file my_data.tgz containing a directory with a large number of files. To create a new compressed archive, that includes also a md5 checksum file for each file we would need to: (1) un-compress and un-pack my_data.tgz, (2) execute md5sum for each file and finally (3) pack and compress the my_data directory again. The second step of the workflow could be executed using a for-loop, but we could also use the loop just to generate a list of md5sum commands, that can be processed with sbatch_commandist.
#!/bin/bash -l #SBATCH -J workfow #SBATCH -o workflow_out_%j.txt #SBATCH -e workflow_err_%j.txt #SBATCH -t 12:00:00 #SBATCH --mem=4000 #SBATCH -n 1 #SBATCH -p serial #open the tgz file tar zxf my_data.tgz cd my_data #generate a list of md5sum commands for my_file in * do echo "md5sum $my_file > $my_file.md5" >> md5commands.txt done #execute the md5commands as an array job sbatch_commandlist -commands md5commands.txt #remove the command file and compress the directory rm -f md5commands.txt cd .. tar zcf my_data_with_md5.tgz my_data rm -rf my_data
Note that the batch job script above is not an array job, but it launches a another batch job that is an array job.
|Previous chapter||One level up||Next chapter|