Job arrays: Difference between revisions

<translate>

''Parent page: [[Running jobs]]''

If your work consists of a large number of tasks which differ only in some parameter, you can conveniently submit many tasks at once using a ''job array,'' also known as a ''task array'' or an ''array job''. The individual tasks in the array are distinguished by an environment variable, <code>$SLURM_ARRAY_TASK_ID</code>, which Slurm sets to a different value for each task. You set the range of values with the <code>--array</code> parameter.

See [https://slurm.schedmd.com/job_array.html Job Array Support] at SchedMD.com for detailed documentation.

   

  sbatch --array=1,3,5,7  # $SLURM_ARRAY_TASK_ID takes the listed values

  sbatch --array=1-7:2    # Step-size of 2, same as the previous example

  sbatch --array=1-100%10  # Allows no more than 10 of the jobs to run simultaneously

{{File

|name=simple_array.sh

}}

This job will be scheduled as ten independent tasks. Each task has a separate time limit of 3 hours, and each may start at a different time on a different host.  

The script references $SLURM_ARRAY_TASK_ID to select an input file, for example ("program_x"), or to set a command-line argument for the application ("program_y").

Using a job array instead of a large number of separate serial jobs has advantages for you and other users. A waiting job array only produces one line of output in squeue, making it easier for you to read its output. The scheduler does not have to analyze job requirements for each array task separately, so it can run more efficiently too.

Note that, other than the initial job-submission step with <code>sbatch</code>, the load on the scheduler is the same for an array job as for the equivalent number of non-array jobs. The cost of dispatching each array task is the same as dispatching a non-array job. You should not use a job array to submit tasks with very short run times, e.g. much less than an hour. Tasks with run times of only a few minutes should be grouped into longer jobs using [[Glost]], [[GNU Parallel]], or a shell loop.

Suppose you have multiple directories, each with the same structure, and you want to run the same script in each directory. If the directories can be named with sequential numbers then the example above can be easily adapted. If the names are not so systematic, then create a file with the names of the directories, like so:

  pacific2016

  pacific2017

  atlantic2017

There are several ways to select a given line from a file; this example uses <code>sed</code> to do so:

{{File

|name=directories_array.sh

#SBATCH --array=1-4

echo "Starting task $SLURM_ARRAY_TASK_ID"

DIR=$(sed -n "${SLURM_ARRAY_TASK_ID}p" case_list)

cd $DIR

# Place the code to execute here

pwd

}}

Cautions:

* Take care that the number of tasks you request matches the number of entries in the file.