In this exercise, we will submit and observe the outcomes of a few trivial jobs in order to become more familiar with Slurm job dependencies.

1. Log on to Stampede2 or Frontera (or a different Slurm cluster) and create a simple batch script that will sleep for two minutes, as shown below. You can call it "sleep.sh". Edit the --mail-user line with your e-mail address. If you need to specify an account from among several that are available to you, remove one of the hashes from the -A line, and enter your account name there.

2. Create a simple bash script done.sh that does nothing except echo a statement.

3. Submit the sleep job, and examine the terminal output from sbatch. Make note of the job ID, as you will use it immediately in the next step.

   login1.stampede2(1009)$ sbatch sleep.sh
   
   -----------------------------------------------------------------
               Welcome to the Stampede2 Supercomputer                 
   -----------------------------------------------------------------
   
   No reservation for this job
   --> Verifying valid submit host (login1)...OK
   --> Verifying valid jobname...OK
   --> Enforcing max jobs per user...OK
   [...]
   Submitted batch job 8071232

4. Immediately after sbatch queues the sleep job, submit the dependent job. Use the command-line argument --dependency=after: to specify it as dependent on the job you just submitted:

5. Now look at the status of your jobs in the queue. If you submitted the dependent job soon enough, you may see them both concurrently executing in the queue:

6. Now examine the output files. They will be named <slurm-jobId>.out.

This is exactly what happened in this scenario: the main job started, then the dependent job started, and finally the main job finished.

On your own, repeat the exercise, but this time change the command-line argument to --dependency=after: when submitting the dependent job. This specifies that the dependent job will launch only after the main job terminates. Verify that this is the observed behavior.