Job chains

Deploying dependent jobs (job chains)

In certain scenarios, it might be desirable to condition the execution of a certain job on the status of another job. In such cases, the sbatch directive --dependency=<condition>:<jobID> can be used, where <condition> specifies the type of dependency (See table 2), and <jobID> is the slurm jobID upon which dependency is based. To specify more than one dependency, the , separator is used to indicate that all dependencies must be specified, and, ? is used denotes that any dependency may be satisfied.

For example, assume the slurm job scripts, job_1.sbatch, … job_3.sbatch need to run sequentially one after the other. To start this chain, submit the first job and obtain its jobID:

$ sbatch job_1.sbatch
Submitted batch job 8580135

Next, submit the second job to run only if the first job is successful:

$ sbatch --dependency=afterok:8580135 job_2.sbatch
Submitted batch job 8580136

And, now, to run the third job only after the first two jobs have both run successfully:

$ sbatch --dependency=afterok:8580135,8580136 job_3.sbatch
Submitted batch job 8580140

Alternatively, if the third job is dependent on either job running successfully:

$ sbatch --dependency=afterok:8580135?8580136 job_3.sbatch
Submitted batch job 8580141