During production job failures are to be expected either due to infrastructure failures such as network or node failures but also due to faulty applications. The goal is to understand:

• what are conditions making Pegasus to consider a job as failed one,
• what happens after a job enters a failed state, i.e.:
  • how a failure affect transferring job's intermediate/output files (if any),
  • what happens if failure occurs in a job cluster,
• how retries work in practice, i.e.:
  • can an operating organization control number of retries,
  • can an operating organization make Pegasus retry on certain failures but not other,