HOWTO make an error report

When reporting an error or problem encountered on HPC machines we kindly request that you:

  1. Tell us the machine you were working on
    yes, you have one and only one account, so you were obviously on <younameit>; and yes, of course we know how to find out where your only account is located; but it is a (relatively) long and boring work for us, while it costs nothing at all to you

  2. Always attach the complete error message
    we can do really little with generic reports of "MPI errors" or "memory allocation problems"

  3. Always tell us the job number
    this is the most important piece of information

  4. Always attach the complete PBS script
    not just the resources, or just the launch command; if you work interactively, post the session history on the masternode ("history" command); if your job is qsub'ed by another script, please attach that script too

  5. Add optional information
    information that you consider helpful for us to pinpoint (and possibly solve) your problem
    e.g.: nodes on which your job is/was running, compiler and optimization flags used for compilation, libraries linked, ...

  6. Use e-mail, not phone
    ALWAYS send your reports to <cluster-admin AT democritos.it>, error codes are quickly corrupted by AWGN on phone lines (and silently dropped by content filters on private accounts)

Please note that it is really difficult for us to diagnose problems if we do not know where and when they happened, so please follow this micro-HOWTO. Error reports not complying to these guidelines will be dropped without further action.

Thank you for your understanding.

Further reading: