login5 ran out of memory yesterday (27.02.2017) around 18:16 and took about 15 minutes to recover.

During this time the compute nodes were unable to contact the application scheduler running on login5 and some jobs might have crashed.
A typical error message for this case is: "aprun: Apid nnnnnnn: close of the compute node connection after app startup barrier".

We apologise for any inconvenience caused.