We had a first class on our OnDemand server last Tuesday which resulted in about 30 users simultaneously logging into the server. It did not go that well, and dmesg shows the following:
[Tue May 12 11:55:25 2020] Pid 8081(ps) over core_pipe_limit
[Tue May 12 11:55:25 2020] Skipping core dump
That gave us a hint to change the core_pipe_limit to unlimited from the default of 4, in the sysctl.conf file. We increased the maximum available number of file handles from 6000 to 60000.
That seems to have taken care of this issue as today the class is not complaining (yet).
Since I haven’t seen any notion of these settings in the docs, I am wondering if we just happened to have them too low from the start, or, what other people do when they set up the OOD server.