Currently, only the 3.2 Upgrade Document recommends disabling core dumps in production systems.

Could we also update the Production notes and/or the Operations Checklist to include the recommendation to disable core dumps on production systems?

On the Unix ulimit Settings page there is a sample output (which includes a core limit of 0) which looks like a recommendation but it's not really clear.

This is usually done with ulimit -c 0.

Under Ubuntu, if the sysctl kernal.core_pattern = |/usr/share/apport/apport %p %s %c (which is the default core_pattern) then apport would be triggered and would write a crash report even if core ulimit is set to 0 (documented on apport webpage). In addition to writing the crash report, if a process crashes, it must wait for apport before it can finish exiting, which might take some time.

One way to stop apport from being triggered is to set:

sudo sysctl -w kernel.core_pattern="core.%e.%p.%h.%t"