yeah, this is what i found strange because i did not do anything yet to try and suppress the default 5.9 behavior which should print the backtrace.
my first suspicion is this might be the same buffering problem i’ve had before with print
and journalctl
. i need to look into this more.
it smells to me like a query-of-death situation. i don’t know how to reproduce this locally, because it likely only happens on a single page that exists in the production database. the first step is to get the backtrace to flush to disk before the process exits.