Re: BUG #18469: OOM occurs and backend processes are kept in Zombie state.

PG Bug reporting form <noreply(at)postgresql(dot)org> writes:
> I was performing a lot of operations on a server deployed with postgresql
> 12.16. As heavy operations performed continuously. memory consumption has
> been increased, the OS eventually got OOM and some background connection
> processes that were taking up too much memory were killed. However, these
> processes were not successfully killed and remained in Zombie state. In the
> meantime, the whole database process seems to be stuck and time out happened
> while connect via psql.

It sounds to me like the OOM killer decided to kill the postmaster
process, rather than the child process(es) that were actually eating
memory. That's *extremely* unhelpful behavior. There is some advice
in our manual about configuring your system to not do that.

> Below is the status after OOM happened:
> Ruby 7822 0.0 0.6 4485088 110940 の May06 10:24 /usr/pgsql/bin/postmaster -D /var/lib/pgsql/data

It's not clear to me where this postmaster process came from,
but it appears to be younger than the other postgres-related
processes you're showing, so they are not its children.

I'd manually nuke all of these processes and start fresh.

regards, tom lane