[Pgpool-general] More fun with pgpool errors

[Tatsuo Ishii]

> > Oh I see. However it seems it's hard to attack the problem. I think
> > even PostgreSQL has the same problem. If postmaster dies hard enough,
> > backend(postgres) processes will remain.
> 
> Yes and no.  It's true that the postmaster can die and leave children
> around; but the interlock prevents the postmaster from restarting,
> and the children can't answer _new_ connections (since the new
> connections are always a new child process).  What Brad has seen is
> that the children pgpool stay around _and answer new connections_,
> which makes things much worse.

Question is die "hard" is how hard. If parent pgpool is killed in a
truely hard way (killed by SIGKILL) there's no way to prevent that
kind of situation. I believe postmaster behaves same if it is killed
by SIGKILL, i.e. new postmaster will not start.
--
Tatsuo Ishii