Neil Conway <nconway@klamath.dyndns.org> writes:
> I don't see a major problem with allowing postgres to login if the
> connection limit is hit (although I'm not sure it's worth the worry,
> when 'kill a backend executing SELECT ; psql template1 postgres' works
> as-is).
max_connections is a hard limit; you do not have the option of letting
people in anyway, because there'll be no PROC slot for them.
We could consider establishing a "soft" connection limit that's somewhat
less than max_connections, and allowing non-superusers to log in only
if the soft limit hasn't been exceeded. This does not guarantee that
superusers can always get in: the extra slots might have been filled by
other superuser connections. But it'd give them better odds than the
rabble.
I tend to concur with Neil that the usefulness of such a feature is
dubious. But OTOH such a practice has always existed for Unix disk
space --- maybe we should respect that precedent.
regards, tom lane