Hans-Juergen Schoenig <postgres@cybertec.at> writes:
> i would suggest to replace the existing parameter but something else:
> - a switch to define the global size of the lock pool (e.g. "max_locks")
> - a switch which defines the upper limit for the current backend /
> transaction
The problem with that is that it's pretty much guaranteed to break
pg_dump, as pg_dump always needs a lot of locks. We could perhaps
change pg_dump to increase its limit value (assuming that that's not a
privileged operation), but the fact that a counterexample is so handy
makes me doubt that this is a better design than what we have.
regards, tom lane