"David G. Johnston" <david.g.johnston@gmail.com> writes:
> Or not, since it does appear that the reload signal is propagated to active
> sessions and take effect after the most recent command finishes.
Yeah. I had been wondering about long-lived open transactions, but AFAICS
from the code, backends should re-read the config file at the next client
command submission, whether inside a transaction block or not. So the
thing to look for is what might be overriding the config file's value.
In interactive sessions, examining the pg_settings view would be a
promising way to debug that. I suspect though that the OP is guessing
about what's happening inside application-driven sessions, where it would
be hard to do that kind of debugging :-(
regards, tom lane
