Andres Freund <andres@anarazel.de> writes:
> Worth noting that this doesn't necessarily suffice to avoid race conditions in
> tests, if the test depends on *other* backends having seen the configuration
> changes.
True, but do we have any such cases?
> It might be worth to use the global barrier mechanism to count which backends
> have reloaded configuration and to provide a function / option to pg_sleep
> that waits for that.
That ... seems like a lot of mechanism. And it could easily result
in undetected deadlocks, if any other session is blocked on you.
I seriously doubt that we should go there.
regards, tom lane