Re: PostmasterContext survives into parallel workers!?

Andres Freund <andres@anarazel.de> writes:
> Personally I think the whole logic should be reworked so we do most of
> that that only from one place.

Alvaro already mentioned a couple of reasons why that might not be so
easy.

> Especially the signal handler stuff
> should imo just be replaced by setting latch, which then does the work
> inside the normal main loop.

Of course, then we're utterly dependent on the latch logic to be
zero-failure, as any bug in it takes out the postmaster.  That idea
would've been rejected out of hand earlier in the development of
the latch code.  Maybe it's safe enough today, but I still wonder what
it is that we're buying after we do such a massive rewrite of the
postmaster.
        regards, tom lane