Re: Persistent wait event sets and socket changes

Hi Craig,

On 2017-07-31 14:08:58 +0800, Craig Ringer wrote:
> Hi all
>
> I've been looking into the wait event set interface added in 9.6 with an
> eye to using it in an extension that maintains a set of non-blocking libpq
> connections to other PostgreSQL instances.
>
> In the process I've been surprised to find that there does not appear to be
> any interface to remove a socket once added to the wait event set, or
> replace it with a new socket.
>
> ModifyWaitEvent(...) doesn't take a pgsocket. There doesn't seem to be a
> way to remove an event from the set either.

Yea, and what's even more annoying, you can't "disable" waiting for
readyness events on a socket, we've an assert that we're waiting for
something.

> Discussion in
> https://www.postgresql.org/message-id/flat/20160114143931.GG10941%40awork2.anarazel.de
> talks about a proposed SetSocketToWaitOn(...)/AddSocketToWaitSet and
> RemoveSocketFromWaitSet etc. But it looks like it petered out and went
> nowhere, apparently mainly due to not being needed by any current core
> users.

I think it just needs somebody to push this forward.

> I'd like to add such interfaces at some point, but for now can work around
> it by destroying and re-creating the wait event set when the fd-set
> changes. So I'm posting mostly to confirm that it's not supposed to work,
> and ask if anyone thinks I should submit a comment patch to latch.c
> documenting it.

It doesn't work, right. I'm not sure I see the point of adding comments
explaining that a nonexistant interface doesn't exist?

Greetings,

Andres Freund