Re: Wrong usage of pqMsg_Close message code?

On Tue, Aug 29, 2023 at 10:04:24AM +0900, Tatsuo Ishii wrote:
>> Yeah, both of you are right here. Anyway, it seems to me that there
>> is a bit more going on in protocol.h. I have noticed two more things
>> that are incorrect:
>> - HandleParallelMessage is missing a message for 'P', but I think that
>> we should have a code for it as well as part of the parallel query
>> protocol.
>
> I did not know this. Why is this not explained in the frontend/backend
> protocol document?

Hmm. Thinking more about it, I am actually not sure that we need to
do that in this case, so perhaps things are OK as they stand for this
one.

>> - PqMsg_Terminate can be sent by the frontend *and* the backend, see
>> fe-connect.c and parallel.c. However, protocol.h documents it as a
>> frontend-only code.
>
> I do not blame protocol.h because our frontend/backend protocol
> document also stats that it's a frontend only message. Someone who
> started to use 'X' in backend should have added that in the
> documentation.

Actually, this may be OK as well as it stands. One can also say that
the parallel processing is out of this scope, being used only
internally. I cannot keep wondering whether we should put more
efforts in documenting the parallel worker/leader protocol. That's
internal to the backend and out of the scope of this thread, still..
--
Michael