Re: pipe chunks protocol

Tom Lane wrote:
> Andrew Dunstan <andrew(at)dunslane(dot)net> writes:
>
>>> This patch implements the protocol Tom suggested for writing to the
>>> syslogger pipe. It seems to pass my tests (basically "make
>>> installcheck" against a server with stderr redirection turned on and
>>> log_statement set to 'all').
>>>
>
> I didn't like this patch much --- it broke the API of
> write_syslogger_file, which is supposed to just write what it's given
> (and it is used from outside syslogger.c with that expectation). Also
> the way it slung unconsumed data back and forth between two buffers
> seemed both confusing and inefficient. Here's a revised version.
>

Well. it was like the curate's egg :-) Anyway, thanks for the cleanup.

> In my testing, I found that a standard "make installcheck" run produces
> only one message large enough to be split (the "infinite_recurse" thing
> in errors.sql), so this is definitely not a Good Enough test. Maybe
> we could get Ed L. or one of the other complainants to try it.

Yeah, what I did was to wind back the chunk size - try 128 and you'll
see plenty of chunked messages :-) But we really need to do this with
installcheck-parallel to exercise it properly.

> (The
> patch seems to need some adjustment to apply against 8.2, though.)
>
>
>

I know we normally try not to do this, but I'd be happy to wait for the
back branches in this case.

cheers

andrew