On Wed, 12 Jan 2005 21:21:13 +0000 (UTC)
dieter <dieter.NetBSD@pandora.be> wrote:
> On Tue, 11 Jan 2005, D'Arcy J.M. Cain wrote:
> 
> > If no one can help with this, does anyone have any idea why
> > PostgreSQL would suddenly start getting into a busy loop in 2.0 but
> > not RC4 or-current?
> 
> I suspect it has something to do with dlopening modules that use
> pthreads. I saw the same a few days ago, and now that I've built
> postgresql47-server with DLOPEN_REQUIRE_PTHREADS set to "no", it seems
> to keep running.

Are you suggesting that because the server is not built with pthreads to
begin with?  I had that problem before and fixed (kluged) it by adding
the following line to postgresql74/Makefile.common:

.include "../../mk/pthread.buildlink3.mk"

So all postgresql programs get built with threads anyway.

What was the error you were seeing?  Did it keep running but get into a
state that it would not answer queries and could only be killed with -9?
 If not you may have been seeing the same issue we discussed about a
month ago.  That issue was killing the backend in certain situations. 
In my case it was when using plpgsql because that caused it to
dynamically load a threaded library.  The above fix took care of that by
making everything start out linked with libpthread to start with.

This still would not explain why the exact same setup fails on 2.0
release but not 2.0 RC4 or -current.

-- 
D'Arcy J.M. Cain <darcy@NetBSD.org>
http://www.NetBSD.org/