I see in the list archives that some people have experience problems
(given up?) on pound running well after being compiled with Redhat 9.

My problem is that pound will proxy a few requests as intended (usually
about 10), then hang.  It is still listening on the incoming port, but
snooping reveals no forwarding request is made on the reverse proxy
side.  Only a kill and restart will make it serviceable (and then only
again for a handful of requests).

Could someone suggest a way in which I could try and debug this?  The
suggestion that it may be the new back-ported threading libraries
doesn't make so much sense to me given that I'm running from-source
load-tested builds of Apache 2, Zope (Python), Postgresql, OpenLDAP
across half a dozen SMP-based RH9 hosts without problem (all glibc
updates applied).  Could there be a difference?

On a side note - there are configure or Makefile problems.  If you
compile with --prefix=/apps, then a make install copies pound to
/apps/sbin (rather than /apps/sbin/pound) and the program barfs on the
lack of a man/man8 directory.  Basically you have to make etc, sbin and
man/man8 by hand, then do a make install should you wish to use
--prefix.

Thanks, Nick

I tried and was not successful--when you resolve the problem, please let
me know.

As I remember, there were two problem areas in RH9 (assuming that you've
upgraded to the bleeding edge): the NPTL and ssl. Since you have gotten
Pound to compile, the SSL problems must have been resolved.  (Or did you 
build and install SSL from source?)

Given the behavior (I saw pretty much the same thing you did) I suspected
the NPTL.  One possible clue--folks running on uniprocessors don't appear
to see the problem.  Could it be that Pound needs to be run with processor
affinity enabled?


On Tue, 23 Mar 2004, Nick Bower wrote:
[...]

On Tue, 2004-03-23 at 08:15, Dennis Allison wrote:[...]

Try running it like this:

$ LD_ASSUME_KERNEL=2.4.19 /usr/bin/pound

which will force pound to use LinuxThreads rather than NPTL.

Sean

I'll do some experimentation and report my results in a day or so.  

On Tue, 23 Mar 2004, Nick Bower wrote:
[...]

Didn't even bother with Redhat's SSL, particularly because of
vulnerabilities like last friday.  Didn't realize about the SSL though.

I'm using affinity (in the form of a small C program) to attach init
derived processes to CPU 0 and various servers to the others (4 virt, 2
real that is).  I wouldn't know where to start guessing.

On Tue, 2004-03-23 at 17:15, Dennis Allison wrote:
[...]
[...]