On Fri, Jul 20, 2001 at 03:18:51PM -0700, GUMMALAM,MOHAN (HP-Cupertino,ex2) wrote:
> This tar-ball doesn't contain the POD patches that Justin had released! I'm
> seeing the symptoms of the 2.0.20 problem on 2.0.21 as well. Apache starts
> up with 3 processes, reduces to 0, and then comes back to 1 (not including
> the watchdog process, and the cgid process). I guess, I do not need to
> explain why that is happening, as we all went over it together last week.
>
> Is the patch going to be included in 2.0.22?
AFAIK, there are two outstanding problems with the threaded MPM:
- The default config sucks. StartServers is 3, MaxThreadsPerChild is
25, and MaxSpareThreads is 10. This is bogus. That's part of the
problem you are seeing. I've submitted a patch for that. Not much
more I can do. You should be able to change it manually.
- Pipes are a tad hyper from everything I can tell here on Solaris.
I'm not sure what is going on - I see a read from a pipe that is
returning success (0) and a length of 0. The threaded MPM would treat
that as if a POD were received (it wasn't since we would have received
a length of 1).
This behavior contradicts the Solaris man pages for when O_NONBLOCK
is set on a pipe (but looks correct IF we had set O_NDELAY). Jeff
Trawick and I are going back and forth to try and figure out why we
need this one-line patch (i.e. check the length of what we read) in
the first place. We probably need people to try and review my posted
testpipe.c (see [EMAIL PROTECTED]) on their respective platforms to
see if this condition occurs or if I am doing something wrong in my
test (very possible).
I said +0 on httpd 2.0.21 T&R for these reasons. I'd *like* to see
this behavior fixed before we do another T&R. But, I'm not a
committer - so take what I say with a grain of salt. -- justin
