So this sounds like a regression in 2.0.47 with worker. Coming from the
debian POV, there is no difference in build environment between the two
builds, and no patches are applied to the mpms.
Just a heads up, I'll try and dig further in a bit.
-Thom
--- Begin Message ---
Hi Thom May and Cliff Woolley,
Cliff I have BCCed you this message because you tried to help on the
official apache devel mailing list.
I have finally found a way to work around what causes my server to
create detached and defunct processes:
http://lists.debian.org/debian-apache/2003/debian-apache-200307/msg00065.html
Cliff gave some suggestions for code to revert on the devel
mailing list on 21 July (thanks Cliff):
- server/core.c r1.225.2.7 (not likely)
- modules/generators/mod_cgid.c r1.145.2.6 (possible but unlikely)
- srclib/apr/file_io/unix/readwrite.c r1.86 (possible)
I tried reverting these changes individually (to 2.0.46). Note that
reverting core.c wasn't successful because it now defines
ap_is_recursion_limit_exceeded, so I reverted request.c instead. I also
tried reverting ssl_engine_kernel.c. The problems persisted.
Noting that some of the mpm code had changed (prefork code, not worker
code) I decided to try changing from the mpm-worker package to
mpm-prefork. All the problems went away and now my server is working
with Debian's 2.0.47-1 packages just as it worked with previous packages
(though always using mpm-worker).
If the cause of this regression is ever tracked down please let me know.
I had always used the Apache2 worker model and I thought it was stable.
But I'm glad to find an alternative that works.
Regards,
Adam
--- End Message ---
