One very tired fish, a bit overdue to be soup.
Cambridge Fishy… Fishy fishy fishy…
Oh shit! It’s mr. Creosote! Play dead.
I recently did a portupgrade -ra on my server after some period of complacence. It was instigated by having to clean out my mySQL logs after they ate up 30GB of disk space and caused some table corruption.
Anyway, the key details were that
apache+mod_ssl-1.3.37+2.8.28 > needs updating (port has 1.3.39+2.8.30)
php5-5.2.3 > needs updating (port has 5.2.4)
(among about 50 others)
Some foreshadowing.. once I updated and rebooting I get in /var/log/messages only
kernel: pid 1127 (httpd), uid 0: exited on signal 11 (core dumped)
and in /var/log/httpd-error.log only
[info] mod_unique_id: using ip addr 220.127.116.11
every time I “
apachectl start” (and after setting apache.config log level to “debug”)
Much email searching ensued, but Torfinn Ingolfsen on the free-bsd-stable mailing list suggested looking at PHP. Turned out disabling php.so in httpd.conf got apache sort of working, but that was no help, so I thought, eh, why not migrate to apache 2.2.6?
That helped a lot. First the default config didn’t get run with SSL (crash) but that was hinted in the config files
Oct 02 11:30:26 2007] [info] mod_unique_id: using ip addr 18.104.22.168
[Tue Oct 02 11:30:27 2007] [info] Init: Seeding PRNG with 136 bytes of entropy
[Tue Oct 02 11:30:27 2007] [info] Loading certificate & private key of SSL-aware server
[Tue Oct 02 11:30:27 2007] [error] Server should be SSL-aware but has no certificate configured [Hint: SSLCertificateFile]
so I disabled that for the moment. But I was also getting periodic seg faults from Apache. No details (even less than with 1.3.39). Disabling PHP made them go away, but at least apache 2 was self-restarting, so aside from log pollution, no problem…
It occurred to me that my make.conf -O2 compiler specification might be part of the problem, so I changed just that to -O1 and recompiled with portupdate -rf PHP and no more seg faults. 5.2.3 had no trouble with -O2, but 5.2.4 doesn’t seem stable with O2 optimization.
The SSL problem was that /usr/local/etc/apache22/httpd.conf had a bit at the end about the following being present to support starting SSL… blah blah, 3rd line from the bottom “SSLEngine on.” It was turning on the engine twice since I was using extra/httpd-ssl.conf already. I commented that line out and everything seems fine now.