OK, there is a solution. It was not APR which was high on our suspect
list. Mea culpa. The test servers built to evaluate upgrading from
3.1.3 to 3.1.8 were built on CentOS 5 with SELinux enabled. There,
I've said it. Self inflicted. We don't use SELinux, not sure how it
was missed on the test servers, but not its obvious it should have
made the troubleshooting list.
During the build so much focus was poured into C, Apache, and
potential effects of breaking the mod_caucho build we never went
further. Once I switched my focus to implementing without mod_caucho
I focussed on Apache logs rather than stack traces and debug.
Error logs during mod_caucho testing were along these lines:
mod_caucho.c:934:caucho_request(): no connection: cluster(0x89a0350)
But with only Apache and reverse proxy to deal with, this output
provided a huge lead:
[error] proxy: HTTP: disabled connection for (192.168.1.2)
[error] (13)Permission denied: proxy: HTTP: attempt to connect to
localhost:8080 (192.168.1.2) failed
[error] ap_proxy_connect_backend disabling worker for (192.168.1.2)
That led me to SELinux, and the rest of the tale quickly revealed
itself. It has led me to evaluate using mod_proxy, and potentially
resin.xml for rewrites which will loosen the tying of our various apps
to the one instance of Apache. So its not all bad news.
resin-interest mailing list