Rainer Jung wrote:
On 06.03.2009 08:39, Mladen Turk wrote:
Can you cc the dist files to your people's dir as well
Did that:
http://people.apache.org/~rjung/mod_jk-dev/
OK, I'll build the bins from there
But as described above, it would be nice to first check BZ 46808 before
building binaries.
I saw your comment, but also added another one. Please have a look.
BZ 46808 is valid but it brings us back where we were before.
It solves (well doesn't actually) one thing,
but breaks the sticky sessions.
Unless the patch reliably detects the cause of failure
I'm -1 for committing that. The problem is that returned
codes from socket operations are platform specific, and
then what if some one pulls the cable in the middle of
AJP request? The possibilities are endless, and trying to
solve each and every one is over engineering thought.
We should keep the things simple:
"If there are opened connections to the backend do not
mark that worker in error"
We simply do not know the status of the network beyond
the current connection. It might be Tomcat rejecting
new connections, firewall reaching the limit, ulimit,
ssh tunnel restart, etc, etc ...
So although our connection fails other might still
work. If we can for sure detect something as absolute
failure then we could save some concurrent client
cycles, but then again we have a watchdog that can
check all the client connections in background.
Regards
--
^(TM)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org
For additional commands, e-mail: dev-h...@tomcat.apache.org
