> -----Original Message-----
> From: [EMAIL PROTECTED]
> Modified: jk/native2/common jk_worker_lb.c
> Log:
> Introduced 3 new configuration parameters to the lb.
> 1. attempts -- replaces harcoded MAX_ATTEMPTS
> 2. recovery -- replaces hardcoded WAIT_BEFORE_RECOVER
> 3. timeout -- this one is new.
> The timeout if set will force the lb to cycle through
> workers if all are in the
> error_state for the specified amount of seconds. This is usefull for
> situations when the TC is overloaded and refuses new connections.
> The lb will wait and after timeout will report 500 to the client.
>
Some more clarification:
The reason for introducing timeout inside lb is from my point of view
and the applications I'm using jk2 for, fact that even using most
powerful system, there will be situations that returns 500 to the client
simply because the TC is to busy and refuses new connections. One way to
solve that is to force the TC to allow more connections eating system
resources, and the other is using lb timeout.
One drawback of using timeout is the fact that the client will get
delayed response even for the non-existent configured TC instance.
I've made such behavior at first inside the socket connector, but Nacho
(he was right) made a strong objection to that. So instead trying to
connect to the instance from connector point of view, all that is put
inside the load balancer.
MT.
--
To unsubscribe, e-mail: <mailto:[EMAIL PROTECTED]>
For additional commands, e-mail: <mailto:[EMAIL PROTECTED]>
