Backend selection, "sticky" connections?

Hello,

i'm experiencing a strange (to me) problem/behavior with haproxy.

HAProxy accepts http connections (default mode http) one a frontend, and based upon the user-agent, uses different backends for the request. This basically works. BUT:

Sometimes, it does not. Then the request gets processed by the wrong backend, and in my case, resulting in bad errors. I'm using this setup to be able to use the same host & url both for production and staging/testing environment for our webpage.

I assume that somehow haproxy uses some still existing connection, maybe based on the requesting ip?

default configuration:

        log global

        mode http
        retries 3
        contimeout 5000
        clitimeout 50000
        srvtimeout 50000

        balance roundrobin
        option httpchk GET / HTTP/1.0
        option redispatch
        option abortonclose
        option httpclose
        option httplog

frontend and backend(s) have no additional configuration. (Except from the acl's necessary to choose the correct backend in the frontend)

If you have any ideas what would be causing this, i would highly appreciate it ;)

Thanks & Regards

Moritz Received on 2010/10/20 19:05