Hi Baptiste, Thanks for your help and your analyse.
I will test the cookie, I first need to read the documentation :-) Best regards, Christophe On 26/03/13 15:37, "Baptiste" <[email protected]> wrote: >> backend Managers >> mode http >> server Manager01 192.168.0.60:80 check on-marked-down >>shutdown-sessions >> server Manager02 192.168.0.61:80 check on-marked-down >>shutdown-sessions >> server Manager03 192.168.0.62:80 check on-marked-down >>shutdown-sessions >> option httpchk HEAD /cfadmin/ping.cfm HTTP/1.0 >> stats enable >> balance source > > >Hi Christophe, > >There were two main reason why you could observe such behavior: >#1 cookie persistence with several server sharing the same cookie value >#2 a source IP hash persistence > >With source hash, you stick a session to a server as long as the >number of servers in the farm remains the same. >When it changes, then the hash changes and everybody is balanced... >A better solution would to use consistent hashing (hash-type >consistent). But using this method, you may still balance people >connected to the failed server. And balance them back to the server >when it wakes up... >For source IP persistence, the best solution would to use a >round-robin algorithm and a stick table for persistence. that way, >when server goes down, you're balanced, that's normal, when the server >comes back, only new sessions can reach him. Sticked one remains on >their attributed server. > >Actually, the best would to do cookie based persistence, either >through cookie insert or cookie prefix. > >Baptiste >
