Hi, You can either play with :
- balance url_param ld - sticky table, and then set the param you want to stick on : stick on urlp(ld) table ... - appsession ld len 3 timeout <holdtime> request-learn mode query-string w/ something like : server server_A ... cookie lb=100 weight 10 server server_A ... weight 10 But I think the most simple configuration for you here is to use multiple backend if you can : frontend App ... acl url_ld100 url_sub ld=100 use_backend bk_ld100 if url_ld100 default_backend default backend bk_ld100 ... balance roundrobin server server_A <ip:port> check server server_B <ip:port> check backup backend default ... balance roundrobin server server_A <ip:port> check server server_B <ip:port> check It means that, when your front parse ld=100 in the url query string, it matches the acl url_ld100, then use a dedicated backend : bk_ld100. In this backend only server A server request since server B is defined as a backup server. If A goes down, B will answer request. So you need "check" on your servers. If ld=100 is not matched, your request will end to the default server, which will roundrobin btw A and B. Regards, On Wed, Sep 3, 2014 at 11:47 PM, S. Zhou <[email protected]> wrote: > We are thinking of the following LB algorithm but we are not sure if current > HAProxy supports it: > given a http request, LB should always forward it to a certain backend > server (say Server A) based on its http parameter (e.g. request with > parameter "Id=100" always go to server A). The only exception is: when the > designated server (e.g. Server A) is down, then the request should be > forwarded to another (fixed) server (e.g. Server B). > > Thanks > -- Steven Le Roux Jabber-ID : [email protected] 0x39494CCB <[email protected]> 2FF7 226B 552E 4709 03F0 6281 72D7 A010 3949 4CCB
