Hi, On Wed, Aug 04, 2010 at 12:44:33PM +0800, overred wrote: > Hi, > I am a Haproxy user.and had some problomes. > > my load balancer policy is: > #haproxy.cfg > global > daemon > maxconn 409600 > chroot /usr/local/haproxy > uid 99 > gid 99 > #nbproc 4 > #debug > #quiet > defaults > > balance hdr(X-Up-Calling-Line-ID) > hash-type consistent > # hash-type map-based > mode http > #option httplog > #option dontlognull > option forwardfor > retries 5 > #option httpclose > option dontlognull > option forwardfor > no option redispatch > > maxconn 409600 > contimeout 5000 > clitimeout 50000 > srvtimeout 50000 > #frontend http-in > #j2me-server > # bind *:80 > listen j2me-server 0.0.0.0:80 > # cookie SERVERID rewrite > # balance roundrobin > # capture request header X-Up-Calling-Line-ID len 11 > # balance hdr(hdr_sub(X-Up-Calling-Line-ID)) > # balance source > balance hdr(X-Up-Calling-Line-ID) > hash-type consistent > stats uri /site-stats > stats realm Haproxy\ statistics > stats auth fetion:fetion > no option redispatch > cookie SERVERID insert indirect > option httpchk HEAD /test.htm HTTP/1.0 > # server j2me89_1 192.168.147.10:80 check inter 2000 rise 2 fall 5 > server j2me89_1 192.168.98.89:80 check inter 2000 rise 2 fall 5 > server j2me90_2 192.168.98.90:80 check inter 2000 rise 2 fall 5 > server j2me91_3 192.168.98.91:80 check inter 2000 weight 3 rise 2 > fall 5 > server j2me92_4 192.168.98.92:80 check inter 2000 weight 3 rise 2 > fall 5 > server j2me93_5 192.168.98.93:80 check inter 2000 weight 3 rise 2 > fall 5 > server j2me94_6 192.168.98.94:80 check inter 2000 weight 3 rise 2 > fall 5 > server j2me102_7 192.168.98.102:80 check inter 2000 weight 3 rise 2 > fall 5 > server j2me103_8 192.168.98.103:80 check inter 2000 rise 2 fall 5 > server j2me104_9 192.168.98.104:80 check inter 2000 rise 2 fall 5 > server j2me105_10 192.168.98.105:80 check inter 2000 rise 2 fall 5 > > > the hash type is: balance hdr(X-Up-Calling-Line-ID),"X-Up-Calling-Line-ID" > is a http request header like : > GET /test.aspx HTTP/1.1 > X-Up-Calling-Line-ID: 13888888888 > > > but 200 balance is wrong with 200,000 requests(the X-Up-Calling-Line-ID is > random). > this 200 X-Up-Calling-Line-ID load to the different realserver.
I don't understand what you mean here. Do you mean that when the ID equals 200 it does not always go to the same real server ? > In principle,will load to the same realserver with the same header of > X-Up-Calling-Line-ID(hash-type consistent too). Yes, in principle, as long as none of your servers fail, a same header value will always go to the same server. If this is not the case for you, you should check your logs to ensure that your servers are not randomly failing. Well, check your stats instead, sine your logs are not enabled ;-) > Is something wrong with my haproxy.cfg or other? The only thing I see is the very high maxconn which probably does not match what you need because reaching such levels requires very fine system tuning. Ah I see something. You have disabled "option httpclose", which means that only the first request of each connection is analysed. If your clients have keep-alive enabled, then it is possible that you observe incorrect balance for the 2nd and up requests of each connection. You'd better use "option http-server-close" so that client keep-alive remains active, still allowing haproxy to parse all requests. Regards, Willy
