Hi Brian, Thanks for the response.
I had previously tried this, but setting the default gateway on the web servers to point to the HAProxy server's eth1 results in the web servers losing all external connectivity, as the source address is always a private address. root@web:~# ping -c 5 8.8.8.8 PING 8.8.8.8 (8.8.8.8) 56(84) bytes of data. ^C --- 8.8.8.8 ping statistics --- 5 packets transmitted, 0 received, 100% packet loss, time 4008ms root@haproxy:~# tcpdump -n -i eth1 icmp tcpdump: verbose output suppressed, use -v or -vv for full protocol decode listening on eth1, link-type EN10MB (Ethernet), capture size 96 bytes 16:33:57.024999 IP 192.168.3.65 > 8.8.8.8: ICMP echo request, id 6180, seq 1, length 64 16:33:58.034080 IP 192.168.3.65 > 8.8.8.8: ICMP echo request, id 6180, seq 2, length 64 16:33:59.034036 IP 192.168.3.65 > 8.8.8.8: ICMP echo request, id 6180, seq 3, length 64 16:34:00.034104 IP 192.168.3.65 > 8.8.8.8: ICMP echo request, id 6180, seq 4, length 64 16:34:01.034266 IP 192.168.3.65 > 8.8.8.8: ICMP echo request, id 6180, seq 5, length 64 Any other ideas? I'm currently running a similar setup to load balance a mail cluster that's been in place for almost 4 years. The HAProxy servers use an older kernel with the cttproxy patch. The mail servers all receive connections from the HAProxy boxes on their eth1 interfaces and route back out to the them on their eth0s - without any iptables rules or routing tables. Thanks, REW On Tue, Apr 12, 2011 at 4:24 PM, Brian Carpio <bcar...@broadhop.com> wrote: > Randy, > > The problem is the gateway on the backend webservers needs to be set as a > VIP (or eth1 interface) on the HAproxy servers on their private interface > (assuming you have two HAproxy servers and are using heartbeat for > failover). It looks like from your routing table that eth0 on the > webservers’ gateway is pointed to the eth0 interface on haproxy this is why > it works perfectly when you configure haproxy to use the public IPs on the > webservers. > > > > Once you change the default gateway on the backend webserver’s to use eth1 > on the haproxy server (or a VIP which lives on eth1 using heartbeat for > failover between two haproxy servers) then it will work. > > > > *Brian Carpio * > > *Senior Systems Engineer* > > *[image: Description: Description: BroadHop Home Page]* > > *Office: +1.303.962.7242* > > *Mobile: +1.720.319.8617* > > *Email: bcar...@broadhop.com* > > > > *From:* Randy Wilson [mailto:randyedwil...@gmail.com] > *Sent:* Tuesday, April 12, 2011 8:29 AM > *To:* haproxy@formilux.org > *Subject:* Tproxy with multiple interfaces > > > > Hi, > > I'm trying to setup an HAProxy instance to transparently load balance a > group of web servers. The HAProxy server and web servers each have two > interfaces; eth0 as the public interface and eth1 the private. I'm trying to > configure the load balancer to accept requests on port 80 on eth0 and > transparently proxy the connections to the web servers over the private > interfaces on eth1. I've configured the load balancer in the normal way for > tproxy, and have the web servers routing out through it. > > i.e. > > iptables -t mangle -N DIVERT > iptables -t mangle -A PREROUTING -p tcp -m socket -j DIVERT > iptables -t mangle -A DIVERT -j MARK --set-mark 1 > iptables -t mangle -A DIVERT -j ACCEPT > > ip rule add fwmark 1 dev eth0 lookup 100 > ip rule add fwmark 1 dev eth1 lookup 100 > ip route add local 0.0.0.0/0 dev lo table 100 > > root@haproxy:~# sysctl net.ipv4.ip_forward > net.ipv4.ip_forward = 1 > > root@web:~# route -n > Kernel IP routing table > Destination Gateway Genmask Flags Metric Ref Use > Iface > xxx.xxx.97.0 0.0.0.0 255.255.255.0 U 0 0 0 > eth0 > 192.168.0.0 0.0.0.0 255.255.252.0 U 0 0 0 > eth1 > 0.0.0.0 xxx.xxx.97.155 0.0.0.0 UG 0 0 0 > eth0 > > ... > server web1 192.168.3.65:80 source 192.168.3.64 usesrc clientip > ... > > When testing this setup, the connection is correctly proxied to the web > server through eth1 and a response is sent back to the HAProxy server on > eth0, but it's ignored and the connection hangs. > > > Here's some netstat output during the connection: > > HAProxy: > > Proto Recv-Q Send-Q Local Address Foreign Address > State PID/Program name > tcp 0 1 xxx.xxx.48.1:42424 192.168.3.65:80 > SYN_SENT 1386/haproxy > tcp 0 0 xxx.xxx.97.155:80 xxx.xxx.48.1:42424 > ESTABLISHED 1386/haproxy > > Web Server: > > Proto Recv-Q Send-Q Local Address Foreign Address > State PID/Program name > tcp 0 0 192.168.3.65:80 xxx.xxx.48.1:42424 > SYN_RECV - > > > And the relevant entries from a tcpdump on each interface on each server: > > HAProxy/eth0 IP xxx.xxx.48.1.42424 > xxx.xxx.97.155.80: Flags [S], seq > 1021535895, win 5840, options [mss 1418,sackOK,TS val 2448718 ecr > 0,nop,wscale 7], length 0 > > HAProxy/eth0 IP xxx.xxx.97.155.80 > xxx.xxx.48.1.42424: Flags [S.], seq > 504489330, ack 1158274356, win 5792, options [mss 1460,sackOK,TS val 230477 > ecr 1407043,nop,wscale 7], length 0 > > HAProxy/eth0 IP xxx.xxx.48.1.42424 > xxx.xxx.97.155.80: Flags [.], ack 1, > win 46, options [nop,nop,TS val 1407043 ecr 230477], length 0 > > HAproxy/eth1 IP xxx.xxx.48.1.42424 > 192.168.3.65.80: Flags [S], seq > 391399045, win 5840, options [mss 1460,sackOK,TS val 230550 ecr 0,nop,wscale > 7], length 0 > > Web/eth1 IP xxx.xxx.48.1.42424 > 192.168.3.65.80: Flags [S], seq 391399045, > win 5840, options [mss 1460,sackOK,TS val 230550 ecr 0,nop,wscale 7], length > 0 > > Web/eth0 IP 192.168.3.65.80 > xxx.xxx.48.1.42424: Flags [S.], seq > 4033028970, ack 391399046, win 5792, options [mss 1460,sackOK,TS val 6751967 > ecr 230550,nop,wscale 7], length 0 > > HAproxy/eth0 IP 192.168.3.65.80 > xxx.xxx.48.1.42424: Flags [S.], seq > 4033028970, ack 391399046, win 5792, options [mss 1460,sackOK,TS val 6751967 > ecr 230550,nop,wscale 7], length 0 > > > NB. In this example I had set the usesrc setting to client so the client's > port was used for readability, but the same occurs with clientip. > > I'm sure this is occurring because the response connection is arriving on a > different interface to the one HAProxy originated the connection to the web > server on. Does anyone know of a way around this? Is there an iptables or ip > rule that can be set to switch the return traffic from eth0 to eth1? > > I have tried testing the setup by proxying the connections to the web > server's public interface on eth0 instead: > > ... > server web1 xxx.xxx.97.156:80 source xxx.xxx.97.155 usesrc clientip > ... > > And the transparent proxying works perfectly. > > Any thoughts or suggestions appreciated. > > > Many thanks, > > REW >
