How does loadbalancers like haproxy and keepalived implement failover/high-availability for themselves, in case the machine they are running dies?

Question

I'm trying to understand how my load balance software does not become a single-point-of-failure. Do load-balancing software usually support failover/ha for themselves?

They usually don’t, you dot hat by failover solutions on the machine (keepalived), virtualized or with the help of router/switch. It works best if no session state needs to be kept (this works well for cookie based sticky http sessions) — eckes, Oct 14 '17 at 14:49

score 2 · Answer 1 · answered Oct 14 '17 at 14:43

2

Ideally, you would use a pair of clustered load balancers in a active/passive failover configuration.

answered Oct 14 '17 at 14:43

ewwhite

197,159
92
443
809

score 2 · Answer 2 · answered Oct 16 '17 at 06:37

Yes, Traffic is configured to hit a Floating Virtual IP, when the master load balancer fails the slave detects this, brings up the Floating IP and sends a load of gratuitous ARPs to the network. You can achieve this with VRRP, Keepalived, CARP etc. The Loadbalancer.org appliances use HA-Linux & HAProxy.

How does loadbalancers like haproxy and keepalived implement failover/high-availability for themselves, in case the machine they are running dies?

2 Answers2