Question 954

One Address, Two Boxes

A single load balancer instance fronts your whole site and is an obvious single point of failure. You need failover in under a second, with clients continuing to use one unchanged address. Which design does this?

A

Let the backends notice the dead balancer and re-register themselves with the survivor through service discovery

B

Two balancers share a virtual IP; the standby claims it on failure and sends a gratuitous ARP so switches relearn the MAC

C

Publish both balancer addresses as A records with a 30 second TTL and let clients fail over through DNS on error

D

Give each balancer its own address and rely on the client's TCP connect retry to reach whichever one survives