Ask HN: How are you doing redis failover?

I've seen several large entities using redis in production, such as GitHub.

Due to Redis Cluster being at least 6 months from an RC, how are you guys doing failover and upgrades without downtime?

There seems to be two options that I've seen, and they are:

1. Juggle around virtual IPs upon a redis-server dying, so that the clients can still use the same IP, without knowing a secondary IP in advance.

2. Modify each redis client implementation to switch over to the secondary IP upon connection refuse.

What are you guys actually doing, or what are some suggestions for making Redis more fault tolerant?