Bobby Tables, that’s right, Elasticache would be a much better option. At Netflix, we don’t treat any of the Redis instances as something we can just lose, but we do lose them & can recover: Every Redis has a replica that we can promote to master as well as S3 backups. Here’s a couple things to keep in mind on Redis usage:

• Igor: Losing the index will re-notify all builds (there are guards against mass notifications, though).
• Clouddriver: Losing Redis will cause any in-flight atomic operations to fail. Atomic operations typically don’t last longer than 30–60s. Everything else (the cache) will rebuild.
• Orca: All running executions will fail and become orphaned. Execution history will be lost.
• Gate: Users will need to login again.
• Echo: We don’t yet use Redis for Echo (but will when we adopt Pubsub triggers internally). From the sounds of it, losing Redis will cause pubsub events may cause redelivery of events.
• Front50: Never use Redis for Front50. :)

I’m not particularly keen on how much we use Redis. It’s great to develop against, but has some operational challenges around HA.