We recently had an incident where we had a Redis [http://redis.io/] instance blocked on writing to disk, hung inside the kernel. Through a variety of other circumstances this meant the only up to date copy of business critical data was in memory on a single machine, multiple independent