> I see this line of thought that somehow self-hosted "clouds"/clusters look after themselves, but thats usually not the case. They don't really look after the

> I see this line of thought that somehow self-hosted "clouds"/clusters look after themselves, but thats usually not the case. They don't really look after themselves, but do handle failures on the highest possible level. Which makes it unnecessary to keep each OS in check. What's critical for a single NAS box is critical for a distributed storage only if all boxes of some replica have the same problem at the exact same time, otherwise you just reboot and move on and it doesn't matter if it happens again, it doesn't cause any downtime. I actually speak from my own experience. I run and maintain a distributed key-value storage for many years. Although I designed and implemented it myself (and redesigned a bunch of times), I don't see how experience with those other distributed storages, like Swift, would be any different.