Used it for fully transient services. If search died, it could be rebuilt from scratch. If chat died, data about active chats would be lost, but no one really c

Used it for fully transient services. If search died, it could be rebuilt from scratch. If chat died, data about active chats would be lost, but no one really cared. The main issue with using it with other services was answering questions like "how do we restore a database backup?" or "what do we do if Kafka blows up, and we lose messages?". It's possible to create a design that won't get into an inconsistent state when bad things happen, but people tend to greatly underestimate how hard it is.