[throwaway20371]

> Sadly, when the 1% of the time hits this also means that nobody on the team has built up the DBA skills required for the big hairy problem. I see this pattern repeated all the time when consulting for startups that just had their first real traction and now the magic database box is no longer working as before.

I call this the Kubernetes effect. 99% of the time, people think it's easy. Then the 1% hits and they have to completely rebuild it from the ground up. For us it was cluster certificate expiration and etcd corruption, and some other problems we couldn't ever pin down.

The other effect (which I'm sure has a name) is when they depend on it for a long time, but then a problem starts to occur, and because they can't figure out a way to solve it, they move to a completely different tech ("We had problems with updating our shards so we moved to a different database") and encounter a whole new problem and repeat the cycle.