> "We've hit an error an error that the manual says nothing about, but our DB expert saw this once, 7 years ago, and remembers how to recover" Yup, this rings

> "We've hit an error an error that the manual says nothing about, but our DB expert saw this once, 7 years ago, and remembers how to recover" Yup, this rings true. Once worked on a large Cassandra deployment – holding people's bank accounts, so, ya know, somewhat critical – and we eventually had to get one of the authors of Cassandra onto our on-call team. Even he was stumped by some of the bugs and failure modes. (And, incidentally, the word 'anticompaction' is still enough to trigger my PTSD.) The notion that the documentation is an accurate and exhaustive explanation of every single thing about a product is a junior engineer mistake that no one with rough ops experience would say. Nor, for that matter, anyone who's written a database that's in use (as I unfortunately have), or really any other piece of software. And the same goes for the notion that "everyone should go with MySQL or Postgres for everything because they are Proper Databases that are scalable and won't go down". If I were being charitable, I'd say it's a 'no one got fired for buying IBM' argument; if I were being honest, I'd say it puts me in mind of 'MongoDB is webscale' (https://www.youtube.com/watch?v=b2F-DItXtZs https://www.youtube.com/watch?v=b2F-DItXtZs). I don't know why bad engineers are so terrified of thinking for themselves specifically – well, generally, but especially – when it comes to databases. They aren't magical contraptions. They are just portable filesystem wrappers.