if we shut down on a fatal error (like unique index violation), that can be good as we would just go infinitely stale if we just halted replication. however there is an issue though which is we can no longer then vote for a primary. so if other nodes halt but primary is healthy it might step down. staying up isn't a good option either but in the future this is something to solve.

another way to phrase this is that if a secondary is down, ideally it can still do its duties as an arbiter.