This message was deleted.

killall is about the worst way to do that. It doesn’t let anything shut down cleanly, it just `kill -9`s all the pods. You should drain/cordon the node first, THEN killall if you absolutely must. Although just doing a normal shutdown and letting init send sigterm before killing things would be better.

In my experience a rough shutdown of ceph requires a bit of patience for it to restart and clean itself up.