can confirm that connections do not die while taking the coordinator down during update (not long time, but still)