When shutting down the server, BalancerStatsRegistry can deadlock. The shutdown thread waits for all threads to terminate here, while holding a mutex taken at this point. Meawhile, one of the threads it is waiting for can get stuck here waiting for the same mutex. In principle, this deadlock could also happen when stepping down It can't happen during stepdown, because _initializeAsync acquires the global lock here, which serializes with stepdown, the same is not true when shutting down.