Basically same as Cannot add or remove masters when master is dead · Issue #5211 · yugabyte/yugabyte-db · GitHub
IMO, this is a an unintuitive from operations perspective and quite hard to debug when happens.
Compared to something like cockroach where cluster is able to detect node replacements and heal by itself. If -master-adresses changed and no longer include dead master, yugabyte probably should automatically start remote bootstrap for new master from the two old ones.