Description
I've found that in this case we'll have both ip and ip_start and ip is latest (because as part of joining cluster somebody trying to assign bad name to it). But when cluster is restarted ip_start takes precedence node starts under old name. Which leads to auto-ejection.