Two or more health check instances started at around the same time would
read the partition assignment state, selfishly add only themselves, and
the last to write would be the result. The loser health check instance
would get into an endless loop of:
Unable to find broker's partition in topic \"broker-replication-check\" in metadata
Two or more health check instances started at around the same time would read the partition assignment state, selfishly add only themselves, and the last to write would be the result. The loser health check instance would get into an endless loop of: