Hi, just wondering how does replicas work for realtime tables in terms of choosing which replicas to query? From what I can see, it appears that the broker randomly chooses which replica to use when querying.

yes, it randomly selects the replica. As of today, when a replica goes and returns, broker does not wait to recover and catch up before querying. We thought of adding this but we did not because in practice, the catch up is very fast and hardly noticeable.. we have seen speed of 100k event/sec during catch up..

thanks!