Generally when redis cluster expand/shrink, we would depend on a MOVED error to update the node table when using a client like Jedis.

However, if we use MULTI/EXEC or lua script, the MOVED error would be covered by transaction error message like -ERR Transaction discarded because of previous errors., which means MOVED would never be received.

Therefore, if there is a redis user who only use MULTI/EXEC data, after cluster expand/shrink, everything would stop until someone update the node table manually or restart the client.

Shall we treat this as a redis problem? or just leave it to clients.