SWITCH MASTER TO

We could use multiple source replication on master to create a domain pointing on each replica that fetch a single blackhole query queuing table, from each replica we use that domain to send any write query in a transaction to the master instead of error and forcing the slave to be readonly . The master execute the query and send the result back to the slave via a replicated temporary table and send an acknowledge via inserting into a replicated acknowledge system table, the slave put the query on hold waiting for the acknowledge event could be via waiting for the ack containing GTID to wait for, on such event forward the temporary table to the client. Having this in place we allow load balancing to all nodes and can implement a SWITCH MASTER command. That solution is losing snapshot isolation unless we implement a way to propagate READ lock as well on the master but offer easy clustering by enabling all nodes to read in COMMITTED READ without the requirement of read write splitting