Re: Federated as MultiMaster Replication Replacement

[Vinayak Javaly]

I'm thinking of implementing something similar. Here's the server flow diagram I've come up with:

Master1 Merge1
\ / \
replication / \
\ / \
[Slave1A & Slave1B] or [Slave2A & Slave2B] or [...]
/ \ /
replicaton \ /
/ \ /
Master2 Merge2

All writes will go thru the Master servers. The data being inserted will not be updated / deleted. The data needs to be online for long-term future access.

Initially, both Masters 1 & 2 are replicating to Slaves 1A & 1B. When this set of slaves reach a certain limit (we have not yet determined the cutoff limit), we'll do the following:
- stop replication
- reconfigure Masters 1 & 2 to replicate to new Slaves 2a & 2B
- truncate tables on Masters 1& 2
- restart replication from Masters 1 & 2 to new Slaves 2a & 2b

Another set of MySQL servers, Merge 1 & 2, will perform all read operations. Each Merge server will Merge the corresponding tables from Slave1A and Slave2A and ... (same thing for Slave1B and Slave 2B and ...). All reads will go thru the Merge servers.

If the MyISAM Merge allows me to merge tables across different servers, I might not have to use the Federated approach.

Please comment.

Thanks.