Steps:
1) Use rsync -avz to copy the data folder to the new slave. This should take a long time and it will likely complain files change during read, which is fine.
2) Repeat step 1. This should take considerably less time; basically only transfers the data that has changed during the initial copy.
3) Shut down master mysqld and repeat step 1 again.
4) Start master and slave mysqld.

On small dbs step 2 can be skipped since very little is expected to have changed during initial copy, but on larger dbs it is beneficial to do two live synchronizations.

This is an excellent article and real food for thought. Your workload simulation idea using a sniffer looks very tractable for a couple of reasons. First, the basic problem with read/write splitting is making an up-front decision when it’s safe to dispatch synchronously to a slave. Here we don’t care about the result of the query or even very greatly when it actually runs. Second, it seems it would not be necessary to run all queries or even a lot of them: a representative distribution should be sufficient to load popular pages into cache.

If we could get a subset of sniffed packets delivered as consistent tcp/ip sessions I think we could adapt our Myosotis proxy to perform this kind of cache heating operation. It could be a simple add-on that has little or no impact on underlying HA mechanisms (i.e., does not make them more complex).

Cheers, Robert

p.s, thanks for the link

You should check out MMM.

A very reliable daemon for running a highly available master-master cluster.

Oren

I know of only three ways to get a consistent backup of myisam suitable for a replication seed. Maybe one of the guys here has a fourth (in which case I’d be curious myself), but here’s the three I know of:

1) Stop the database server, copy the data directory (this is an outage)
2) use mysqldump with the lock all tables option (this is an outage)
3) if you’re using LVM, FLUSH TABLES WITH READLOCK … snap … RELEASE LOCK (this is a very brief outage)

If you’re using innodb, you can just do a dump with –single-transaction in the flags and all should be well.

We run a site with a large number of myisam tables that we’re seriously considering moving to innodb; myisam has a lot of issues around the whole concept of 100% availability.

thanks

However, I don’t see any connection between m-m replication and nested sets, so maybe you have several things you need to think about. It sounds like a complex topic.