I like method a) better, because then even if the slave lags a little bit, you can know a particular table is caught up. And if necessary you can wait and check back.

It would be great to see something like this on the Forge though, it’s something that does get asked for quite often.

I’m looking at getting this kind of procedure in to the “MySQL Netork Monitoring and Advisory Service” as well, so this can be done in the background and alerted on, hopefully in the coming versions - however something for the Community is certainly needed as well!

Great article. I had one thing that I wanted to note, to give you a little tip. You say:

“To take checksums of your master and slave, you’ll obviously have to halt the master and let the slave catch up to it. The safest way I know is to stop the master, let all slaves catch up and then stop them, re-initialize one slave directly from a filesystem snapshot of the master, start the master again, and keep all the slaves stopped — thus ensuring that you have all slaves stopped at a single point in time relative to the master. Then you can compare each of the slaves to the brand-new snapshot from the master.”

Actually, this all seems a little complex, you can in fact do something like:

/* On slave(s) */

STOP SLAVE;

/* On master */

FLUSH TABLES WITH READ LOCK;
SHOW MASTER STATUS;
/* CHECKSUM or SELECT COUNT(*) on table(s) */
UNLOCK TABLES;

/* On slave(s) */

START SLAVE UNTIL MASTER_LOG_FILE = '‘, MASTER_LOG_POS = ;
/* Wait for slave to run to position */
/* CHECKSUM or SELECT COUNT(*) on table(s) */
START SLAVE;

This allows you to not shutdown the master (which means you have to restart, warm caches again, etc.), and the same on the slaves as well - whilst still providing a 100% consistency check.

Cheers

Mark