Replication - High Performance MySQL - page 516

semisynchronous replication could be a lower-cost way to raise a system's data dura-
bility than some of the alternatives.
In addition to semisynchronous replication, MySQL 5.5 also sports replication heart-
beats, which help replicas stay in touch with the master and avoid silent disconnections.
If there's a dropped network connection, the replica will notice the lack of a heartbeat.
There's an improved ability to deal with differing data types between master and replica
when row-based replication is used, and there are several options to configure how
replication metadata files are actually synced to disk and how relay logs are treated after
a crash, reducing some of the opportunities for problems after a replica crashes and
recovers.
That said, we haven't yet seen wide production deployment of any of MySQL 5.5's
improvements to replication, so there is certainly more to learn.
Aside from the above, here's a quick rundown of improvements in the works, either in
MySQL or in third-party branches such as Percona Server and MariaDB:
• Oracle has many improvements in MySQL 5.6 lab builds and development mile-
stone releases:
—Transactional replication state—no more metadata files to get out of sync on
a crash. (Percona Server and MariaDB have had this for a while in a different
form.)
—Binary log event checksums to help detect corrupted events in a relay log.
—Time-delayed replication to replace Percona Toolkit's pt-slave-delay tool.
—Row-based binary log events can contain the original SQL executed on the
master.
—Multi-threaded replication apply (parallelized replication).
• MySQL 5.6, Percona Server, Facebook's patches, and MariaDB have three different
fixes for the group commit problems introduced in MySQL 5.0.
Other Replication Technologies
Built-in replication isn't the only way to replicate your data from one server to another,
although it probably is the best for most purposes. (In contrast to PostgreSQL, MySQL
doesn't have a wide variety of add-on replication options, probably because built-in
replication was added early in the product's life.)
We've brushed elbows with a few of the add-on technologies for MySQL replication,
such as Oracle GoldenGate, but we're really not familiar enough with most of them to
write much about them. There are two that we want to mention, however. The first is
Percona XtraDB Cluster's synchronous replication, which we'll discuss in Chapter 12
because it fits better into a chapter on high availability. The second is Continuent's
Tungsten Replicator ( http://code.google.com/p/tungsten-replicator/ ) .