Continuent Tungsten Stack Targets MySQL Database Replication

Continuent wants to plug what it sees as a hole in the market when it comes to MySQL replication.

Continuent has a long history with MySQL, going back to when Continuent was called Emic Networks and was focused on providing high-availability clustering for the open-source database. Now, Continuent has expanded its open-source stack to include the Tungsten Replicator, technology Continuent CEO Eero Teerikorpi said will help address long-standing challenges.

“MySQL’s own replication [technology] is very good and is one of the key reason why MySQL adoption is as high as it is, especially in Web-based applications,” the CEO said. “But because of this popularity, [its] weaknesses have also been very visible and have created major headaches for the companies running, for example, one master [database] and tens or even hundreds of slaves.”

When master databases fail in this type of setup, it creates a lot of manual work and typically fairly long outages in service, he continued. Continuent's Tungsten stack helps to run these larger sites much more smoothly by providing automatic master failover and online maintenance and upgrades, he contended.

With Tungsten Replicator, the company has targeted a number of challenges around replication laid out by Monty Widenius, an original developer of MySQL, at the 2008 MySQL Conference & Expo in Santa Clara, Calif. Specifically, the company has put a bull's eye on MySQL statement replication, proper handling of master failover in presence of multiple slaves for high availability and checksums on replication events. In addition, Continuent also added a table consistency check mechanism.

“We solve the failover problem using global sequence numbers that allow any slave to connect to any other slave and commence replicating data,” said Continuent CTO Robert Hodges. “More specifically, we add a sequence number to every request we pick up from the logs. The requests are stored in a structure known as a transaction history log [THL], whose contents are copied to slaves and applied there.”

The master and each slave agree on the sequence numbers, which allows slaves to be promoted to master in the event of a failover based on who has the highest sequence number. This, Hodges explained, is a simple way to pick the most up-to-date slave.

Besides the Tungsten Replicator, the Tungsten stack also includes Sequoia, the synchronous multi-master replication project, as well as Tungsten Manager for cluster-aware service management over LAN or WAN, Tungsten Connector for native wire protocol proxying, Bristlecone for scale-out performance testing and Hedera for group communications.

In addition, the company plans to announce support for Oracle Sept. 16, and has support for PostgreSQL on its road map for late 2008 or early 2009 as well.

“There will be a follow-up release on Oracle replication within two weeks, just before Oracle OpenWorld,” Teerikorpi said. “PostgreSQL is lacking some key features for us to provide our solution to that market, but we are working with the PostgreSQL community to get this addressed."