Tokutek sponsored the Percona Live MySQL / New York 2012 Conference which took place this past Monday and Tuesday.  I spent much of the time at our booth discussing TokuDB with conference attendees but also managed to attend the following presentations:

• Understanding Performance through Measurement, Benchmarking and Profiling by Rene Cannao of PalominoDB.  Rene presented 3 hours of content covering his methodologies, toolbox, and tips & tricks to help identify performance bottlenecks. Rene has obviously been optimizing systems for some time as I’ve never seen such a comprehensive list of tools and techniques.  This …

Overview
MySQL replication does not always scale up well. A common reason for
that is the single-threaded nature of the slave server applier.
Indeed the performance race between the master and the slave
has been unfair. Contrary to the master, which executes transactions
concurrently by multiple threads, the standard MySQL slave is limited
to install changes sequentially by the only thread.
However, if logical partitioning of data per database
takes place, the 5.6 server's Multi-Threaded Slave (MTS) framework may
be found helpful. It allows to install changes done to different
databases in parallel.
For instance the simplest use case would be when the master server has
just two databases and transactions against the master server update
only one of them at a time. Those transactions when replicated will
be executed by two separate slave worker …

There's definitely more and more buzz on and around cloud computing and different solutions to support it nowadays!

This was evident while visiting LinuxCon/CloudOpen in San Diego a few weeks ago, where I spoke about high availability solutions for MySQL. Most cloud sessions were highly attended, and I could sense an excitement in the air with regards to the cloud. It was also evident, however, that there is only a small percentage of people today truly taking advantage of the cloud. It seems that despite all the excitement, there is a large portion of potential users who don't know how to use the cloud, or what the benefits and potential drawbacks are.

read more

This second day at the Oracle Open World is hotter than the previous one! Today, we focused on Oracle Linux. Oracle Linux roadmap, the Unbreakable Enterprise Kernel (UEK) release 3 is coming in summer 2013. Oracle's strategy consists of developing the new enterprise-class kernel on state of the art hardware, which guarantees to customer to get the most of their hardware (Like Suse for SLES 11).

The Oracle Linux development team is working on many topics. But before, we will review what changed with Oracle Linux 6.x:

• Oracle-rdbms-server-11gR2-preinstall RPM is a replacement for "oracle-validated". This RPM aids in the installation of the Oracle Database.
• The package manager (yum) has been improved with several features like patching your kernel with only CVE updates for fixing security leaks
• BTRFS as root file system

In combination with the right yum command, you can install kernel …

As Todd Farmer points out in Understanding mysql_config_editors security aspects, the new .mylogin.cnf file generated by mysql_config_editor does not securely store the password used to login to the database. It just obfuscates it.

The format of the file is as follows (as of MySQL 5.6.7-RC):

• 4 Bytes Zero (Version Information)
• 20 Bytes Key Generation Matter
• Repeated:
  • 4 Bytes Length information
  • Length bytes crypted matter. The crypt is done using the AES ENCRYPT function, which in itself is insecure: It is an aes-128-ecb with a NULL IV.
  …

At MySQL Connect last weekend the MySQL 5.6 was released as a Release Candidate. It contains a large set of improvements over MySQL 5.5. In this blog I will focus on what has changed in the area of scalability.

There are four main areas of improvement of scalability in the MySQL 5.6:
1) Splitting InnoDB Kernel mutex
2) The G5 discovery
3) Splitting LOCK_open mutex
4) Binlog group commit

In InnoDB as in so many other products there was an infamous kernel mutex. This mutex was a sort of catch all mutex used to protect the system from concurrent use of various data structures. For an implementer it was easy enough to use the kernel mutex as a way to achieve safe concurrency. The problem of course is that it hurts our scalability. The kernel mutex was a major bottleneck in many benchmarks, and even more important it also made it very difficult to resolve other scalability issues.

…

So we tested the 5.6.7-RC. And ran into a strange problem:

Because of a test, a preexisting configuration with GTID enabled existed, and suddenly we did not have properly initialized grants in mysql.* created for a new installation. Turns out: GTID and non-transactional tables are no friends, and that is even documented.

When using GTIDs, updates to tables using nontransactional storage engines such as MyISAM are not supported. This is because updates to such tables mixed with updates to tables that use a transactional storage engine such as InnoDB can result in multiple GTIDs being assigned to the same transaction.

Also, this is supposed to work with GRANT and REVOKE, but not with INSERT and DELETE. Now guess what mysql-install-db and friends are using?

server:~ # less …

Global Transaction Identifiers – why, what, and how Next post: Advanced use of Global Transaction Identifiers
This post was kindly translated to Japanese by Ryusuke Kajiyama.

In MySQL 5.6 we introduced a new replication feature called Global Transaction Identifiers, or GTIDs. While there are many use cases, our primary motivation for introducing GTIDs is that it allows for seamless failover. By this, we mean promoting one of the slaves to be come a master, if the master crashes, with minimal manual intervention and service disruption.

This is the first in a series of several blog posts. We will go through several use cases and show how easy it is to do a failover. We explain …

I’ve just finished presenting my work on adaptive fault detection at Surge and Percona Live NYC. My slides are available on Slideshare, and embedded below.

Adaptive Fault Detection from xaprb

Further Reading:

• Products that scale linearly to hundreds of servers
• …

Every so often you hear an idea that makes you wonder why it had not been thought of it before. At the recent Northeast PHP Conference, Sheeri Cabral and I were at the speakers dinner when we heard how the Boston PHP group was training novices. Many folks were joining the Boston PHP book but were just starting out and needed some formalized training. So the brilliant Gene Babon came up with the idea of PHP Peculate. The idea is simple: Everyone buys the same book, reads one chapter a week, and posts answers to exercises online. They had great success with PHP and some other subjects. Sheeri has adapted the idea to MySQL and has just announced MySQL Marinate.

Membership in the …