A quick look at MySQL 5.5 GA

Congratulations to Oracle and the MySQL team for getting 5.5 out the door!

I first have to acknowledge that I have not been able to follow the MySQL 5.5 development as closely as I would have wanted during the last 2 years as most of the planning of MySQL 6.0, 5.4 and 5.5 has happened behind closed doors, without insight for the community. The commits have been open, until recently, but it's not easy to follow what is happening just based on the commits. I am sure that I am missing below some of the important features in 5.5 and forgetting to acknowledge some of the people that have done great work on 5.5.

That said, the 5.5 release contains some things that a lot of heavy users of MySQL have waited a long time for. Here follows some of the main things that I know about.

InnoDB improvements:

• The new version 1.1 of the already improved InnoDB plugin is now the default (and only) InnoDB (in 5.1 the InnoDB plugin version 1.0.x was an optional engine).
• Multiple buffer pools and rollback segments, splitting of log_sys and flush_list mutex, and other improvements that together give a major contribution to the improved scalability and performance of 5.5.
• Thanks to the InnoDB team for continuing to do impressive work!
  

Adding DDL locking.

• In earlier MySQL versions LOCK_open, which protected table opening (among other things), was one of the main things that stopped MySQL from scaling on many CPU's, especially when many tables were used. By introducing proper DDL locking this issue is now mostly fixed.
• A big thanks to Konstantin Osipov and Dimitry Lenev for the hard work they have done in getting this to work.
  

Reducing scalability bottlenecks in the core server, such as LOCK_open, LOCK_alarm, LOCK_thread_count, Query cache and binlog mutexes and so on.

• Thanks to Mikael Ronström for driving this!

A big speed improvement for multi-cpu on windows.

• This was actually the result of a minor bug fix for how mutex and condition variables were used on Windows. By using native Vista primitives instead of homegrown ones, a big speedup was achieved.
  
• Thanks to Vladislav Vaintroub for getting this fixed!

Optional Semi-synchronous replication (the server will wait for at least one slave to catch up before continuing) and replication heart beat are good improvements for those that use replication.

• Thanks to Google and Mark Callaghan's team for providing the initial patches for this.
  

The performance schema gives a lot of more insight in which mutex and IO calls are the bottleneck for your queries.

• Thanks to Marc Alff for driving and developing this.
  

Pluggable authentication

• Done by MIT student R.J.Silk and Sergei Golubchik for MariaDB 5.2. Donated to Oracle by Monty Program Ab.
  

And of course a big thanks to everyone else who has been involved in this and been fixing bugs in 5.1 and later.

Some good links for more information about the improvements in 5.5:
- New In MySQL 5.5 Performance_scale Unleashed Presentation
- Introduction to MySQL 5.5
- MySQL 5.5: What's New in Replication

It's now almost exactly 2 years since the MySQL 5.1 GA release, so what's the verdict of 5.5?

The two things that impress me are the work on the InnoDB storage engine and the DLL locking. Most of the other new features are minor (technically) adjustments even if the results, like with the windows speed improvement, can be impressive. As a technical person I have a hard time getting impressed by something that could be done in a few hours and could easily have been done in 5.1.

What is worrying me the most is the things that are not in 5.5. Some if the things that were supposed to be in the next MySQL release are:

Pool of threads (instead of one thread per connection like it is in MySQL 5.1).

• As far as I know, this feature has been moved to appear in 5.6 enterprise and will not be part of the community server.

An open source, free backup utility and backup API that would work for all storage engines. This has probably been the most awaited feature of MySQL ever!

• This project was discontinued by Oracle the same day the feature was deemed ready to be pushed into 5.5.

In 5.5 some internal subsystems, like safemalloc, a portable memory checker, were removed.

Another worry I have is how long Oracle will be able to continue development of MySQL while important core developers continue to leave. Of the above, Konstantin Osipov and Vladislav Vaintroub have already left Oracle and there are not many old-timers left.

When it comes to MariaDB we are continuing to work making MariaDB 5.3 stable. MariaDB 5.1, 5.2 and 5.3 use the InnoDB-plugin compatible XtraDB storage engine, which already has many of the speed improvements of the new InnoDB plugin, but not yet the windows speed improvement.

We have already merged MariaDB 5.3 with MySQL 5.5 and we will release the MariaDB 5.5 tree shortly. It has taken some time as there are a lot of things done in MariaDB 5.3 that conflict with new code in MySQL 5.5 and we also have to add back some of the features removed in MySQL 5.5.

The next big decision is to decide which version we should release next as a stable release; MariaDB 5.3 or MariaDB 5.5.

The problem is that it's very hard to evaluate how stable 5.5 really is as that 5.5 has not been tested by the public enough and some of the bugs in the public bug system, like the critical Bug #33082, can't be accessed. The bug database shows 294 open bugs for MySQL 5.1 and 150 for MySQL 5.5. I haven't however had time to evaluate all public bugs, so I don't know how severe they really are. A lot depends also how many bugs will be filed during the next few weeks after the GA release.

Note: I corrected the above sentence after getting confirmation that the MySQL team has not yet switched to Oracle's internal bug system. This is scheduled to happen first in the spring. Thanks for helping me get the facts right!

The good news is that many of the bugs that I originally reported for the 5.1 GA release have been fixed. However there are still things that could have been better:

• regression in parser compared to 5.1.x. This was reported just after 5.5 was released, but something like this should have been noticed during testing.
• regression compared to 5.1.x.
• cannot select from innodb_trx when trx_query contains blobs that aren't strings
• Object names not accessible via case sensitive reference on lctn=1 systems
• Change in execution plan for count_distinct_group_on_key causes 400 % performance drop
  

And there are still bugs that were supposed to be fixed before the 5.1 GA which are still not fixed:

• cost calculations (in optimizer) are off
• Slowdown (related to logging) in 5.1.21 vs. 5.1.20
• Stored routines do not detect changes in meta-data.
• crash when federated table loses connection during insert ... select
• comparisons with Information schema tables don't honor collation
• Partitioning performance drops drastically with hundreds of partitions
• Event set to DISABLED and ON COMPLETION NOT PRESERVE is deleted at server start
  

So our plan is to continue to fix the last issues in MariaDB 5.3 and at the same time keep MariaDB 5.5 up to date. Early next year when it's time to decide which one to release as stable, we will have more information about the stability of MySQL 5.5. Let's all hope for the best!