Yesterday we had our third MySQL User Group meeting in Trondheim with close to 40 participants, a mix of database administrators, application developers, database developers, and even university professors.
We had all gathered to listen to Truls Bergskaug from Basefarm a leading hosting provider for mission critical business applications. Truls Bergskaug is an experienced database administrator who deals with around 500 MySQL instances in production, all MySQL versions starting from 4.1 up to 5.6. Truls presented the hosting environment, typical set up scenarios, issues related to database upgrades, monitoring, and management. As an example Truls outlined Basefarm’s internally made rpm packages, configuration files and database layout. Then we spent some good time on questions and answers and general discussions. A few highlights are:
Q: What is most important to you, new features, better performance, better monitoring, or something else? A: The single most important thing is performance (well… stability goes first, I like my databases to just run and run…)
Q: Do you see performance improvements between 5.1 and 5.6? A: Improvements are seen for complex queries (the optimizer is getting better) and it scales better with many CPU cores.
Q: What is the most desired features in a new release? A: Online. New abilities to do things without stopping the database.
Q: How do you do upgrades? A: Replication. We replicate over to a new instance running a newer version and then do a manual switchover.
“What else would simply your life as a DBA?” One fundamental problem in hosting environment is to know what the customers will do or use. For example, most customers use InnoDB storage engine but then they occasionally use MyISAM. If there existed a feature to lock the MySQL instance to only use InnoDB we could do more powerful assumptions, e.g. to rely on Global Transaction IDs. Furthermore, monitoring index usage is important to us. For example, I would like to have easy access to the number of times a given index has been used. Also, it can be very expensive to monitor buffer pool usage for large buffer pools.
On behalf of the audience I want to thank Truls for sharing his experience with us!