Thu 04 Aug 2011
Transactions and Locks in SQL
Category : Technology/TransactionProcessing.txt
I'm done comparing the SQL calls I need to do across MySQL and PostgreSQL in order to support transaction processing (i.e., Locks, Begin, Commit or Rollback) in a multi-user version of Luca that I'm working on.
Fortunately, the SQL calls for transaction processing are consistent across MySQL and PostgreSQL. They're the same - BEGIN, COMMIT or ROLLBACK - for each platform.
While PostgreSQL has always supported transactions for as long as I can remember, MySQL with MyISAM tables do not. We'll need Innodb, which is the default setting for newly created tables in MySQL 5.5.
Once we can get our Lion installer for MySQL 5.5 done, Hai Hwee has a functionality built into her installer such that it'll migrate forward all the data in your current databases to the new version of MySQL that you are installing. I'm hoping that when that kicks in, all the data in the MySQL tables will then automatically support transaction processing, and then we'll be able to do BEGIN, COMMIT or ROLLBACK from applications like Luca, knowing that we'll be accessing databases that support it.
We also need advisory locks - putting a lock on database updates among consenting apps (I almost said adults). This is a simple form of record locking, to prevent the database from being written to the same place at the same time by different people, using Luca on different client machines. I don't think Luca needs the more sophisticated form of record-locking at the moment, so advisory locks will do for now.
On MySQL it's very easy. SELECT GET_LOCK('Lock1', 10); To test for the lock, it is SELECT IS_FREE_LOCK('lock1'); To release the lock, just do SELECT RELEASE_LOCK('lock1'); I just tested it, accessing MySQL from two client instances. It works great.
On Postgres, the equivalents are SELECT pg_advisory_lock(5), SELECT pg_try_advisory_lock(5), and SELECT pg_advisory_unlock(5).
So two differences - one is the name. The other is that MySQL uses a string for the semaphore while Postgres uses a big integer.
I can hide these differences in my database frameworks so that an app like Luca calls the same code, no matter which type of database it is accessing. So we're good to go.