Incremental MigrationWith this approach, every time you make a change to your schema, you create two scripts: one to apply it and one to revert it. The scripts are numbered or timestamped in the order they need to be applied. So, deploying a new version of the database is simply a matter of applying these scripts in the right order. This technique is quite popular and widely used (Ruby Migrations is probably the most notable utilizer of it). This approach is nice because it’s simple. Delta scripts are easy to create and apply, database schemas are easy to version, changes are easy to roll back. The primary complication is that you need to track what version of the database you’re deploying to. Otherwise, you run the danger of not applying all of the changes required.
Schema SynchronizationWith this approach, your entire database schema is first checked into source control, with each artifact being represented by a script. It’s important to note that these scripts usually contain only CREATE operations (and not ALTER). To make a change to the schema, you simply modify the corresponding script. The really interesting part is how deployments happen. To deploy, you point at an existing physical database. Its schema is compared against the schema represented by scripts in source control. A delta script is then generated and ran against the target. Visual Studio database projects use this approach to manage schema changes and deployments, as does Red Gate’s SQL Source Control. The cool thing about schema synchronization is that it works with any version of the database. If you point it at a empty local database, it will rebuild it from scratch. If you point at production, it’ll apply only the changes which are actually needed.
Choosing The Right SolutionI find that the schema sync option works a bit better with DBAs because the deployment process produces a single delta script, which they can examine and get comfortable with before applying changes to production. On the other hand, devs prefer the incremental migration because it’s more widely used, it’s more transparent (i.e. there is no magic in figuring out what the delta is), and it’s more dev friendly. There are even frameworks which allow you to code your migrations as objects (a good example in the .NET space is Fluent Migrator, which lets you code db changes using a DSL in C#). Whatever option you end up choosing, you should make sure that it works for everyone who will be making db schema changes: devs and DBAs (and anyone else who needs to touch the database). There is danger in picking a solution that’s really dev-friendly, but alien to DBAs. Remember: if the solution you picked is not accepted by the entire team, it would defeat the purpose of having your db schema under source control in the first place.
You may also like:
Did you love / hate / were unmoved by this post?
Then show your support / disgust / indifference by following me on Twitter!