Specify DB migrations in terms of your persistent model
persistent-migration is a Haskell library that lets you describe your DB migrations in terms of
persistent entities, and it handles the details of the
persistent backend for you. Whenever you change your
persistent model, all you need to do is to list a typed description of the change in the list of migrations, and the library has everything else covered.
Currently only the SQL backend is supported, and only PostgreSQL (because that’s what I’ve been using), but it’s perfectly possible to add typeclass instances for other backends and database management systems.
There is a variety of database migration tools out there, including in the Haskell ecosystem and Hackage. However these migration tools work with migration instructions written directly in SQL. This makes sense of course, but it also makes it more difficult to write and read them. If your code has a high-level persistent model, translating it in your head into SQL and figuring out the types and small details is cumbersome.
In the Haskell
persistent library, these details are handled automatically for you. The migration process tries to guess which changes you made, and adapts the database for you. When it doesn’t, or when its guess involves changes it considers unsafe, it gives you the details so that you can manually run the necessary migration SQL queries.
I’m not in the software industry and I don’t write single instance software-as-a-service or centralized web app code, but I’m guessing that this way of migration your database works nicely when you, the developer, run the instance(s) of the web app and you can manually tweak the migrations and DB schema when needed. And it’s a one-time operation for each change made to the DB.
What I have been writing, however, is web app code that is free-as-in-freedom software and is meant to be run by everyone and have an unlimited number of instances out there. For this reason, I can’t rely on any manual tricks and tweaks - the migration process has to be automatic and run without user intervention when someone builds my web app and runs it and later upgrades it to newer versions with changes in the DB schema.
To summarize, the situation is like this:
persistententities, not in terms of SQL, and that’s how I think of them in my mind and I don’t need to remember the SQL details behind them.
So when, for example, I change some entity field’s type from A to B in my model file, I just want to tell my web app’s migration tool “change this field’s type from A to B and use the following function to convert the values” and not worry about the SQL details, and just move on to my next coding task.
persistent knows all the SQL behind my nice elegant model, and I’d like it to handle these details for me. I can figure out the SQL names of my entities and fields, I can figure out the constraints and the SQL types behind my Haskell types, it’s just that I don’t want to have to think about it every single time I change something in my persistent model.
An additional point is that a web app with many instances may support more than one DBMS, and each DBMS has its own SQL queries for schema changes, and in general all kinds of small and big differences from other DBMSs. By describing changes in terms of
persistent models, the migrations are able to cover the various
persistent backends and DBMSs your web app may support.
There is more than one way to provide these requirements. This library doesn’t necessarily do it the best way, and I do want to put more thought into that eventually, but right now here’s how it works.
In your Haskell code, you define a list of DB schema changes and DB transactions. You can use the provided typed ones, or write custom SQL. You then pass this list of changes to a function
[Change] -> IO () which connects to the database and runs the necessary migrations. And you can use this function in the initialization code of your app.
Perhaps the biggest issue with this approach is that you may use some detail from your code in the description of a change, for example some type that you defined, and later, weeks or months later, change that type and break the migrations. For now the solution is to be careful: Whenever you change a type that is used by the persistent model, check in the model and in the list of changes, and update where necessary to avoid changing old migrations.
A better approach for the future of this library may be for the developer to specify the change using the same terms, except it is instantly converted to SQL and stored. And the high-level change description is stored along with it, at least as a comment, for the convenience of the developer and users running the app.