Best practices for embedding migrations

[maxcountryman]

Hi folks,

I'm looking for some guidance on the current best practice for shipping migrations from a library that's meant to work alongside other projects already using SQLx.

Currently the best I've come up with is to provide a Migrator as a static that users can then combine with their own migrations. Something like:

let combined_migrations: Vec<_> = APP_MIGRATOR
    .iter()
    .chain(underway::MIGRATOR.iter())
    .cloned()
    .collect();

let combined_migrator = Migrator {
    migrations: combined_migrations.into(),  // semver-exempt (!)
     ..Migrator::DEFAULT
};

While I think this technically works there's a couple of downsides:

1. It's unfortunate to combine app migrations with third-party migrations,
2. Constructing a Migrator like this requires using semver-exempt fields for now.

If there's a blessed path I'm overlooking I'd appreciate someone pointing it out! 🙂

[abonander]

A goal of #3383 is to allow you to change the table that tracks migrations for an app, making it a lot easier to have multi-tenant databases.

You would want your app to have a dedicated schema that it owns and put the migrations table in there. Currently, the schema would need to exist before the migrations run but I could add a migrate.create-schemas key for that.

Then you wouldn't need to merge the sets, the user would just run your migrations and then their own or vice versa.

I plan to land that feature in 0.9.0 which I hope to get out in the next month or so.

[maxcountryman]

You would want your app to have a dedicated schema that it owns and put the migrations table in there.

Underway already defines its own schema, so it's really just a matter of having some facility for asking SQLx to use that schema for these migrations.

I plan to land that feature in 0.9.0 which I hope to get out in the next month or so.

I'm wondering what to do in the meantime: for my own projects, I'm using the technique above. But when 0.9.0 lands I'm not exactly sure how to unwind that and similarly if other folks want to use Underway before then they'd be in the same boat.

[maxcountryman]

@abonander is there a reason why setting search path wouldn't work as a workaround for the time being? For example, I've done something like this and it seems like it's doing the right. (Note that the search path also must be set in the migration files.)

[abonander]

If it works, it works. But I hesitate to recommend modifying search_path because it can be very hard to debug issues with it.