Enable loading migrations from multiple directories.

[pedestrianlove]

I'm new to Rust, so I may have missed some obvious hints. If that's the case, please pardon my ignorance. 🙏

---

I'm looking for a way to implement modular design in my project using SeaORM.

e.g. Each module folder may correspond to a distinct set of features in the system, and they have their migrations, models, controllers, and routes.

However, I had a difficult time finding a way to put migrations into separate module folders. (at least I cannot find it in docs)

Is it possible to achieve this kind of behavior in SeaORM?

Thanks!

[pedestrianlove]

After looking into the source of the scaffolded crate, I just found that you can customize the behavior of sea-orm-cli migrate by modifying the content of main.rs. In theory, you should be able to stack multiple customized Migrator instances so that they're all run within the same main.rs, triggered by the same sea-orm-cli command.

Therefore, I'm changing the category of this discussion from Ideas to Q&A for now to avoid possible confusion. (I'm going to try if this approach works or not)

(btw, is this a recommended practice to do so, or does SeaORM have a requirement or limitation that depends on that cli line of code to be run only once?)

[pedestrianlove]

tl;dr: The "Multiple migrator" approach either doesn't work or requires additional modification.

The following is my attempt making use of sea_orm_cli crate to migrate programmatically:

---

migrations/src/main.rs

use sea_orm_migration::prelude::*;

#[async_std::main]
async fn main() {
    cli::run_cli(migration::KernelMigrator).await;
    cli::run_cli(migration::NotesMigrator).await;
}

migrations/src/lib.rs

#![allow(elided_lifetimes_in_paths)]
#![allow(clippy::wildcard_imports)]
pub use sea_orm_migration::prelude::*;

mod m20220101_000001_users;
mod m20231103_114510_notes;

pub struct KernelMigrator;
pub struct NotesMigrator;


#[async_trait::async_trait]
impl MigratorTrait for KernelMigrator {
    fn migrations() -> Vec<Box<dyn MigrationTrait>> {
        vec![
            Box::new(m20220101_000001_users::Migration),
        ]
    }
}

#[async_trait::async_trait]
impl MigratorTrait for NotesMigrator {
    fn migrations() -> Vec<Box<dyn MigrationTrait>> {
        vec![
            Box::new(m20231103_114510_notes::Migration),
        ]
    }
}

Result:

➜  tester $ DATABASE_URL="mysql://test:test@localhost:3306/tester" sea-orm-cli migrate up                                                                                                                                                                                                                                                                                                                                                                                         (base) 
Running `cargo run --manifest-path ./migration/Cargo.toml -- up -u mysql://test:test@localhost:3306/tester`
    Finished dev [unoptimized + debuginfo] target(s) in 0.15s
     Running `target/debug/migration up -u 'mysql://test:test@localhost:3306/tester'`
Applying all pending migrations
Applying migration 'm20220101_000001_users'
Migration 'm20220101_000001_users' has been applied
thread 'main' panicked at /home/jsl/.cargo/registry/src/index.crates.io-6f17d22bba15001f/tracing-subscriber-0.3.18/src/util.rs:91:14:
failed to set global default subscriber: SetGlobalDefaultError("a global default trace dispatcher has already been set")
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
Fail to run migration

[lucas-labs]

Did you find a workaround? I'm facing the same issue trying to call cli::run_cli multiple times with different migrators. In my case, I have several libraries and each has their own migrators creating the tables they need. Then I have a "main" project which generates the final binary. My idea was to call run_cli for each migrator in the main, but... nope.

[pedestrianlove]

Not yet. And our team just decided to move on to golang instead, so we're unlikely to actively explore this issue in near future.

[lucas-labs]

Btw this is what I'm doing for now, unless there's a better way (create a "Main" migrator, that combines the migrations of all the other migrators:

use sea_orm_migration::prelude::*;

/// main migrator, which combines all the other migrations
pub struct MainMigrator;

#[async_trait::async_trait]
impl MigratorTrait for MainMigrator {
    fn migrations() -> Vec<Box<dyn MigrationTrait>> {
        let mut migrations = vec![];

        // get migrations from the specific migrators
        migrations.extend(lib_foo::migration::Migrator::migrations());
        migrations.extend(lib_bar::migration::Migrator::migrations());
        .... add as many migrators as we want ...
  
        // return the combined migrations vector
        migrations
    }
}

#[async_std::main]
async fn main() {
    // execute em!
    cli::run_cli(MainMigrator).await;
}