Steps

1. Start postgresql
2. Test a few times

cargo test bug_test

It ramdomly fails with the following error PoolTimedOut:

running 2 tests
INIT ModelManager
test tests::test2 ... ok
thread 'tests::test1' panicked at src/lib.rs:38:33:
called `Result::unwrap()` on an `Err` value: PoolTimedOut
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
test tests::test1 ... FAILED

Reason

This is because every tokio::test runs in a separate runtime, but share the same OnceCell initialized by one of them.

If the runtime which initialized the OnceCell exits and the IO resources just stop working, the left runtimes can nerver get the OnceCell.

Fix

Fix in this way:

use lazy_static::lazy_static;

pub fn test_rt() -> &'static tokio::runtime::Runtime {
    lazy_static! {
        static ref RT: tokio::runtime::Runtime = {
            let rt = tokio::runtime::Builder::new_current_thread()
                .enable_all()
                .build()
                .unwrap();
            rt.block_on(init_test());
            rt
        };
    }
    &RT
}

/// All the test about database should be run in this function.
/// Because we need to share the same database connection,
/// which cannot keep alive between runtimes.
/// In other words, if the database connection is closed
/// due to the end of the runtime which initialized it, the next runtimes will fail.
pub fn run_test<F: std::future::Future>(f: F) -> F::Output {
    test_rt().block_on(f)
}
#[test]
fn test() {
    // run all test in run_test()
    run_test(async {
        let _mm = init_test().await;
    })
}

This can ensure the all the tests share the same runtime.

You can try:

cargo test fixed_test