"spawn_local" panic when using Actors within Actix in a sqlx::test runtime

[esmevane]

Hey there!

I'm experiencing a strange issue when running actors alongside actix-web within a #[sqlx::test] harness. Here's a reproduction that demonstrates the problem when you run cargo test. The reproduction has 3 test cases:

• A test which uses actix_web::test, which is capable of spawning an actor.
• A test which uses sqlx::test, which panics when it attempts to spawn an actor.
• A test which uses sqlx::test, which does not attempt to spawn an actor, and does not panic.

The issue seems a bit beyond me to diagnose, but I'm guessing that the sqlx::test macro sets up a runtime that is just a little bit different from the one actix-web sets up internally, and that leads to the panic in question:

---- another_account::registration_creates_unverified_credentials_and_token stdout ----
thread 'another_account::registration_creates_unverified_credentials_and_token' panicked at '`spawn_local` called from outside of a `task::LocalSet`'

I'm wondering mainly if this is a legit issue, since I can't turn up any similar problems that don't amount to PEBCAK. Is this a legit problem with trying to use these 2 systems at once? Am I doing something wrong here?

If they're just incompatible, that's unfortunate, but I'll figure out how to work around it, because IMO sqlx::test is a far bigger win for test harness setup than any other macros, and I'm so glad to have it (thanks for that).

Ideally I'm just missing something obvious, though, that y'all know about?

[abonander]

#[sqlx::test] doesn't start an actix_rt::System. The runtime-actix features were changed to aliases of the runtime-tokio features, actually by a PR from @robjtede in #1679, because actix-web is meant to be fully compatible with Tokio now.

You can probably start your actor from within the closure passed to HttpServer::new() as I'd expect that to be run on each worker thread when it starts. If you only want it to be run once you could protect that with a std::sync::Once or a simple AtomicBool in a static.

[esmevane]

Oh! Thank you so much; putting invocations in the HttpServer::new closure did it. I can definitely design with that in mind - glad this was just me and not a larger thing!

[mike-lloyd03]

@abonander I'm getting the same error in my tests but I am not using actors. Some of my API endpoints are sending get requests using awc. How can I run these tests on the same runtime that is started by the sqlx::test macro?

[robjtede]

I guess the sqlx::test macro simply isn't compatible with things that expect implicit access to a LocalSet.

[esmevane]

Seconding @robjtede - there are some limitations to using sqlx::test and LocalSet is one. It's still very worth it IMO but it constrains what sort of testing approaches you can take.

[tolumide-ng]

Excellent job with sqlx, @abonander — thank you! Do you have any suggestions or hints on how to work around using sqlx::test with something like actix_rt::test in this situation? As mentioned earlier in the thread, it seems that sqlx::test is not compatible with using LocalSet.

[abonander]

You should be able to run your own LocalSet within a test function: https://docs.rs/tokio/latest/tokio/task/struct.LocalSet.html#use-with-run_until

[tolumide-ng]

that works fine, thank you!