Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Drop Rocket inside the tokio async context when using launch #2822

Open
wants to merge 6 commits into
base: master
Choose a base branch
from

Conversation

the10thWiz
Copy link
Collaborator

Moves error printing and the implicit drop inside the tokio runtime when using the #[launch] attribute.

Based on an issue reported on Element. See https://matrix.to/#/!kDIcCXWSVfdahNCJWq:mozilla.org/$2CSzlNC4rQw7Wd7Y7L2fV-2_fTxsvq93elL1foVfQ1s?via=mozilla.org&via=matrix.org&via=catgirl.cloud.

Basically, the Rocket instance gets dropped (after a successful shutdown or a Shutdown error) outside the tokio runtime. Some minor restructuring of the generated code ensures the tokio context outlives Rocket, if only slightly.

Copy link
Member

@SergioBenitez SergioBenitez left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What is this fixing/making possible? Can we add that as a test? If it's system-wide, perhaps an addition to the testbench would be reasonable.

Copy link
Collaborator Author

@the10thWiz the10thWiz left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Needs tests

@the10thWiz the10thWiz force-pushed the drop-in-async-launch branch 2 times, most recently from c905bcb to b47211c Compare August 10, 2024 04:18
Copy link
Member

@SergioBenitez SergioBenitez left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The changes to launch() are good, but I want the test to live in testbench. It'll reduce the amount of code considerably while making the test robust.

Comment on lines 40 to 67
mod main {
use rocket::tokio::net::TcpListener;

#[rocket::main]
async fn main() {
super::rocket()
.try_launch_on(TcpListener::bind("localhost:8001"))
.await
.unwrap();
}
#[test]
fn test_main() {
main();
}
#[test]
fn test_execute() {
rocket::execute(async {
super::rocket()
.try_launch_on(TcpListener::bind("localhost:8002"))
.await
.unwrap();
});
}
}
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is extremely fragile. The only correct way to start a TCP server in tests is to do so with an unnumbered port.

Given what this code is doing, I really believe this needs to be a test in testbench. The tests are easier to write this way, too. All you need to do is introduce a new module in testbench/src/servers. The returned handle allows you to shutdown the server as well.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yeah, I should switch to using an unnumbered port.

That being said, I'm not confident I can use testbench for this specific case - at least not without some issues. The challenge is that I'm trying to test the #[launch] macro itself - it looks like most cases pass a rocket instance to the spawn macro, which then attaches a fairing. I can't do that - since I'm directly executing the generated (sync) main function.;

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I've taken the time to make an attempt (it required some modifications to testbench, and still doesn't work). You can check it out from the 'drop-in-async-launch-testbench' branch of my fork.

Copy link
Member

@SergioBenitez SergioBenitez left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Take a look at the new test I added, which currently fails.

Comment on lines +64 to +66
fn test_execute_directly() {
rocket::execute(super::rocket().launch()).unwrap();
}
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I added this test, which I think should pass for the same reasons?

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What's happening here, is Rocket is created and launched, on the async runtime. However, launch returns Result<Rocket<Ignite>, Error>. In the success case, Rocket is then returned by rocket::execute, and then dropped outside of it (and the async runtime). For 0.6, we should consider changing the interface of rocket::execute to fn execute<P>(f: impl Future<Output = Rocket<P>>) -> Result<(), Error>, and calling launch internally.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is why the one I wrote explicitly calls unwrap inside an async block.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The main use-case is exactly calling it with rocket.launch() directly, so it should be able to do that without any footguns or extra ceremony.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The main issue with execute right now is that it returns the Rocket instance. The runtime only exists inside the execute function, so it has to be dropped before returning.

For my proposal, rocket::execute(build()) would be roughly the same as rocket::execute(build().launch()) right now, but drop the rocket instance before returning, and the return type would be Result<(), Error> instead of Result<Rocket<Ignite>, Error>. This would have the effect of making execute only be useful for launching a Rocket instance.

We could (equivalently) change Rocket::launch() to evaluate to Result<(), Error>. This would also allow reverting the changes in the launch macro, since the launch method would take of dropping the Rocket instance in the async context.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm wondering if there's a better approach altogether. In addition to trying to drop Rocket in an async context, why not also specifically drop managed state on shutdown? That is, as part of the shutdown procedure, which is async, drop managed state.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think that might be a valid approach, but the tradeoff would be that we can't support relaunching a Rocket application after it's been shutdown.

Given all this, what is the value to Rocket returning the instance after shutdown? The two values I can see is a) relaunching Rocket after shutdown, and b) inspecting managed state after shutdown. The fact that Rocket is dropped in the error case mean it's likely not super useful to inspect managed state, since that's when you would actually want to inspect it. I'm not sure what the exact use case of relaunching Rocket is, and I'm not convinced it wouldn't be better to recommend simply building a new instance of Rocket.

I think the best compromise would be changing launch to not return the Rocket instance, ever. This ensures Rocket is dropped inside an async context (since it's dropped before the launch future completes). Alternatively, we could create a Shutdown phase to return, which would have already dropped managed state, and be safe drop outside an async context.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants