Evaluate the AsyncIOTest test

[jamezp]

The Netty org.jboss.resteasy.test.AsyncIOTest is very flaky. There is an issue with determining if something is running on the IO thread or not. This is determined with the test writer:

String resp = t.expectOnIoThread == Context.isOnEventLoopThread() ? "OK" : "KO";

It looks like the expectation for the test is that it it's not supposed to be on the IO thread, a io.vertx.core.Context.isOnEventLoopThread() thread, then it's expected to be on the thread used in a CompletableFuture.supplyAsync(). This does not always seem to be true for some reason. Whether there is a real race condition on when the current thread would be changed or there is an issue with the test alone is undetermined.

I did however, remove all other invocations of endpoint checks from the test and left one which fails every time. This could be an indicator on the test just being bad. This needs to be looked at closer though.

[Yaytay]

Hi,
Any luck figuring out when this fails?

[jamezp]

@Yaytay I've not had the time to dig into this yet.

[jamezp]

I send the response too soon :) I do, however, think it's just a test issue. I think there is a race on the response line:

String resp = t.expectOnIoThread == Context.isOnEventLoopThread() ? "OK" : "KO";

I think the test itself likely just needs to be updated to either avoid the race, not sure how that could be done yet, or we just need to update the test to do something completely differently.

[Yaytay]

I haven't seen the failure yet, when does it bite?

[jamezp]

The test is disabled now, but it was failing maybe 20% or so of the time. The test attempts to use the CompletableFuture.asyncSupply() so there is a race between which thread the Context.isOnEventLoopThread() is executed on. I'm 99% sure it's just issue though.

[Yaytay]

This has been doing my head in, but I finally have some kind of understanding of why it's failing.
It is never running on a Vert.x worker thread, it is either running in the Vert.x IO thread or in the JDK thread pool and the race condition comes in whether the completion of the Future from the supplyAsync happens immediately (in the thread pool) or via the Future kicking off the event loop in the IO thread.

My understanding it doesn't really matter to you, but I am very keen to get a release of a Resteasy that works with Vert.x 5 and I couldn't push for that with this being outstanding :)

[jamezp]

hehe, you're analysis is exactly the same I came up with @Yaytay. I think I'll merge my PR and I can definitely get a release out for you.

[jamezp]

@Yaytay FWIW there is a https://github.com/resteasy/resteasy-vertx/releases/tag/v2.0.0.Beta1 release which is based on Vert.x 5.