Quarkus 3.0.1 behaves different in reactive native mode

[michael-simons]

This is a reminder for @maxandersen for

quarkiverse/quarkus-neo4j#161 (comment)

and quarkiverse/quarkus-neo4j#165 (comment)

We didn't change anything on hour side between 3.0.0 and 3.0.1 and I am out of options for that.

Issues persists with 3.0.2, too

[maxandersen]

@Sanne @cescoffier difference between hibernate reactive
Or mutiny on native?

[michael-simons]

Mutiny most likely, we are not using hibernate in any way or form.

[gsmet]

/cc @jponge

[jponge]

There's nothing native-specific in Mutiny, and the fact that the tests pass in JVM mode suggest that it might rather be an issue with the driver / integration itself.

What exact version of Mutiny is involved?

/cc @cescoffier

[michael-simons]

Well it worked with quarkus 3.0.0 and fails on 3.0.1, that’s the only changed piece so far. Am 03.05.2023 um 18:50 schrieb Julien Ponge ***@***.***>:  There's nothing native-specific in Mutiny, and the fact that the tests pass in JVM mode suggest that it might rather be an issue with the driver / integration itself. What exact version of Mutiny is involved? /cc @cescoffier <https://github.com/cescoffier> — Reply to this email directly, view it on GitHub <#33106 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAEAQL7PKTKWDSZD5GLJLKLXEKEF3ANCNFSM6AAAAAAXURD7FY> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[jponge]

Both Quarkus 3.0.0.Final and 3.0.1.Final use Mutiny 2.1.0.

[michael-simons]

So, I have now managed to create a full reproducer.

Please bring up a Neo4j instance with

docker run --publish=7474:7474 --publish=7687:7687 -e 'NEO4J_AUTH=neo4j/verysecret' neo4j:5.7

Download the attached project full-reproducer.zip and compile and run with

full-reproducer.zip

./mvnw -Dnative clean package
./target/code-with-quarkus-1.0.0-SNAPSHOT-runner

After that, compare the both the following resources GreetingResource which is broken and FixedResource, which works.

Check with

curl -v  localhost:8080/fixed # returns 10 items, 
curl -v  localhost:8080/broken # hangs, while the application clearly queries the database and closes the session

The difference for the resources? The methods behind the paths above are literally the same:

@Path("/broken")
public class GreetingResource {

	@Inject
	Driver driver;

	static Uni<Void> sessionFinalizer(ReactiveSession session) { // <.>
		return Uni.createFrom().publisher(session.close()).onItem()
			.invoke(() -> System.out.println("Session closed"))
			.replaceWithVoid();
	}

	private Multi<Long> get0() {
		return Multi.createFrom().resource(() -> driver.session(ReactiveSession.class), // <.>
				session -> session.executeRead(tx -> {
					var result = tx.run("MATCH (f:Foobar) RETURN f.value as value ORDER BY f.value");
					System.out.println("Created result, now publishing");
					return Multi.createFrom().publisher(result).flatMap(ReactiveResult::records);
				}))
			.withFinalizer(GreetingResource::sessionFinalizer) // <.>
			.map(record -> record.get("value").asLong());
	}

	@GET
	@Produces(MediaType.SERVER_SENT_EVENTS)
	public Flow.Publisher<Long> get() {

		return get0();

	}
}

The difference in the working resource is one additional method:

	@GET
	@Produces(MediaType.SERVER_SENT_EVENTS)
	@Path("/asMulti")
	public Multi<Long> asMulti() {

		return get0();

	}

Both - the broken and the fixed version - work in JVM mode. That indicates to me that native image and the method type detection of the JAX RS resources changed between Quarkus 3.0.0 and 3.0.1

The last indicator:

This resource does not work at all in native Mode:

@Path("/noNeo")
public class NoNeoResource {

	@GET
	@Produces(MediaType.SERVER_SENT_EVENTS)
	public Flow.Publisher<Long> get() {

		return Multi.createFrom().items(LongStream.rangeClosed(1, 10).boxed());

	}
}

Curling it curl -v localhost:8080/noNeo yields in

2023-05-04 11:05:19,973 ERROR [io.qua.ver.cor.run.VertxCoreRecorder] (vert.x-eventloop-thread-3) Uncaught exception received by Vert.x: java.lang.RuntimeException: java.lang.NoSuchMethodException: org.acme.NoNeoResource.get()
	at org.jboss.resteasy.reactive.server.spi.ResteasyReactiveResourceInfo.getMethod(ResteasyReactiveResourceInfo.java:72)
	at org.jboss.resteasy.reactive.server.spi.ResteasyReactiveResourceInfo.getAnnotations(ResteasyReactiveResourceInfo.java:89)
	at org.jboss.resteasy.reactive.server.core.ResteasyReactiveRequestContext.getMethodAnnotations(ResteasyReactiveRequestContext.java:553)
	at org.jboss.resteasy.reactive.server.core.ResteasyReactiveRequestContext.getAllAnnotations(ResteasyReactiveRequestContext.java:531)
	at org.jboss.resteasy.reactive.server.core.SseUtil.serialiseDataToString(SseUtil.java:137)
	at org.jboss.resteasy.reactive.server.core.SseUtil.serialiseEvent(SseUtil.java:74)
	at org.jboss.resteasy.reactive.server.core.SseUtil.send(SseUtil.java:39)
	at org.jboss.resteasy.reactive.server.handlers.PublisherResponseHandler$SseMultiSubscriber.onNext(PublisherResponseHandler.java:61)
	at io.smallrye.mutiny.helpers.HalfSerializer.onNext(HalfSerializer.java:30)
	at io.smallrye.mutiny.helpers.StrictMultiSubscriber.onItem(StrictMultiSubscriber.java:84)
	at io.smallrye.mutiny.operators.multi.builders.StreamBasedMulti$StreamSubscription.pull(StreamBasedMulti.java:113)
	at io.smallrye.mutiny.operators.multi.builders.StreamBasedMulti$StreamSubscription.request(StreamBasedMulti.java:87)

I have created both examples for pure mutiny and Quarkus without JAX RS, attached as well. They don't expose the problem.

pure-mutiny-no-issues.zip
quarkus-cmd-no-issues.zip

Have fun ;)

[cescoffier]

Thanks for all the information @michael-simons!

It might be because of f73ec98, but I'm not totally sure.

(Note: we should add a test for this)

[michael-simons]

My pleasure, Clement!

This is a good candidate for test in native functionality

@Path("/noNeo")
public class NoNeoResource {

	@GET
	@Produces(MediaType.SERVER_SENT_EVENTS)
	public Flow.Publisher<Long> get() {

		return Multi.createFrom().items(LongStream.rangeClosed(1, 10).boxed());

	}
}

it works as expected in JVM, not in native. Didn't get anything smaller reproduced.

[michael-simons]

That condition here f73ec98#diff-f687ef5f49c43c73a60a5b8fbd573e7d38a827fce537bd3097b71a4753635f68R553 would explain why the above resource would start working when you add another method returning Multi (but not Uni). So in fact I think there are two cases missing, checking for Uni and Flow.Publisher return types.

[cescoffier]

Uni cannot be used for SSE (does not make sense). But yes...

[cescoffier]

@michael-simons Can you try with #33146?

It passed my local test, and I also added an IT.

(I need to catch a plane, I will continue on Tuesday if it does not work).

[michael-simons]

I can confirm that it works for all my use cases in the full-reproducer project above, with and without Neo4j. Excellent work thank you!