Fix flaky test in reactive-messaging-hibernate-orm

[ozangunalp]

I think when the hibernate orm state store test has been added to that IT the test using Emitter in @Transactional method reproduced the issue in #21948 .

To fix the flaky test the method using transactional no longer returns a completion stage.

The original issue is still to investigate.

[gsmet]

I spotted a small thing.

Now, tbh, I have no real idea of what you're doing :).

[ozangunalp]

@Sanne @cescoffier to give more context: This test was stable before I add the other one involving Hibernate transactions.

So this is the code that started to behave flaky:

@Incoming("fruits-in")
@Transactional
public CompletionStage<Void> persist(Message<Fruit> fruit) {
    fruit.getPayload().persist();
    return emitter.sendMessage(fruit).subscribeAsCompletionStage();
}

I verified that the completion stage completes on the same Vert.x context (blocking) as the method call. But then I cannot reproduce it locally :)

[Sanne]

@ozangunalp thanks! I wonder if it could relate to #29159

[ozangunalp]

The MutinyEmitter makes sure that the caller Vert.x context is preserved when the send Uni returns, but in this case the thread won't be the same, as it'll return on the event loop instead of the worker thread.

Looking at the TransactionalInterceptorBase, when handling the CompletionStage return type, the transaction will end in another thread, (but same Vert.x context if it makes any difference). It seems to handle the case and install the right transaction to end.

[ozangunalp]

The test is still flaky

[ozangunalp]

Trying to remove the named persistence-unit configured to use with the checkpointing state store in the test.

[ozangunalp]

First run was green. Triggered a re-run

[ozangunalp]

The test is still flaky. I'll disable the test in this PR.

I don't think that I am doing anything wrong on HibernateOrmStateStore, maybe it reproduces that issue more consistently - I still cannot reproduce it locally. @Sanne can you take a look when you have some time?

[gsmet]

Let's keep this PR open as a basis to dig into this issue but I created a simple PR that we can merge right away there: #29478 .

[ozangunalp]

@gsmet @cescoffier I think we can take this in. I added a note on the Kafka guide example
The issue was sending the Hibernate entity object to Kafka during the transaction. It will run Kafka serialization on the sender thread and somehow Hibernate is not happy about it.

[github-actions]

🙈 The PR is closed and the preview is expired.