Event Handler called twice with the same event

Sara_Pellegrini · January 9, 2018, 4:41pm

Hi everybody,
I’m facing some problems using TrackingProcessors.

Seems that sometimes, some handlers are called twice.
It happens expecially if the handler method is very fast, this made me suppose it is a critical race issue.
It happens only where there is a single event to process.

My axonframework version is 3.1, spring boot version 1.5.7.RELEASE.

I’m using an EmbeddedEventStore persited on MySql Database.
The event handler method doesn’t persist anything on db.

Temporarily, I solved the problem using an interceptor:

`


@Slf4j
public class PreventEventDuplicationInterceptor implements MessageHandlerInterceptor<EventMessage<?>> {

    private String lastEventIdentifier = "";

    @Override
    public Object handle(UnitOfWork<? extends EventMessage<?>> unitOfWork, InterceptorChain interceptorChain) throws Exception {

        Message<Object> message = (Message<Object>) unitOfWork.getMessage();
        String eventName = message.getPayloadType().getSimpleName();
        String eventIdentifier = message.getIdentifier();
        unitOfWork.onCommit(u -> lastEventIdentifier = eventIdentifier);

        if (lastEventIdentifier.equals(eventIdentifier)) {
            log.error("Duplication in " + Thread.currentThread().getName() + " event " + eventName + " " + eventIdentifier + "[" + this + "]");
            return null;
        }

        return interceptorChain.proceed();
    }
}

`

Anyone else is experiencing the same issue?

Thanks
s

Troy_Hart1 · January 9, 2018, 8:57pm

I am also seeing the same behavior with axon 3.1.1 and spring boot 1.5.9.RELEASE. I’m not sure if my processors or SubscribingEventProcessor or TrackingEventProcessor. I believe they are subscribing because I’ve not specifically configured the processors…I’ve simply configured the aggregate repositories. I’m still totally new to axon so there are many disconnects in my understanding/comprehension.

I’m simply debugging aggregate tests (through AggregateTestFixture) and when I put a break point in one of my @EventSourcingHandler methods I notice that the same event is coming in twice in response to a single command.

allardbz · January 9, 2018, 9:28pm

Hi Troy, Sara,

in tests, events do come in twice, and that’s by design. After executing the test scenario, Axon will recreate the aggregate another time, this time based only on events, to verify that the state of the aggregate is identical. Axon uses this to ensure state is not modified (accidentally) outside of EventSourcingHandlers.

Sara, with “it only happens when there is a single event to process”, do you mean that the processor is just starting and processing a single event, or it is still polling at the head of the event stream, and then a single event arrives?

Cheers,

Allard

Troy_Hart1 · January 9, 2018, 9:55pm

Ah, thanks Allard! That makes sense.

Sara_Pellegrini · January 10, 2018, 5:35pm

Hi Allard,

I mean that the processor is still polling at the head of the event stream, and then a single event arrives.
I hope to have some time during this weekend to try to reproduce it in a sample project.

Thanks
s

Sara_Pellegrini · January 11, 2018, 11:59am

I have READ-COMMITTED isolation level
BTW, It seems not related to isolation level, it happens also with repeteable read and serializable.

allardbz · January 12, 2018, 10:53am

Hi Sara,

are you using the JpaEventStorageEngine or the JdbcEventStorageEngine?

What I’m curious about, is where the duplicate appears. It sounds like you’re able to reproduce it. Can you share that code?
It could show whether the duplicate appears in the EventStorageEngine itself (returning the same event as part of 2 batches), or whether it is the EmbeddedEventStore, where an issue arises in the combining and switching of streams.

Thanks.

Allard

Sara_Pellegrini · January 16, 2018, 9:56am

Hi Allard,
I’m using the JpaEventStorageEngine.

I’m trying to find a way to reproduce the case. Seems that if the new events are more than one, it doens’t happen.
I suppose it happens when the size of the last batch is = 1 and only if the processor is behind global cache.

As soon as I find a way to make it replicable I send you the code.

Many thanks

Sara_Pellegrini · January 17, 2018, 9:36am

Hi Allard,

I’ve isolated what happens in EmbeddedEventStore for a single processor thread affected by the problem.

When I have 2 or more events:

…