Same Event Applied Twice when Generating Aggregate Snapshot

[pmclellan]

Hi,

I'm currently experiencing some unexpected behaviour related to aggregate snapshotting and I suspect it may be due to a bug.

I'm using the EventCountSnapshotTriggerDefinition to control when aggregate snapshots are created. This is working fine and snapshots are being triggered when required. However, when the AggregateSnapshotter constructs the snapshot event it publishes the most recent event twice. For example, if the event count threshold is set to 100 then the snapshotter will publish event 99 twice. In most cases this is not an issue. However, there are some events in my current implementation for which the handling is not idempotent and this is creating aggregates with an invalid state.

When I debug the snapshotting process I can see that the duplicate event is being added to the event stream when a call is made to AbstractEventBus.queuedMessages() This method returns a list of all events held by the current unit of work and its parents. In our case, the parent BatchingUnitOfWork and its child DefaultUnitOfWork both hold a reference to the same event, which leads to it being duplicated in the resultant list.

Please let me know if you require any further details regarding this issue.

Thanks,
Paul

[abuijze]

Hi Paul,

thanks for the elaborate report. Since you mentioned the BatchingUnitOfWork and the DefaultUnitOfWork, am I correct in assuming you're using an in-process snapshotter? In other words, you did not configure an ExecutorService on the snapshotter?

[pmclellan]

Hi Allard,

thanks for the prompt response. You're correct, we haven't specified an ExecutorService for the snapshotter. We're using the SpringAggregateSnapshotterFactoryBean with its default configuration.

[abuijze]

Hi Paul,

I'm unable to reproduce the problem on the master branch. Looking at the code, it seems that Axon already takes into account that some of the events in the "staged events" may overlap with what's returning from the Event Store, as these events may, or may not have been stored there yet.

Which version of Axon are you seeing this behavior on?

[pmclellan]

Hi,

we're currently using version 4.3. I think the problem we're seeing isn't due to an overlap between "staged events" and the contents of the event store, it's an overlap between the "staged events" held by units of work with a parent->child relationship. In our case the parent BatchingUnitOfWork and its child DefaultUnitOrWork hold the same staged event so the queuedMessages method is adding it to the list twice.

[abuijze]

Thanks for the clarification. I managed to find the culprit and resolved it in #1416

[pmclellan]

Great! Thanks Allard.

[smcvb]

Closing this issue as it has been resolved in pull request #1416.