Improve ReactiveEventPublisher throughtput

[fjtirado]

EventPublisher has the list of events generated at the end of a UnitOfWork

These events are currently sent one by one by ReactiveEventMessagePublisher, which is the case of Http Smallrye connector heavily affects system throughput, which in turn increase memory consumption because of event queuing.

One possible improvement is to set the message body to the entire collection, minimizing the number of external calls. However this will affect consumers. For example, in DataIndex, ReactiveMessageEventConsumer will have to be changed to check if the message body contains one event or a collection of them.

[rgolangh]

What is the root cause of the sending being too slow? I remember that the numbers of events in the stress tests were not big.

[fjtirado]

What is the root cause of the sending being too slow? I remember that the numbers of events in the stress tests were not big.

@rgolangh The number of processes was not too big, but the number of events per process is.

[fjtirado]

This issue depends on this one quarkiverse/quarkus-reactive-messaging-http#281

[fjtirado]

@tkobayas
its really scary ;)
anyway, it can be easily fixed if it ever happen again, thanks for the heads up!

[tkobayas]

^ @fjtirado Probably because the commit message Fix apache/incubator-kie-issues#1457 contains apache so, it can close the upstream issue even from a fork?