Enno Runne
Enno Runne
In the scheduled build https://travis-ci.org/github/akka/alpakka-kafka/jobs/725973526
https://travis-ci.org/akka/alpakka-kafka/jobs/640635810#L1385 in #1026
https://travis-ci.org/akka/alpakka-kafka/jobs/648052161#L446 in master
Thank you for reporting this. We changed the internals of committing quite a bit in 1.1 and 2.0, something might have slipped in there.
In the scheduled build https://travis-ci.org/github/akka/alpakka-kafka/jobs/725973527#L953
Thank you @gabrielreid for pointing this out. Not taking commit callbacks into account for the backpressure is by design and desired behaviour. But I agree, it could be documented. Not...
I wonder if we could come up with something that would support your use with the `CommittingProducerSinkStage`. Can you give an indication of what numbers we're talking about? Do you...
What was the idea behind `max-batch = 25`? With the new aggregation of commits in the consumer actor, this only leads to more actor messages being sent to the consumer...
> Would it work to just increment and decrement a counter of the currently in-flight `commitAsync` calls in `KafkaConsumerActor#commit`, and trigger the `requestDelayedPoll()` based on that counter? That is one...
I believe this would be best implemented as something that retrieves the current offset and creates subscriptions from those offsets. This would benefit from implementing https://github.com/akka/alpakka-kafka/issues/765 first, I guess.