Enno Runne
Enno Runne
> Delay consumer actor stop only for non-failure cancellations As `CommittingSpec` shows, it is relevant to leave the consumer alive so that emergency commits from the `CommitCollectorStage` have a chance...
> Is this PR changing anything else than logging? If so, could you highlight what that it is? The original idea didn't work out so easily. The remaining bit is...
The improvements in this PR are too small for this to be relevant.
https://travis-ci.org/akka/alpakka-kafka/jobs/640635808#L649 in #1026
https://travis-ci.org/akka/alpakka-kafka/jobs/648052162#L586 in master
It shows in a new variant in `master`: ``` - should collect commits to be sent to commitAsync *** FAILED *** (1 minute, 40 seconds) the commits are aggregated to...
Let's deprecate them in 0.10 and remove the sources some later release.
After the switch to Akka 2.6 in #1209 the sources can easily distinguish the reason for cancellation/downstream finish and could shut down without any stop timeout in case of errors.
The consumer is kept around to allow for more commits to reach the Kafka broker. Even in the case of a stream failure, the commit collector will try to fire...
Thank you for this report and the reproducer. The Akka Streams internal `FixedSizeBuffer.ModuloFixedSizeBuffer` is used by the `mapAsync` operator to ensure the order of outgoing elements. This buffer is not...