Hi Mike,
FYI, support for "max.poll.records" was added in
https://github.com/apache/kafka/pull/931 (KAFKA-3007) which was not present
in the streams tech preview release. It will however be in 0.10
Cheers,
Geoff
On Thu, Apr 7, 2016 at 4:58 AM, Michael D. Coon
wrote:
> One more thing I'm notici
One more thing I'm noticing in the logs.
I see periodic node disconnection messages due to "timeout". I set my
metadata.fetch.timeout.ms to 6, request.timeout.ms to 3 and timeout.ms
to 3 and those should be more than enough time waiting for metadata
responses. I also set my offset co
Guozhang,
Thanks for the advice; however, "max.poll.records" doesn't seem to be
supported since it's not affecting how many records are coming back from the
consumer.poll requests. However, I agree that the likely culprit in rebalancing
is the delay in processing new records. I'm going to try
Hi Michael,
Your issue seems like a more general one with the new Kafka Consumer
regarding unexpected rebalances: as for Kafka Streams, it's committing
behavior is synchronous, i.e. triggering "consumer.commitSync" of the
underlying new Kafka Consumer, which will fail if there is an ongoing
rebala
