It depends on your version. The behavior is known and we put one improvement into 1.1 release: https://github.com/apache/kafka/pull/4410
Thus, it's "by design" (for 1.0 and older) but we we want to improve it. Cf: https://issues.apache.org/jira/browse/KAFKA-4969 -Matthias On 5/13/18 7:52 PM, Liam Clarke wrote: > Hi all, > > We are running a KStreaming app with a basic topology of > > consume from topic A -> transform and write through topic B (making the app > a consumer of topic B also) -> finally write to topic C > > We are running it with two instances of the application. Topic A has 100 > partitions, topics B and C both have 10 partitions. > > What we've noticed in the consumer group info that is confusing us is that > instance 1 of the app will own 45 partitions of topic A and all 10 > partitions of topic B, whereas the 2nd instance will own 55 partitions of > topic A and 0 of topic B. > > Our expectation is that we'd see an even split of partitions across both > instances - is this a realistic expectation? Or is this working as intended? > > Regards, > > Liam Clarke >
signature.asc
Description: OpenPGP digital signature
