Hello Jiang, In Kafka parallelism is achieve through partitioning, meaning you can specify n partitions for a topic and m (<=n) consumers each consuming from one or more partitions.
Guozhang On Wed, Aug 14, 2013 at 12:59 PM, Wu, Jiang2 <[email protected]> wrote: > Hi, > > In our experiments, we find that if multiple consumers in the same group > listen to the same partition, then one consumer will receive all messages > on this partition, and others get none. Is this inherent to Kafka design, > or it can be changed by some configuration? Basically we expect ems queue > behavior, i.e., each of the n consumers receive about 1/n of the total > messages. > > Thanks, > Jiang > > -- -- Guozhang
