Hey guys,

A couple questions/comments:

1. The callback and user-controlled commit offset functionality is already
in the new consumer which we are working on in parallel. If we accelerated
that work it might help concentrate efforts. I admit this might take
slightly longer in calendar time but could still probably get done this
quarter. Have you guys considered that approach?

2. I think partitioning on the hash of the topic partition is not a very
good idea because that will make the case of going from a cluster with
fewer partitions to one with more partitions not work. I think an intuitive
way to do this would be the following:
a. Default behavior: Just do what the producer does. I.e. if you specify a
key use it for partitioning, if not just partition in a round-robin fashion.
b. Add a --preserve-partition option that will explicitly inherent the
partition from the source irrespective of whether there is a key or which
partition that key would hash to.

3. You don't actually give the ConsumerRebalanceListener interface. What is
that going to look like?

4. What is MirrorMakerRecord? I think ideally the MirrorMakerMessageHandler
interface would take a ConsumerRecord as input and return a ProducerRecord,
right? That would allow you to transform the key, value, partition, or
destination topic...

5. Have you guys thought about what the implementation will look like in
terms of threading architecture etc with the new consumer? That will be
soon so even if we aren't starting with that let's make sure we can get rid
of a lot of the current mirror maker accidental complexity in terms of
threads and queues when we move to that.

-Jay

On Tue, Jan 20, 2015 at 4:31 PM, Jiangjie Qin <j...@linkedin.com.invalid>
wrote:

> Hi Kafka Devs,
>
> We are working on Kafka Mirror Maker enhancement. A KIP is posted to
> document and discuss on the followings:
> 1. KAFKA-1650: No Data loss mirror maker change
> 2. KAFKA-1839: To allow partition aware mirror.
> 3. KAFKA-1840: To allow message filtering/format conversion
> Feedbacks are welcome. Please let us know if you have any questions or
> concerns.
>
> Thanks.
>
> Jiangjie (Becket) Qin
>

Reply via email to