Hello Francesco,
Streams auto-created repartition topics's num.partitions are determined by
the num.tasks of the writing sub-topology, which is then determined by the
source topic's num.partitions in turn. There are some proposals about
extending this coupling but not yet implemented:
https://cwiki.apache.org/confluence/display/KAFKA/KIP-221%3A+Repartition+Topic+Hints+in+Streams
So for your scenario, for now you'll have to manually create the "through"
topic explicitly before starting the streams app with the num.partitions
you want.
Guozhang
On Tue, Jul 31, 2018 at 6:37 AM, Francesco Frontera <
francesco.front...@radicalbit.io> wrote:
> Hi,
>
> I have a question about topic repartitioning in Kafka Streams using
> `through` function.
>
> I try to explain the context Briefly:
>
> I have single topic A with two partitions:
> A:1:9
> A:0:0
>
> I try to create a repartitioned topic using Kafka Streams API:
>
> builder.stream("A").map<>((key, val) => KeyValueMapper
> {..selectKey(val)}).through("B", Produced.`with`(....))
>
> the streaming job produced a topic with a single partition (B:0:9) instead
> of a topic with 2 partitions, which is something I want to achieve (i.e.
> retaining the upstream partitions number).
>
> Is there a way to create an auxiliary topic with specific partitions
> directly from Kafka Streams API without creating topic explicitly (similar
> to join operation)?
>
> If the join neither does it, am I then forced to create manually the
> internal topics with the desired number of partitions for internal
> repartitioning? Is this valid also for all the KafkaStreams operator's
> output topics?
>
>
> Thanks,
> Francesco Frontera.
>
--
-- Guozhang
