In Kafka Streams the producer config `PARTITIONER_CLASS_CONFIG` does not take effect, because Kafka Streams computes and set partition numbers explicitly and thus the producer does never use the partitioner to compute a partition, but accepts whatever Kafka Streams specifies on each `ProducerRecord`.
If you want to change the partitioning strategy, you need to implement a custom `StreamPartitioner` and pass it into the corresponding methods. For example, `to()` or `through()`. Hope this helps. -Matthias On 11/14/19 9:51 AM, Mikkel Gadegaard wrote: > I've set up a POC using KafkaStreams with microservices consuming and > producing from/to topics. In the beginning I hadn't thought about > partition strategy, and so I was using the DefaultPartitioner for producer > partition assignments. My messages have keys (I use these for > forking/joining), and the keys are time based UUIDs, this causes some > rather uneven distribution on my topics. I looked around google and > stumbled on KIP-369 (Alternative Partitioner to Support "Always > Round-Robin" Selection) and figured that would be what I needed, so since > 2.4 isn't out yet I borrowed the class from the PR on github, added it to > my project and added the property to my config, like so: > > streamProperties.put(ProducerConfig.PARTITIONER_CLASS_CONFIG, > RoundRobinPartitioner.class.getCanonicalName()); > > > And the round robin strategy works on a newly added topic, spreading > messages evenly over 4 partitions. But, and I'm finally getting to my > question, it doesn't seem to have any effect on existing topics, in other > words, it seems to be continuing to use the DefaultPartitioner for topics > created before I added the RoundRobinPartioner class to my > project/properties. > > Is it me that just hasn't understood that it is impossible to change > strategy for an existing partition or do I have to do something specific > apart from re-deploying the Microservice containing the producer? > > Thanks > Mikkel > > -- >
signature.asc
Description: OpenPGP digital signature
