Kafka Streams RoundRobinPartitioner

问题

I wrote a kafka streams code that uses kafka 2.4 kafka client version and kafka 2.2 server version. I have 50 partition on my topic & internal topic.

My kafka stream code has selectKey() DSL operation and I have 2 million of record using same KEY. In the stream configuration, I have done

props.put(ProducerConfig.PARTITIONER_CLASS_CONFIG, RoundRobinPartitioner.class);

So that I am able to use different partitions with the exactly same key. If I dont use Round Robin as expected all my messages go to same partition.

Everything is ok untill now but I realized that; when I use RoundRobinPartitioner class my messages go like ~40 partitions. 10 partition is in the idle state. I wonder what am I missing ? It should use 50 of them about 2 million of records right?

      final KStream<String, IdListExportMessage> exportedDeviceIdsStream =
            builder.stream("deviceIds");

        // k: appId::deviceId, v: device
        final KTable<String, Device> deviceTable = builder.table(
            "device",
            Consumed.with(Serdes.String(), deviceSerde)
        );
            // Some DSL operations
            .join(
                deviceTable,
                (exportedDevice, device) -> {
                    exportedDevice.setDevice(device);

                    return exportedDevice;
                },
                Joined.with(Serdes.String(), exportedDeviceSerde, deviceSerde)
            )
            .selectKey((deviceId, exportedDevice) -> exportedDevice.getDevice().getId())
            .to("bulk_consumer");

And

   props.put(StreamsConfig.STATE_DIR_CONFIG, /tmp/kafka-streams);
   props.put(StreamsConfig.REPLICATION_FACTOR_CONFIG, 3);
   props.put(StreamsConfig.NUM_STANDBY_REPLICAS_CONFIG, 2);
   props.put(StreamsConfig.COMMIT_INTERVAL_MS_CONFIG, 100);
   props.put(StreamsConfig.PROCESSING_GUARANTEE_CONFIG, EXACTLY_ONCE);
   props.put("num.stream.threads", 10);
   props.put("application.id", applicationId);

RoundRobinPartitioner.java

public class RoundRobinPartitioner implements Partitioner {
    private final ConcurrentMap<String, AtomicInteger> topicCounterMap = new ConcurrentHashMap();

    public RoundRobinPartitioner() {
    }

    public void configure(Map<String, ?> configs) {
    }

    public int partition(String topic, Object key, byte[] keyBytes, Object value, byte[] valueBytes, Cluster cluster) {
        List<PartitionInfo> partitions = cluster.partitionsForTopic(topic);
        int numPartitions = partitions.size();
        int nextValue = this.nextValue(topic);
        List<PartitionInfo> availablePartitions = cluster.availablePartitionsForTopic(topic);
        if (!availablePartitions.isEmpty()) {
            int part = Utils.toPositive(nextValue) % availablePartitions.size();
            return ((PartitionInfo)availablePartitions.get(part)).partition();
        } else {
            return Utils.toPositive(nextValue) % numPartitions;
        }
    }

    private int nextValue(String topic) {
        AtomicInteger counter = (AtomicInteger)this.topicCounterMap.computeIfAbsent(topic, (k) -> {
            return new AtomicInteger(0);
        });
        return counter.getAndIncrement();
    }

    public void close() {
    }
}

回答1:

You cannot change the partitioning using ProducerConfig.PARTITIONER_CLASS_CONFIG configuration -- this only works for the plain producer.

In Kafka Streams, you need to implement the interface StreamsPartitioner and pass your implementation into corresponding operators, e.g., to("topic", Produced.streamPartitioner(new MyPartitioner()).

来源：https://stackoverflow.com/questions/59645127/kafka-streams-roundrobinpartitioner