我编写了一个Akka应用程序,它从Kafka获取输入,然后使用分片actor处理数据并输出到Kafka。
但在某些情况下,分片区域无法处理负载,我得到:
您应该实施流量控制以避免泛滥 远程连接。
如何在此链/流中实施背压?
Kafka Consumer - >共享演员 - >卡夫卡制片人
代码中的一些片段:
ReactiveKafka kafka = new ReactiveKafka();
Subscriber subscriber = kafka.publish(pp, system);
ActorRef kafkaWriterActor = (ActorRef) Source.actorRef(10000, OverflowStrategy.dropHead())
.map(ix -> KeyValueProducerMessage.apply(Integer.toString(ix.hashCode()), ix))
.to(Sink.fromSubscriber(subscriber))
.run(materializer);
ConsumerProperties cp = new PropertiesBuilder.Consumer(brokerList, intopic, consumergroup, new ByteArrayDeserializer(), new NgMsgDecoder())
.build().consumerTimeoutMs(5000).commitInterval(Duration.create(60, TimeUnit.SECONDS)).readFromEndOfStream();
Publisher<ConsumerRecord<byte[], StreamEvent>> publisher = kafka.consume(cp,system);
ActorRef streamActor = ClusterSharding.get(system).start("StreamActor",
Props.create(StreamActor.class, synctime), ClusterShardingSettings.create(system), messageExtractor);
shardRegionTypenames.add("StreamActor");
Source.fromPublisher(publisher)
.runWith(Sink.foreach(msg -> {
streamActor.tell(msg.value(),ActorRef.noSender());
}), materializer);
答案 0 :(得分:1)
也许您可以考虑将主题并行化为分区(如果适用),并通过调整this example中的ConsumerWithPerPartitionBackpressure
来使用mapAsync and ask与您的演员集成来创建具有每分区背压的消费者。< / p>