Kafka消费者与Spark-Kafka-Consumer的区别

Question

我有一个Kafka主题，我正在通过Kafka生产者发送数据。现在在消费者方面，我有两个选择。

  @EnableKafka
@Configuration
@PropertySource("kaafka.properties")
public class RawEventKafkaConsumer {

    private static final Logger logger = LoggerFactory.getLogger(RawEventKafkaConsumer.class);
    @Autowired
    private DataModelServiceImpl dataModelServiceImpl;

    private PolicyExecutor policyExecutor;

    public RawEventKafkaConsumer() {
         policyExecutor = new PolicyExecutor();
    }


    @Value("${spring.kafka.topic}")
    private String rawEventTopicName;

    @Value("${spring.kafka.consumer.auto-offset-reset}")
    private String autoOffsetReset;

    @Value("${spring.kafka.consumer.bootstrap-servers}")
    private String bootStrapServer;

    @Value("${spring.kafka.consumer.group-id}")
    private String groupId;

    @Value("${spring.kafka.producer.key-serializer}")
    private String keySerializer;

    @Value("${spring.kafka.producer.value-serializer}")
    private String valueSerializer;

    @Value("${spring.kafka.consumer.key-deserializer}")
    private String keyDeserializer;

    @Value("${spring.kafka.consumer.value-deserializer}")
    private String valueDeserializer;



    @Bean
    public DefaultKafkaConsumerFactory<String, BaseDataModel> rawEventConsumer() {
        Map<String, Object> consumerProperties = new HashMap<>();
        consumerProperties.put(ConsumerConfig.BOOTSTRAP_SERVERS_CONFIG, bootStrapServer);
        consumerProperties.put(ConsumerConfig.KEY_DESERIALIZER_CLASS_CONFIG, keyDeserializer);
        consumerProperties.put(ConsumerConfig.VALUE_DESERIALIZER_CLASS_CONFIG, valueDeserializer);
        consumerProperties.put(ConsumerConfig.GROUP_ID_CONFIG, "group1");
        consumerProperties.put(ConsumerConfig.AUTO_OFFSET_RESET_CONFIG, autoOffsetReset);
        consumerProperties.put(ConsumerConfig.ENABLE_AUTO_COMMIT_CONFIG, true);
        return new DefaultKafkaConsumerFactory<>(consumerProperties);
    }

    @Bean(name="kafkaListenerContainerFactory")
    public ConcurrentKafkaListenerContainerFactory<String, BaseDataModel> kafkaListenerContainerFactory() {
         logger.info("kafkaListenerContainerFactory called..");
        ConcurrentKafkaListenerContainerFactory<String, BaseDataModel> factory = new ConcurrentKafkaListenerContainerFactory<>();
        factory.setConsumerFactory(rawEventConsumer());
        return factory;
    }

    @KafkaListener(topics = "rawEventTopic",  containerFactory = "kafkaListenerContainerFactory")
    public void listen(String baseDataModel) {

        ObjectMapper mapper = new ObjectMapper();
        BaseDataModel csvDataModel;
        try {
            csvDataModel = mapper.readValue(baseDataModel, BaseDataModel.class);

            //saving the datamodel in elastic search.
            //dataModelServiceImpl.save(csvDataModel);
            System.out.println("Message received " + csvDataModel.toString());
        } catch (Exception e) {
            // TODO Auto-generated catch block
            e.printStackTrace();
        }

    }

}

 @Service
    public class RawEventSparkStreamConsumer {

        private final Logger logger = LoggerFactory.getLogger(RawEventSparkStreamConsumer.class);

        @Autowired
        private DataModelServiceImpl dataModelServiceImpl;


        @Autowired
        private JavaStreamingContext streamingContext;

        @Autowired
        private JavaInputDStream<ConsumerRecord<String, String>> messages;


        @PostConstruct
        private void sparkRawEventConsumer() {

            ExecutorService executor = Executors.newSingleThreadExecutor();
            executor.execute(()->{
                messages.foreachRDD((rdd) -> {
                    System.out.println("RDD coming *************************______________________________---------------------.." + rdd.count());
                    rdd.foreach(record -> {
                        System.out.println("Data is comming...." + record);
                    });
                });

                streamingContext.start();

                try {
                    streamingContext.awaitTermination();
                } catch (InterruptedException e) { // TODO Auto-generated catch block
                    e.printStackTrace();
                }
            });

        }
    }

null

Answer 1

简单的回答是，您需要一个Spark集群以分布式方式运行Spark代码，而Kafka使用者只在一个JVM中运行，您手动运行同一个应用程序的多个实例来扩展它。

换句话说，您将以不同的方式运行它们。spark-submitvsjava-jar。我不相信使用Spring会改变

另一个区别是“普通消费者”对Kafka配置有更多的控制，您一次只获得一个记录。Spark RDD可以是许多事件，并且它们必须都属于相同的“模式”，除非您需要复杂的解析逻辑，否则使用RDD对象编写复杂的解析逻辑比使用为您提取的ConsumerRecord值更难。

总的来说，我认为把它们结合起来不是个好主意。

如果他们读的是相同的主题，那么Kafka使用者协议只能为每个分区分配一个使用者...不清楚您的主题有多少个分区，但这可以解释为什么一个可以工作，而另一个不行