Question

我正在使用以下示例来使用Spring Kafka使用者读取消息。我的用例要求每次产生一条消息时，侦听器都会从头开始每次读取。

@KafkaListener(
    id = "grouplistener",
    topicPartitions = { 
        @TopicPartition(
            topic = "mycompactedtopic", partitionOffsets = @PartitionOffset(partition = "0", initialOffset = "0")
        )
    }
)

public void onReceiving(
    String payload, @Header(KafkaHeaders.OFFSET) Integer offset,
    @Header(KafkaHeaders.RECEIVED_PARTITION_ID) int partition,
    @Header(KafkaHeaders.RECEIVED_TOPIC) String topic
) {
    log.info(
        "Processing topic = {}, partition = {}, offset = {}, payload= {}",
        topic, partition, offset, payload
    );
}

我似乎只能从应用程序启动时就开始读取它，然后通常只消耗向前发送的消息。

是否有一种方法可以迫使它每次尝试开始？

Answer 1

使用具有1个分区的压缩主题来保存配置列表。然后需要由其他端点调用，并且应该显示配置的完整唯一列表

您应该执行此操作的方法是使用Kafka Streams和KTable，并在REST层后面设置interactive queries。不是需要倒带以获取系统的最新状态的标准使用者。

Kafka Connect框架中已有一个配置示例，其中有一个配置主题，您只能访问GET /connectors/name/config的最新值，并且只有重新启动它或扩展到更多实例时，它将再次消耗所有消息。架构注册表也是这样的一个示例，它在_schemas主题中存储了所有架构的内部Hashmap，并且具有用于读取，插入，删除

的REST API。

基本上，当您为给定密钥获得新配置时，可以用一个全新的密钥“替换”给定密钥的旧值，也可以在新数据中“合并”旧值。某种方式。

Answer 2

我认为您应该尝试编写ConsumerSeekAwareListener，并在每次阅读消息时都尝试将偏移量设置为0。听起来像是疯狂的解决方法，但这可能会有所帮助。希望这会对您有所帮助：-）

class Listener implements ConsumerSeekAware {

 private final ThreadLocal<ConsumerSeekCallback> seekCallBack = new ThreadLocal<>();

   ----Override all methods that are needed----

@KafkaListener(...)
    public void listen(@Payload String message) {

            this.seekCallBack.get().seek(topic, partition, 0);
        }
    }
}

Answer 3

@ Nimo1981这是使用纯Java的实现。我不确定它是否满足您的需求。因此，基本上，我将偏移量设为0，（意味着，即使我从Kafka主题阅读，我也会回到开始时的偏移量。）我不确定您是否考虑过此实现，但是请告诉我是您要寻找的东西

忽略CommitCountObj。您的那不需要。因此，默认情况下， offsetMap 将具有下一个这样的偏移记录

offsetMap.put（new TopicPartition（record.topic（），record.partition（）），新的OffsetAndMetadata（record.offset（）+ 1，“一些提交成功消息”））;

但是对于您的用例，我进行了某种修改，在不重新启动使用者时它可以很好地工作

offsetMap.put（new TopicPartition（record.topic（），record.partition（）），新的OffsetAndMetadata（0，“未完成提交”））;

public class KafkaConsumerClass {

    private static final org.apache.log4j.Logger log = org.apache.log4j.Logger.getLogger(KafkaConsumerClass.class);
    private CommitCountClass commitCountobj = new CommitCountClass();

    public Consumer<String, List<FeedBackConsumerClass>> createConsumer() {
        Map<String, Object> consumerProps = new HashMap<String, Object>();
        consumerProps.put(ConsumerConfig.BOOTSTRAP_SERVERS_CONFIG, "localhost:7070,localhost:7072");
        consumerProps.put(ConsumerConfig.CONNECTIONS_MAX_IDLE_MS_CONFIG, 50000);
        consumerProps.put(ConsumerConfig.CLIENT_ID_CONFIG, "first_group-client1");
        // consumerProps.put(ConsumerConfig.ENABLE_AUTO_COMMIT_CONFIG, true);
        consumerProps.put(ConsumerConfig.GROUP_ID_CONFIG, "first_group");
        // consumerProps.put(ConsumerConfig.INTERCEPTOR_CLASSES_CONFIG, KafkaConsumerInterceptor.class);
        consumerProps.put(ConsumerConfig.KEY_DESERIALIZER_CLASS_CONFIG, StringDeserializer.class);
        consumerProps.put(ConsumerConfig.VALUE_DESERIALIZER_CLASS_CONFIG, JsonDeserializer.class);
        consumerProps.put(ConsumerConfig.SESSION_TIMEOUT_MS_CONFIG, 15000);
        consumerProps.put(ConsumerConfig.ENABLE_AUTO_COMMIT_CONFIG, false);
        consumerProps.put(ConsumerConfig.AUTO_COMMIT_INTERVAL_MS_CONFIG, 1500);
        consumerProps.put(ConsumerConfig.AUTO_OFFSET_RESET_CONFIG, "earliest");

        return new KafkaConsumer<String, List<FeedBackConsumerClass>>(consumerProps);
    }

    public void consumeRecord() {
        log.info("Coming inside consumer consumer");
        ArrayList<String> topicList = new ArrayList<String>();
        topicList.add("topic1");
        commitCountobj.setCount(0);
        Consumer<String, List<FeedBackConsumerClass>> kafkaConsumer = createConsumer();
        kafkaConsumer.subscribe(topicList);
        log.info("after subscribing");

        Map<TopicPartition, OffsetAndMetadata> offsetMap = new HashMap<>();

        while (true) {

            ConsumerRecords<String, List<FeedBackConsumerClass>> recordList = kafkaConsumer.poll(Long.MAX_VALUE);
            // kafkaConsumer.seekToBeginning(kafkaConsumer.assignment());

            log.info("Inside while loop:" + recordList);
            if (!recordList.isEmpty()) {
                recordList.forEach(record -> {
                    int i = 0;
                    System.out.println(record.toString());
                    // we can make the call to the API here
                    // call the db here or any API and process the record
                    // then call the code to commit
                    // since the commit is switched off, it becomes a developers responsibility to do the auto commit
                    offsetMap.put(new TopicPartition(record.topic(), record.partition()),
                            new OffsetAndMetadata(0, "no metadata/offset commited"));
                    // here we are incrementing the offsetMap so that we are making sure we are storing the
                    // next set of offsets in the map
                    if (commitCountobj.getCount() % 1000 == 0) {
                        kafkaConsumer.commitAsync(offsetMap, new OffsetCommitCallback() {

                            @Override
                            public void onComplete(Map<TopicPartition, OffsetAndMetadata> offsets,
                                    Exception exception) {
                                // TODO Auto-generated method stub
                                if (exception != null) {
                                    // retry it now with a sync
                                    // possibility of error occuring here as well
                                    // so capture the exception and exit the consumer gracefully
                                    kafkaConsumer.commitSync();
                                    log.error(exception.getMessage());
                                }
                            }
                        });
                    }
                    commitCountobj.setCount(i++);
                });
            }

        }
    }

}

Answer 4

这是我将如何实施。您需要实现ConsumerSeekAware接口并在onPartitionsAssigned方法上进行一些实现。如果在重新启动应用程序时发送环境变量，也可以按需使seekToBegining。我还没有实现！

@Service
@EnableKafka
public class Service implements ConsumerSeekAware {



    @KafkaListener(topics = "${topicName}", groupId = "${groupId}")
    public void listen(@Header(KafkaHeaders.RECEIVED_PARTITION_ID) int partition,
                       @Header(KafkaHeaders.RECEIVED_TIMESTAMP) long ts,
                       @Payload List<String> messageBatch
    ) {
            //do a bunch of stuff
    }



    @Override
    public void onPartitionsAssigned(Map<TopicPartition, Long> assignments, ConsumerSeekCallback callback) {
        String topic= Optional.ofNullable(System.getProperty(TOPIC_NAME)).orElseThrow(()->new RuntimeException("topicName needs to be set"));
        assignments.keySet().stream().filter(partition->topic.equals(partition.topic()))
                .forEach(partition -> callback.seekToBeginning(topic, partition.partition()));
    }

    @Override
    public void onIdleContainer(Map<TopicPartition, Long> assignments, ConsumerSeekCallback callback) {}

    @Override
    public void registerSeekCallback(ConsumerSeekCallback callback) {}
}

Answer 5

@KafkaListener(topicPartitions 
          = @TopicPartition(topic = "test", partitionOffsets = {
          @PartitionOffset(partition = "0", initialOffset = "0")}),groupId = "foo",
        containerFactory = "kafkaListenerContainerFactory")
public void listenAllMsg(@Payload String message,@Header(KafkaHeaders.RECEIVED_PARTITION_ID) int partition) {
    System.out.println(" all msg Received Messasge in group 'foo': " + message+"RECEIVED_PARTITION_ID - "+partition);

}

在kafka 2.3.1中

@KafkaListener每次都从头开始阅读

5 个答案: