使用spring的kafka集群间的事务传输消息

Question

我有两个kafka集群。我需要使用kafka-spring实现它们之间的同步。

[cluster A, topic A]  <-- [spring app] --> [cluster B, topic B]

我创建了注释为@Transactional的侦听器，它使用kafkaTemplate发布消息。当有到两个集群的连接时，这可以完美地工作。当与目标群集的连接丢失时-监听程序似乎仍会确认新消息，但不会发布这些消息。我尝试了手动破解listener，禁用自动提交等，但它们似乎不像我认为的那样工作。当连接重新联机时，消息永远不会被传递。我需要你的帮助。

    @KafkaListener(topics = "A", containerFactory = "syncLocalListenerFactory")
    public void consumeLocal(@Header(KafkaHeaders.RECEIVED_MESSAGE_KEY) String key, @Payload SyncEvent message, Acknowledgment ack) {
        kafkaSyncRemoteTemplate.send("B", key, message);
        ack.acknowledge();
    }

我正在获取日志：

2019-04-26 12:11:40.808  WARN 21304 --- [ad | producer-1] org.apache.kafka.clients.NetworkClient   : [Producer clientId=producer-1] Connection to node 1001 could not be established. Broker may not be available.
2019-04-26 12:11:40.828  WARN 21304 --- [ntainer#0-0-C-1] org.apache.kafka.clients.NetworkClient   : [Consumer clientId=consumer-2, groupId=app-sync] Connection to node 1001 could not be established. Broker may not be available.
2019-04-26 12:11:47.829 ERROR 21304 --- [ad | producer-1] o.s.k.support.LoggingProducerListener    : Exception thrown when sending a message with key='...' and payload='...' to topic B:

org.apache.kafka.common.errors.TimeoutException: Expiring 2 record(s) for sync-2: 30002 ms has passed since batch creation plus linger time

2019-04-26 12:11:47.829 ERROR 21304 --- [ad | producer-1] o.s.k.support.LoggingProducerListener    : Exception thrown when sending a message with key='...' and payload='...' to topic B:

org.apache.kafka.common.errors.TimeoutException: Expiring 2 record(s) for sync-2: 30002 ms has passed since batch creation plus linger time

-编辑

这里的kafkaProperties是从application.properties文件读取的默认kafka-spring属性，但在本例中它们都是默认的

    @Bean
    public ConsumerFactory<String, SyncEvent> syncLocalConsumerFactory() {
        Map<String, Object> config = kafkaProperties.buildConsumerProperties();

        config.put(ConsumerConfig.GROUP_ID_CONFIG, kafkaProperties.getStreams().getApplicationId() + "-sync");
        config.put(ConsumerConfig.KEY_DESERIALIZER_CLASS_CONFIG, StringDeserializer.class);
        config.put(ConsumerConfig.VALUE_DESERIALIZER_CLASS_CONFIG, JsonDeserializer.class);
        config.put(JsonDeserializer.VALUE_DEFAULT_TYPE, SyncEvent.class);
        config.put(JsonDeserializer.TRUSTED_PACKAGES, "app.structures");

        config.put(ConsumerConfig.ENABLE_AUTO_COMMIT_CONFIG, false);

        DefaultKafkaConsumerFactory<String, SyncEvent> cf = new DefaultKafkaConsumerFactory<>(config);
        cf.setValueDeserializer(new JsonDeserializer<>(SyncEvent.class, objectMapper));
        return cf;
    }

    @Bean(name = "syncLocalListenerFactory")
    public ConcurrentKafkaListenerContainerFactory<String, SyncEvent> kafkaSyncLocalListenerContainerFactory() {
        ConcurrentKafkaListenerContainerFactory<String, SyncEvent> factory = new ConcurrentKafkaListenerContainerFactory();
        factory.setConsumerFactory(syncLocalConsumerFactory());
        factory.getContainerProperties().setAckMode(ContainerProperties.AckMode.MANUAL_IMMEDIATE);
        factory.getContainerProperties().setAckOnError(false);
        factory.setErrorHandler(new SeekToCurrentErrorHandler(0));
        return factory;
    }

Answer 1

This website描述了如何设置错误处理程序(使用SeekToCurrentErrorHandler)，它可能会对您有所帮助。从Spring documentation

SeekToCurrentErrorHandler:一个错误处理程序，用于查找剩余记录中每个主题的当前偏移量。用于在消息失败后对分区进行倒带，以便可以重放。

Answer 2

这是因为kafka事务不能跨集群。您的@Transactional注释没有任何意义，因此无论发布到集群B是否成功，偏移量都会提交给集群A。

您目前可以为跨群集流实现的最佳保证是“至少一次”处理，您可以通过确保只有在来自群集B的目标代理确认消息之后才将偏移量提交到群集A来达到它。

有关更多信息，请参阅我的博客帖子- https://medium.com/@harelopler/kafka-cross-cluster-stream-reaching-at-least-once-semantics-c74ed0eb1a54