spring-kafka

spring-kafka

使用spring-kafka的小伙伴,看过来。

说明

因为spring-kafka封装的比较厉害,可能跟你实际使用起来有很大的差别。

一个简单的消费例子

spring-boot基础上添加依赖:

<dependency>
    <groupId>org.springframework.kafka</groupId>
    <artifactId>spring-kafka</artifactId>
    <version>2.2.3.RELEASE</version>
</dependency>

注意要检查下依赖的kakfa-clients,是否与你服务端的匹配。

不要忘记了通过注解@EnableKafka开启自动配置。

采用默认的配置:

spring:
  kafka:
    consumer:
      bootstrap-servers:
        - 127.0.0.1:9092
      # 消费组
      group-id: myGroup
      # 消费者是否自动提交偏移量,默认为true
      enable-auto-commit: false
      # 消费者在读取一个没有偏移量或者偏移量无效的情况下,从起始位置读取partition的记录,默认是latest
      auto-offset-reset: earliest
      # 单次调用poll方法能够返回的消息数量
      max-poll-records: 50

然后写个测试用例试试:

package tk.fishfish.easyjava.kafka;

import org.apache.kafka.clients.consumer.ConsumerRecord;
import org.junit.Test;
import org.junit.runner.RunWith;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import org.springframework.boot.test.context.SpringBootTest;
import org.springframework.kafka.annotation.KafkaListener;
import org.springframework.test.context.junit4.SpringRunner;

/**
 * 消费者
 *
 * @author 奔波儿灞
 * @since 1.0
 */
@SpringBootTest
@RunWith(SpringRunner.class)
public class ConsumerTest {

    private static final Logger LOG = LoggerFactory.getLogger(ConsumerTest.class);

    @KafkaListener(topics = "test")
    public void onMessage(ConsumerRecord<String, String> record) {
        LOG.info("record: {}", record);
        String value = record.value();
        if (value.length() % 2 == 0) {
            throw new RuntimeException("模拟业务出错");
        }
    }

    @Test
    public void run() {
        try {
            // 阻塞5分钟,方便调试
            Thread.sleep(5 * 60 * 1000);
        } catch (InterruptedException e) {
            LOG.warn("sleep error", e);
        }
    }

}

上面通过@KafkaListener来监听topic,处理消息。

为了模拟业务会出现一些异常,我特意在判断value长度为偶数的情况下抛出异常,看在默认配置的情况下,如果业务出错,是否仍会提交offsets

结果发现,仍提交了offsets

spring-kafka-consumer-default

下面是日志:

INFO  2019-05-17 10:25:48.557 [org.springframework.kafka.KafkaListenerEndpointContainer#0-0-C-1] tk.fishfish.easyjava.kafka.ConsumerTest            - record: ConsumerRecord(topic = test, partition = 0, offset = 261, CreateTime = 1558059947687, serialized key size = -1, serialized value size = 0, headers = RecordHeaders(headers = [], isReadOnly = false), key = null, value = )
ERROR 2019-05-17 10:25:48.557 [org.springframework.kafka.KafkaListenerEndpointContainer#0-0-C-1] o.s.kafka.listener.LoggingErrorHandler             - Error while processing: ConsumerRecord(topic = test, partition = 0, offset = 261, CreateTime = 1558059947687, serialized key size = -1, serialized value size = 0, headers = RecordHeaders(headers = [], isReadOnly = false), key = null, value = )
org.springframework.kafka.listener.ListenerExecutionFailedException: Listener method 'public void tk.fishfish.easyjava.kafka.ConsumerTest.onMessage(org.apache.kafka.clients.consumer.ConsumerRecord<java.lang.String, java.lang.String>)' threw exception; nested exception is java.lang.RuntimeException: 模拟业务出错
    at org.springframework.kafka.listener.adapter.MessagingMessageListenerAdapter.invokeHandler(MessagingMessageListenerAdapter.java:302)
    at org.springframework.kafka.listener.adapter.RecordMessagingMessageListenerAdapter.onMessage(RecordMessagingMessageListenerAdapter.java:79)
    at org.springframework.kafka.listener.adapter.RecordMessagingMessageListenerAdapter.onMessage(RecordMessagingMessageListenerAdapter.java:50)
    at org.springframework.kafka.listener.KafkaMessageListenerContainer$ListenerConsumer.doInvokeOnMessage(KafkaMessageListenerContainer.java:1220)
    at org.springframework.kafka.listener.KafkaMessageListenerContainer$ListenerConsumer.invokeOnMessage(KafkaMessageListenerContainer.java:1213)
    at org.springframework.kafka.listener.KafkaMessageListenerContainer$ListenerConsumer.doInvokeRecordListener(KafkaMessageListenerContainer.java:1174)
    at org.springframework.kafka.listener.KafkaMessageListenerContainer$ListenerConsumer.doInvokeWithRecords(KafkaMessageListenerContainer.java:1155)
    at org.springframework.kafka.listener.KafkaMessageListenerContainer$ListenerConsumer.invokeRecordListener(KafkaMessageListenerContainer.java:1096)
    at org.springframework.kafka.listener.KafkaMessageListenerContainer$ListenerConsumer.invokeListener(KafkaMessageListenerContainer.java:924)
    at org.springframework.kafka.listener.KafkaMessageListenerContainer$ListenerConsumer.pollAndInvoke(KafkaMessageListenerContainer.java:740)
    at org.springframework.kafka.listener.KafkaMessageListenerContainer$ListenerConsumer.run(KafkaMessageListenerContainer.java:689)
    at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511)
    at java.util.concurrent.FutureTask.run(FutureTask.java:266)
    at java.lang.Thread.run(Thread.java:748)
Caused by: java.lang.RuntimeException: 模拟业务出错
    at tk.fishfish.easyjava.kafka.ConsumerTest.onMessage(ConsumerTest.java:29)
    at sun.reflect.GeneratedMethodAccessor41.invoke(Unknown Source)
    at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
    at java.lang.reflect.Method.invoke(Method.java:498)
    at org.springframework.messaging.handler.invocation.InvocableHandlerMethod.doInvoke(InvocableHandlerMethod.java:170)
    at org.springframework.messaging.handler.invocation.InvocableHandlerMethod.invoke(InvocableHandlerMethod.java:120)
    at org.springframework.kafka.listener.adapter.HandlerAdapter.invoke(HandlerAdapter.java:48)
    at org.springframework.kafka.listener.adapter.MessagingMessageListenerAdapter.invokeHandler(MessagingMessageListenerAdapter.java:283)
    ... 13 common frames omitted
DEBUG 2019-05-17 10:25:48.557 [org.springframework.kafka.KafkaListenerEndpointContainer#0-0-C-1] o.s.k.l.a.RecordMessagingMessageListenerAdapter    - Processing [GenericMessage [payload=d, headers={kafka_offset=262, kafka_consumer=org.apache.kafka.clients.consumer.KafkaConsumer@f46d581, kafka_timestampType=CREATE_TIME, kafka_receivedMessageKey=null, kafka_receivedPartitionId=0, kafka_receivedTopic=test, kafka_receivedTimestamp=1558059947851}]]
INFO  2019-05-17 10:25:48.557 [org.springframework.kafka.KafkaListenerEndpointContainer#0-0-C-1] tk.fishfish.easyjava.kafka.ConsumerTest            - record: ConsumerRecord(topic = test, partition = 0, offset = 262, CreateTime = 1558059947851, serialized key size = -1, serialized value size = 1, headers = RecordHeaders(headers = [], isReadOnly = false), key = null, value = d)
DEBUG 2019-05-17 10:25:48.557 [org.springframework.kafka.KafkaListenerEndpointContainer#0-0-C-1] o.s.k.l.KafkaMessageListenerContainer$ListenerConsumer - Commit list: {test-0=OffsetAndMetadata{offset=263, metadata=''}}
DEBUG 2019-05-17 10:25:48.557 [org.springframework.kafka.KafkaListenerEndpointContainer#0-0-C-1] o.s.k.l.KafkaMessageListenerContainer$ListenerConsumer - Committing: {test-0=OffsetAndMetadata{offset=263, metadata=''}}
DEBUG 2019-05-17 10:25:48.560 [org.springframework.kafka.KafkaListenerEndpointContainer#0-0-C-1] o.a.k.c.consumer.internals.ConsumerCoordinator     - [Consumer clientId=consumer-2, groupId=myGroup] Committed offset 263 for partition test-0

从日志可以看到offset = 261的记录处理失败了,但最后仍提交了Committed offset 263 for partition test-0

总结:如果你采取我这样的配置,当处理record出错的时候,仍会提交偏移量。那么我们就需要业务处理失败的情况了。比如try...catch之后保存错误的record,然后定时重试。

出错的情况下不提交offsets

那么,能不能在出错的情况下不提交咧?

通过查看文档发现,发现可以使用Acknowledgment去确认该条record是否提交。

修改下配置,配置spring.kafka.listener.*

spring:
  kafka:
    consumer:
      bootstrap-servers:
        - 127.0.0.1:9092
      group-id: myGroup
      # 消费者是否自动提交偏移量,默认为true
      enable-auto-commit: false
      # 消费者在读取一个没有偏移量或者偏移量无效的情况下,从起始位置读取partition的记录,默认是latest
      auto-offset-reset: earliest
      # 单次调用poll方法能够返回的消息数量
      max-poll-records: 50
    listener:
      # Listener AckMode
      ack-mode: MANUAL_IMMEDIATE
      # 并发消费者
      concurrency: 1

然后代码有些调整(就不贴全了),使用Acknowledgment

@KafkaListener(topics = "test")
public void onMessage(ConsumerRecord<String, String> record, Acknowledgment ack) {
    LOG.info("record: {}", record);
    String value = record.value();
    if (value.length() % 2 == 0) {
        throw new RuntimeException("模拟业务出错");
    }
    // 业务处理成功确认
    ack.acknowledge();
}

如果在业务出错的情况下,不会提交offsets,然而真的是这样的吗?

测试发现,在业务出错的情况下,确实不会提交offsets,但是只要后面的记录处理成功,就会提交offsets,这样前面的失败的数据还是需要自己去手动处理。要么重新获取该offset的数据,要么记录错误record,业务重试。

总结

在错误的情况下,建议自己业务记录后,去重试;或者使用spring-kafkaErrorHandler,处理错误的情况。

posted @ 2019-05-17 14:31  bener  阅读(5434)  评论(0编辑  收藏  举报