即使我暂停消费者，Spring Kafka 也总是在 5 分钟后重新平衡

Posted 2023-02-26

【中文标题】即使我暂停消费者，Spring Kafka 也总是在 5 分钟后重新平衡

【英文标题】：Spring Kafka Always rebalance after 5 min even i pause consumer

【发布时间】：2019-04-16 23:22:28

【问题描述】：

有一个耗时的操作（大约 10 分钟），但 kafka 会在 5 分钟后重新平衡，即使我暂停了消费者。消费者方法：

    @KafkaListener(topics = TopicAppoint.EXECUTE_SCHOOL_DATA_STATICS_TASK)
public void receiveMessage(@Payload String payload, Consumer<String, String> consumer) 

    Set<TopicPartition> assignment = consumer.assignment();
    consumer.pause(assignment);

    if (StringUtils.isNotEmpty(payload)) 
        SchoolStatisticsTaskDTO staticsTaskDTO = JSONObject.parseObject(payload, SchoolStatisticsTaskDTO.class);
        Optional<SchoolStatisticsTaskDO> taskOptional = schoolStatisticsTaskRepository.findById(staticsTaskDTO.getTrackId());
        taskOptional.ifPresent(schoolStaticsTaskDO -> 
            // handler
        );
    

    consumer.resume(assignment);

这是我的配置：

  kafka:
bootstrap-servers: 192.168.0.230:9092
producer:
  key-serializer: org.apache.kafka.common.serialization.StringSerializer
  value-serializer: org.apache.kafka.common.serialization.StringSerializer
  retries: 3
  properties:
    max.request.size: 12582912
consumer:
  key-deserializer: org.apache.kafka.common.serialization.StringDeserializer
  value-deserializer: org.apache.kafka.common.serialization.StringDeserializer
  group-id: dc-fitness-data-consumer-group
  properties:
    max.partition.fetch.bytes: 12582912
  #enable-auto-commit: false
listener:
  ack-mode: record
  concurrency: 6

日志

  13:09:20.219 [org.springframework.kafka.KafkaListenerEndpointContainer#2-0-C-1] INFO  o.a.k.c.c.i.AbstractCoordinator - [Consumer clientId=consumer-25, groupId=dc-fitness-data-consumer-group] Attempt to heartbeat failed since group is rebalancing
   13:09:20.219 [org.springframework.kafka.KafkaListenerEndpointContainer#2-0-C-1] INFO  o.a.k.c.c.i.ConsumerCoordinator - [Consumer clientId=consumer-25, groupId=dc-fitness-data-consumer-group] Revoking previously assigned partitions [ft.es.records.incremental.update.task-4, ft.es.records.incremental.update.task-5, ft.es.records.incremental.update.task-2, ft.es.records.incremental.update.task-3, ft.es.records.incremental.update.task-0, ft.es.records.incremental.update.task-1]
   13:09:20.219 [org.springframework.kafka.KafkaListenerEndpointContainer#2-0-C-1] INFO  o.s.k.l.KafkaMessageListenerContainer - partitions revoked: [ft.es.records.incremental.update.task-4, ft.es.records.incremental.update.task-5, ft.es.records.incremental.update.task-2, ft.es.records.incremental.update.task-3, ft.es.records.incremental.update.task-0, ft.es.records.incremental.update.task-1]
  13:09:20.219 [org.springframework.kafka.KafkaListenerEndpointContainer#2-0-C-1] INFO  o.a.k.c.c.i.AbstractCoordinator - [Consumer clientId=consumer-25, groupId=dc-fitness-data-consumer-group] (Re-)joining group
  13:09:20.220 [org.springframework.kafka.KafkaListenerEndpointContainer#9-2-C-1] INFO  o.a.k.c.c.i.AbstractCoordinator - [Consumer clientId=consumer-21, groupId=dc-fitness-data-consumer-group] Attempt to heartbeat failed since group is rebalancing
  13:09:20.221 [org.springframework.kafka.KafkaListenerEndpointContainer#9-2-C-1] INFO  o.a.k.c.c.i.ConsumerCoordinator - [Consumer clientId=consumer-21, groupId=dc-fitness-data-consumer-group] Revoking previously assigned partitions [ft.student.batch.upload.task-17, ft.student.batch.upload.task-14, ft.student.batch.upload.task-13, ft.student.batch.upload.task-16, ft.student.batch.upload.task-15, ft.student.batch.upload.task-12]
  13:09:20.221 [org.springframework.kafka.KafkaListenerEndpointContainer#9-2-C-1] INFO  o.s.k.l.KafkaMessageListenerContainer - partitions revoked: [ft.student.batch.upload.task-17, ft.student.batch.upload.task-14, ft.student.batch.upload.task-13, ft.student.batch.upload.task-16, ft.student.batch.upload.task-15, ft.student.batch.upload.task-12]
  13:09:20.221 [org.springframework.kafka.KafkaListenerEndpointContainer#9-2-C-1] INFO  o.a.k.c.c.i.AbstractCoordinator - [Consumer clientId=consumer-21, groupId=dc-fitness-data-consumer-group] (Re-)joining group
  13:09:20.221 [org.springframework.kafka.KafkaListenerEndpointContainer#6-3-C-1] INFO  o.a.k.c.c.i.AbstractCoordinator - [Consumer clientId=consumer-4, groupId=dc-fitness-data-consumer-group] Attempt to heartbeat failed since group is rebalancing
  13:09:20.221 [org.springframework.kafka.KafkaListenerEndpointContainer#8-5-C-1] INFO  o.a.k.c.c.i.AbstractCoordinator - [Consumer clientId=consumer-18, groupId=dc-fitness-data-consumer-group] Attempt to heartbeat failed since group is rebalancing
  13:09:20.221 [org.springframework.kafka.KafkaListenerEndpointContainer#8-5-C-1] INFO  o.a.k.c.c.i.ConsumerCoordinator - [Consumer clientId=consumer-18, groupId=dc-fitness-data-consumer-group] Revoking previously assigned partitions [ft.record.batch.upload.task-32, ft.record.batch.upload.task-31, ft.record.batch.upload.task-34, ft.record.batch.upload.task-33, ft.record.batch.upload.task-30, ft.record.batch.upload.task-35]
  13:09:20.221 [org.springframework.kafka.KafkaListenerEndpointContainer#8-5-C-1] INFO  o.s.k.l.KafkaMessageListenerContainer - partitions revoked: [ft.record.batch.upload.task-32, ft.record.batch.upload.task-31, ft.record.batch.upload.task-34, ft.record.batch.upload.task-33, ft.record.batch.upload.task-30, ft.record.batch.upload.task-35]
  13:09:20.221 [org.springframework.kafka.KafkaListenerEndpointContainer#8-5-C-1] INFO  o.a.k.c.c.i.AbstractCoordinator - [Consumer clientId=consumer-18, groupId=dc-fitness-data-consumer-group] (Re-)joining group
  13:09:20.221 [org.springframework.kafka.KafkaListenerEndpointContainer#6-3-C-1] INFO  o.a.k.c.c.i.ConsumerCoordinator - [Consumer clientId=consumer-4, groupId=dc-fitness-data-consumer-group] Revoking previously assigned partitions [ft.class.batch.upload.task-18, ft.class.batch.upload.task-19, ft.class.batch.upload.task-20, ft.class.batch.upload.task-21, ft.class.batch.upload.task-22, ft.class.batch.upload.task-23]
  13:09:20.221 [org.springframework.kafka.KafkaListenerEndpointContainer#6-4-C-1] INFO  o.a.k.c.c.i.AbstractCoordinator - [Consumer clientId=consumer-5, groupId=dc-fitness-data-consumer-group] Attempt to heartbeat failed since group is rebalancing
  13:09:20.221 [org.springframework.kafka.KafkaListenerEndpointContainer#0-4-C-1] INFO  o.a.k.c.c.i.AbstractCoordinator - [Consumer clientId=consumer-53, groupId=dc-fitness-data-consumer-group] Attempt to heartbeat failed since group is rebalancing
  13:09:20.221 [org.springframework.kafka.KafkaListenerEndpointContainer#0-3-C-1] INFO  o.a.k.c.c.i.AbstractCoordinator - [Consumer clientId=consumer-52, groupId=dc-fitness-data-consumer-group] Attempt to heartbeat failed since group is rebalancing
  13:09:20.223 [org.springframework.kafka.KafkaListenerEndpointContainer#6-4-C-1] INFO  o.a.k.c.c.i.ConsumerCoordinator - [Consumer clientId=consumer-5, groupId=dc-fitness-data-consumer-group] Revoking previously assigned partitions [ft.class.batch.upload.task-26, ft.class.batch.upload.task-27, ft.class.batch.upload.task-28, ft.class.batch.upload.task-29, ft.class.batch.upload.task-24, ft.class.batch.upload.task-25]
  13:09:20.223 [org.springframework.kafka.KafkaListenerEndpointContainer#6-3-C-1] INFO  o.s.k.l.KafkaMessageListenerContainer - partitions revoked: [ft.class.batch.upload.task-18, ft.class.batch.upload.task-19, ft.class.batch.upload.task-20, ft.class.batch.upload.task-21, ft.class.batch.upload.task-22, ft.class.batch.upload.task-23]
  13:09:20.223 [org.springframework.kafka.KafkaListenerEndpointContainer#0-4-C-1] INFO  o.a.k.c.c.i.ConsumerCoordinator - [Consumer clientId=consumer-53, groupId=dc-fitness-data-consumer-group] Revoking previously assigned partitions [ft.es.class.incremental.update.task-26, ft.es.class.incremental.update.task-27, ft.es.class.incremental.update.task-28, ft.es.class.incremental.update.task-29, ft.es.class.incremental.update.task-24, ft.es.class.incremental.update.task-25]
  13:09:20.223 [org.springframework.kafka.KafkaListenerEndpointContainer#6-4-C-1] INFO  o.s.k.l.KafkaMessageListenerContainer - partitions revoked: [ft.class.batch.upload.task-26, ft.class.batch.upload.task-27, ft.class.batch.upload.task-28, ft.class.batch.upload.task-29, ft.class.batch.upload.task-24, ft.class.batch.upload.task-25]
  13:09:20.223 [org.springframework.kafka.KafkaListenerEndpointContainer#0-4-C-1] INFO  o.s.k.l.KafkaMessageListenerContainer - partitions revoked: [ft.es.class.incremental.update.task-26, ft.es.class.incremental.update.task-27, ft.es.class.incremental.update.task-28, ft.es.class.incremental.update.task-29, ft.es.class.incremental.update.task-24, ft.es.class.incremental.update.task-25]
  13:09:20.223 [org.springframework.kafka.KafkaListenerEndpointContainer#6-4-C-1] INFO  o.a.k.c.c.i.AbstractCoordinator - [Consumer clientId=consumer-5, groupId=dc-fitness-data-consumer-group] (Re-)joining group
  13:09:20.223 [org.springframework.kafka.KafkaListenerEndpointContainer#6-3-C-1] INFO  o.a.k.c.c.i.AbstractCoordinator - [Consumer clientId=consumer-4, groupId=dc-fitness-data-consumer-group] (Re-)joining group
  13:09:20.223 [org.springframework.kafka.KafkaListenerEndpointContainer#0-4-C-1] INFO  o.a.k.c.c.i.AbstractCoordinator - [Consumer clientId=consumer-53, groupId=dc-fitness-data-consumer-group] (Re-)joining group
  13:09:20.223 [org.springframework.kafka.KafkaListenerEndpointContainer#0-3-C-1] INFO  o.a.k.c.c.i.ConsumerCoordinator - [Consumer clientId=consumer-52, groupId=dc-fitness-data-consumer-group] Revoking previously assigned partitions [ft.es.class.incremental.update.task-22, ft.es.class.incremental.update.task-23, ft.es.class.incremental.update.task-18, ft.es.class.incremental.update.task-19, ft.es.class.incremental.update.task-20, ft.es.class.incremental.update.task-21]
  13:09:20.223 [org.springframework.kafka.KafkaListenerEndpointContainer#0-3-C-1] INFO  o.s.k.l.KafkaMessageListenerContainer - partitions revoked: [ft.es.class.incremental.update.task-22, ft.es.class.incremental.update.task-23, ft.es.class.incremental.update.task-18, ft.es.class.incremental.update.task-19, ft.es.class.incremental.update.task-20, ft.es.class.incremental.update.task-21]
  13:09:20.223 [org.springframework.kafka.KafkaListenerEndpointContainer#5-4-C-1] INFO  o.a.k.c.c.i.AbstractCoordinator - [Consumer clientId=consumer-47, groupId=dc-fitness-data-consumer-group] Attempt to heartbeat failed since group is rebalancing
  13:09:20.225 [org.springframework.kafka.KafkaListenerEndpointContainer#0-3-C-1] INFO  o.a.k.c.c.i.AbstractCoordinator - [Consumer clientId=consumer-52, groupId=dc-fitness-data-consumer-group] (Re-)joining group

【问题讨论】：

spring kafka 的哪个版本？

kafka 版本：2.12

你能发布重新平衡日志吗

似乎没有太多有用的日志

【参考方案1】：

让我们看一下文档，从pause()方法开始

public void pause（集合分区） here

暂停从请求的分区中提取。未来对 poll(long) 的调用将不会从这些分区返回任何记录，直到使用 resume(Collection) 恢复它们。请注意，此方法不影响分区订阅。特别是在使用自动分配时，它不会导致组重新平衡。

所以从上面的描述来看，pause() 方法将暂停分区获取消息但不会暂停消费者线程，这意味着消费者线程将执行后续的poll() 请求到暂停的分区但不会获取任何记录

在您的应用程序中： 分区已暂停，但消费者线程忙于执行耗时操作超过 10 分钟而没有执行任何轮询请求。

检测消费者故障：所以当轮询停止时，心跳不会发送到集群

订阅一组主题后，调用 poll(long) 时，消费者会自动加入群组。 poll API 旨在确保消费者的活跃度。只要您继续调用 poll，消费者就会留在组中并继续从分配给它的分区接收消息。在幕后，poll API 会定期向服务器发送心跳； 当你停止调用 poll 时（可能是因为抛出了异常），则不会发送任何心跳。 如果在服务器收到心跳之前经过了配置的会话超时时间，那么消费者将被踢出组，其分区将被重新分配。

解决方案：增加以下属性的超时时间，因为默认时间是 5 分钟，您会看到每 5 分钟重新平衡一次（个人不支持增加超时时间）here

新的 Java Consumer 现在支持来自后台线程的心跳。有一个新配置 max.poll.interval.ms 控制在消费者主动离开组之前轮询调用之间的最长时间（默认为 5 分钟）。配置request.timeout.ms 的值必须始终大于 max.poll.interval.ms 因为这是消费者重新平衡时 JoinGroup 请求可以在服务器上阻塞的最长时间，因此我们将其默认值更改为 just 5分钟以上。最后将session.timeout.ms的默认值调整为10秒，max.poll.records的默认值改为500。

解决方案2：如果你需要用更少的时间处理大量数据，增加更多的分区，每个线程消耗每个分区（这意味着更多的并发），5 内可以处理的数据更少分钟

【讨论】：

非常感谢，我通过调整max.poll.interval.ms属性解决了这个问题。