Python - Kafka：消费者失败

Question

我有一个简单的Producer-Consumer设置：1个生产者（作为一个线程）和2个消费者（作为2个进程）。生产者的运行方法：

    def run(self):
        producer = KafkaProducer(bootstrap_servers=self.bootstrap_servers, 
                                 api_version=(0, 10))
        while not self.stop_event.is_set():
           self.logger.info("Checking for new changes")
           self.check_for_new_changes(producer)
           self.logger.info("Sleeping for {minutes} 
                          minutes...".format(minutes=self.time_to_sleep / 60))
           time.sleep(self.time_to_sleep)
        producer.close()

基本上它会检查更改，如果发现新的更改则发送消息，然后进入睡眠状态5分钟。

运行方法：

def run(self):
    if self.group_id:
        consumer = KafkaConsumer(bootstrap_servers=self.bootstrap_servers,
                                 consumer_timeout_ms=1000,
                                 api_version=(0, 10),
                                 group_id=self.group_id)
    else:
        consumer = KafkaConsumer(bootstrap_servers=self.bootstrap_servers,
                                 consumer_timeout_ms=1000,
                                 api_version=(0, 10))
    consumer.subscribe(['new_change'])
    while not self.stop_event.is_set():
        for msg in consumer:
            self.logger.info("New message:\n{msg}".format(msg=msg))
            self.process_new_change(json.loads(msg.value))
            if self.stop_event.is_set():
                consumer.close()
                return
    consumer.close()

它似乎工作正常，但运行一段时间后，我在协调器日志中收到这些消息：

[2017-12-17 02:06:40,639] INFO [GroupCoordinator 0]: Member kafka-python-1.3.5-f5cdcad3-bc1a-4623-a42b-f5de5e8bded1 in group meta_data_consumer has failed, removing it from the group (kafka.coordinator.group.GroupCoordinator)
[2017-12-17 02:06:40,659] INFO [GroupCoordinator 0]: Preparing to rebalance group meta_data_consumer with old generation 15 (__consumer_offsets-6) (kafka.coordinator.group.GroupCoordinator)
[2017-12-17 02:06:40,659] INFO [GroupCoordinator 0]: Group meta_data_consumer with generation 16 is now empty (__consumer_offsets-6) (kafka.coordinator.group.GroupCoordinator)
[2017-12-17 02:06:41,784] INFO [GroupCoordinator 0]: Member kafka-python-1.3.5-bdea8ce3-922f-4ee1-9959-13341e1730f5 in group failures_consumer has failed, removing it from the group (kafka.coordinator.group.GroupCoordinator)
[2017-12-17 02:06:41,785] INFO [GroupCoordinator 0]: Preparing to rebalance group failures_consumer with old generation 9 (__consumer_offsets-35) (kafka.coordinator.group.GroupCoordinator)
[2017-12-17 02:06:41,785] INFO [GroupCoordinator 0]: Group failures_consumer with generation 10 is now empty (__consumer_offsets-35) (kafka.coordinator.group.GroupCoordinator)

这会杀死我的消费者并停止运行。我没有在消费者日志中看到任何异常或错误。

什么可能导致他们失败？

Answer 1

我认为你的zookeeper.session.timeout.ms低于5分钟。从zookeeper设置调整该超时。看它是否仍然失败。如果是，那么你应该调整kafka配置中的超时。 group.max.session.timeout.ms，rebalance.timeout.ms，heartbeat.interval.ms应相应调整。您的客户端会休眠5分钟，在此期间，超出其中一个超时值，并且Group Coordinator会尝试重新平衡消费者，并认为这些消费者已经失败。

资料来源：Kafka Documentation