kafka-python 消费者从偏移量开始读取(自动) [英] kafka-python consumer start reading from offset (automatically)

问题描述

我正在尝试使用 kafka-python 构建一个应用程序，其中消费者从一系列主题中读取数据.消费者绝不会两次读取同一条消息，而且绝不会错过任何一条消息，这一点极为重要.

I'm trying to build an application with kafka-python where a consumer reads data from a range of topics. It is extremely important that the consumer never reads the same message twice, but also never misses a message.

似乎一切正常，除非我关闭消费者(例如失败)并尝试从偏移量开始读取.我只能读取主题中的所有消息(这会造成双重读取)或仅收听新消息(并错过在故障期间发出的消息).我在暂停消费者时没有遇到这个问题.

Everything seems to be working fine, except when I turn off the consumer (e.g. failure) and try to start reading from offset. I can only read all the messages from the topic (which creates double reads) or listen for new messages only (and miss messages that where emitted during the breakdown). I don't encounter this problem when pausing the consumer.

我创建了一个孤立的模拟来尝试解决问题.

I created an isolated simulation in order to try to solve the problem.

这里是通用生产者:

from time import sleep
from json import dumps
from kafka import KafkaProducer

producer = KafkaProducer(bootstrap_servers=['localhost:9092'])

x=0 # set manually to avoid duplicates 

for e in range(1000):
    if e <= x:
        pass
    else:
        data = dumps(
            {
            'number' : e
        }
        ).encode('utf-8')

        producer.send('numtest', value=data)
        print(e, ' send.')

        sleep(5)

还有消费者.如果auto_offset_reset 设置为'earliest'，则所有消息将被再次读取.如果 auto_offset_reset 设置为 'latest'，则停机期间不会读取任何消息.

And the consumer. If auto_offset_reset is set to 'earliest', all the messages will be read again. If auto_offset_reset is set to 'latest', no messages during down-time will be read.

from kafka import KafkaConsumer
from pymongo import MongoClient
from json import loads

## Retrieve data from kafka (WHAT ABOUT MISSED MESSAGES?)
consumer = KafkaConsumer('numtest', bootstrap_servers=['localhost:9092'],
                         auto_offset_reset='earliest', enable_auto_commit=True,
                         auto_commit_interval_ms=1000)


## Connect to database
client = MongoClient('localhost:27017')
collection = client.counttest.counttest

# Send data
for message in consumer:
    message = loads(message.value.decode('utf-8'))
    collection.insert_one(message)
    print('{} added to {}'.format(message, collection))

我觉得自动提交功能不正常.

I feel like the auto-commit isn't working properly.

我知道这个问题类似于这个，但我想要一个具体的解决方案.

I know that this questions is similar to this one, but I would like a specific solution.

谢谢你帮助我.

推荐答案

您之所以出现这种行为，是因为您的消费者没有使用消费者组.通过消费者组，消费者将定期向 Kafka 提交(保存)其位置.这样，如果它重新启动，它将从上次提交的位置恢复.

You are getting this behavior because your consumer is not using a Consumer Group. With a Consumer Group, the consumer will regularly commit (save) its position to Kafka. That way if it's restarted it will pick up from its last committed position.

为了让你的消费者使用一个消费者组，你需要在构造它时设置group_id.请参阅 docs 中的 group_id 说明:

To make your consumer use a Consumer Group, you need to set group_id when constructing it. See group_id description from the docs:

动态分区要加入的消费者组名称分配(如果启用)，并用于获取和提交抵消.如果没有，自动分区分配(通过组协调器)和偏移提交被禁用.默认值:无

The name of the consumer group to join for dynamic partition assignment (if enabled), and to use for fetching and committing offsets. If None, auto-partition assignment (via group coordinator) and offset commits are disabled. Default: None

例如:

consumer = KafkaConsumer('numtest', bootstrap_servers=['localhost:9092'],
                         auto_offset_reset='earliest', enable_auto_commit=True,
                         auto_commit_interval_ms=1000, group_id='my-group')

这篇关于kafka-python 消费者从偏移量开始读取(自动)的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！