Python Kafka使用者缺少轮询某些消息

Question

The code for my Kafka consumer looks like this

def read_messages_from_kafka():
    topic = 'my-topic'
    consumer = KafkaConsumer(
        bootstrap_servers=['my-host1', 'my-host2'],
        client_id='my-client',
        group_id='my-group',
        auto_offset_reset='earliest',
        enable_auto_commit=False,
        api_version=(0, 8, 2)
    )
    consumer.assign([TopicPartition(topic, 0), TopicPartition(topic, 1)])

    messages = consumer.poll(timeout_ms=kafka_config.poll_timeout_ms, max_records=kafka_config.poll_max_records)

    for partition in messages.values():
        for message in partition:
            log.info("read {}".format(message))

    if messages:
        consumer.commit()

    next_offset0, next_offset1 = consumer.position(TopicPartition(topic, 0)), consumer.position(TopicPartition(topic, 1))
    log.info("next offset0={} and offset1={}".format(next_offset0, next_offset1))

while True:
    read_messages_from_kafka()
    sleep(kafka_config.poll_sleep_ms / 1000.0)

I have realised that this setup of consumer is not able to read all the messages. And I am not able to reproduce this as it's intermittent issue.

When I compare last 100 messages using kafka-cat to this consumer, I found that my consumer intermittently misses few messages randomly. What's wrong with my consumer?

kafkacat -C -b my-host1 -X broker.version.fallback=0.8.2.1 -t my-topic -o -100

There are just too many ways to consume messages in python . There should be one and preferably only one obvious way to do it.

Answer 1

There is a problem of missing messages in your Kafka client. I found solution here :

while True:
    raw_messages = consumer.poll(timeout_ms=1000, max_records=5000)
    for topic_partition, messages in raw_messages.items():
        application_message = json.loads(message.value.decode())

Also there is another Kafka client exists: confluent_kafka. It has no such problem.