Poll several messages from Kafka

Question

I'm using confluent_kafka package for working with Kafka. I create topic in this way:

from confluent_kafka import avro
from confluent_kafka.avro import AvroProducer

def my_producer():
    bootstrap_servers=['my_adress.com:9092',
                    'my_adress.com:9092']

    value_schema = avro.load('/home/ValueSchema.avsc')

    avroProducer = AvroProducer({
        'bootstrap.servers': bootstrap_servers[0]+','+bootstrap_servers[1],
        'schema.registry.url':'http://my_adress.com:8081',
        },
        default_value_schema=value_schema
        )

    for i in range(0, 25000):
        value = {"name":"Yuva","favorite_number":10,"favorite_color":"green","age":i*2}
        avroProducer.produce(topic='my_topik14', value=value)
        avroProducer.flush(0)
    print('Finished!')


if __name__ == '__main__':
    my_producer()

It works. (this get 24820 messages instead of 25000 by the way...) We can check it:

kafka-run-class kafka.tools.GetOffsetShell --broker-list my_adress.com:9092 --topic my_topik14
my_topik14:0:24819

Now I want to consume:

from confluent_kafka import KafkaError
from confluent_kafka.avro import AvroConsumer
from confluent_kafka.avro.serializer import SerializerError

bootstrap_servers=['my_adress.com:9092',
                   'my_adress.com:9092']
c = AvroConsumer(
    {'bootstrap.servers': bootstrap_servers[0]+','+bootstrap_servers[1],
     'group.id': 'avroneversleeps',
     'schema.registry.url': 'http://my_adress.com:8081',
     'api.version.request': True,
     'fetch.min.bytes': 100000,
     'consume.callback.max.messages':1000,
     'batch.num.messages':2
     })
c.subscribe(['my_topik14'])
running = True

while running:
    msg = None
    try:
        msg = c.poll(0.1)
        if msg:
            if not msg.error():
                print(msg.value())
                c.commit(msg)
            elif msg.error().code() != KafkaError._PARTITION_EOF:
                print(msg.error())
                running = False
        else:
            print("No Message!! Happily trying again!!")
    except SerializerError as e:
        print("Message deserialization failed for %s: %s" % (msg, e))
        running = False
c.commit()
c.close()

But there is a problem: I read messages just one by one. My question is How to read batch of messages? I tried different parameters in Consumer config but they didn't cnahge anything!

---

Also I found this question on SO and tried the same parameters - it still doesn't work.

Also read this. But this is against the previous link...

Answer 1

AvroConsumer have no consume method. But it is easy to make my own implementation of this method as there is in Consume class (parent of AvroConsumer). Here is the code:

def consume_batch(self, num_messages=1, timeout=None):
    """
    This is an overriden method from confluent_kafka.Consumer class. This handles batch of message
    deserialization using avro schema

    :param int num_messages: number of messages to read in one batch (default=1)
    :param float timeout: Poll timeout in seconds (default: indefinite)
    :returns: list of messages objects with deserialized key and value as dict objects
    :rtype: Message
    """
    messages_out = []
    if timeout is None:
        timeout = -1
    messages = super(AvroConsumer, self).consume(num_messages=num_messages, timeout=timeout)
    if messages is None:
        return None
    else:
        for m in messages:
            if not m.value() and not m.key():
                return messages
            if not m.error():
                if m.value() is not None:
                    decoded_value = self._serializer.decode_message(m.value())
                    m.set_value(decoded_value)
                if m.key() is not None:
                    decoded_key = self._serializer.decode_message(m.key())
                    m.set_key(decoded_key)
                messages_out.append(m)
    #print(len(message))
    return messages_out

But after that we run test and this method give no any performance increasing. So looks like it just for better usability. Or I need to make some additional work about serializing not single message, but whole batch.