一、简介
python连接kafka的标准库，kafka-python和pykafka。kafka-python使用的人多是比较成熟的库，kafka-python并没有zk的支持。pykafka是Samsa的升级版本，使用samsa连接zookeeper，生产者直接连接kafka服务器列表，消费者才用zookeeper。使用kafka Cluster。

二、pykafka
(1) pykafka安装
根据机器环境从以下三种方式中选择进行一种安装pykafka，版本号是2.7.0。

PyPI安装

pip install pykafka

conda安装

conda install -c conda-forge pykafka

anaconda自带pip安装

/root/anaconda3/bin/pip install pykafka
(2) pykafka的api

     1、http://pykafka.readthedocs.io/en/latest/，https://github.com/Parsely/pykafka
     2、在pykafka安装目录site-packages/pykafka/下，直接查看。

(3) pykafka生产者api

#coding=utf-8
 
import time
from pykafka import KafkaClient
 
 
class KafkaTest(object):
    """
    测试kafka常用api
    """
    def __init__(self, host="192.168.237.129:9092"):
        self.host = host
        self.client = KafkaClient(hosts=self.host)
 
    def producer_partition(self, topic):
        """
        生产者分区查看，主要查看生产消息时offset的变化
        :return:
        """
        topic = self.client.topics[topic.encode()]
        partitions = topic.partitions
        print (u"查看所有分区 {}".format(partitions))
 
        earliest_offset = topic.earliest_available_offsets()
        print(u"获取最早可用的offset {}".format(earliest_offset))
 
        # 生产消息之前看看offset
        last_offset = topic.latest_available_offsets()
        print(u"最近可用offset {}".format(last_offset))
 
        # 同步生产消息
        p = topic.get_producer(sync=True)
        p.produce(str(time.time()).encode())
 
        # 查看offset的变化
        last_offset = topic.latest_available_offsets()
        print(u"最近可用offset {}".format(last_offset))
 
    def producer_designated_partition(self, topic):
        """
        往指定分区写消息，如果要控制打印到某个分区，
        需要在获取生产者的时候指定选区函数，
        并且在生产消息的时候额外指定一个key
        :return:
        """
 
        def assign_patition(pid, key):
            """
            指定特定分区, 这里测试写入第一个分区(id=0)
            :param pid: 为分区列表
            :param key:
            :return:
            """
            print("为消息分配partition {} {}".format(pid, key))
            return pid[0]
 
        topic = self.client.topics[topic.encode()]
        p = topic.get_producer(sync=True, partitioner=assign_patition)
        p.produce(str(time.time()).encode(), partition_key=b"partition_key_0")
 
    def async_produce_message(self, topic):
        """
        异步生产消息，消息会被推到一个队列里面，
        另外一个线程会在队列中消息大小满足一个阈值（min_queued_messages）
        或到达一段时间（linger_ms）后统一发送,默认5s
        :return:
        """
        topic = self.client.topics[topic.encode()]
        last_offset = topic.latest_available_offsets()
        print("最近的偏移量 offset {}".format(last_offset))
 
        # 记录最初的偏移量
        old_offset = last_offset[0].offset[0]
        p = topic.get_producer(sync=False, partitioner=lambda pid, key: pid[0])
        p.produce(str(time.time()).encode())
        s_time = time.time()
        while True:
            last_offset = topic.latest_available_offsets()
            print("最近可用offset {}".format(last_offset))
            if last_offset[0].offset[0] != old_offset:
                e_time = time.time()
                print('cost time {}'.format(e_time-s_time))
                break
            time.sleep(1)
 
    def get_produce_message_report(self, topic):
        """
        查看异步发送消报告,默认会等待5s后才能获得报告
        """
        topic = self.client.topics[topic.encode()]
        last_offset = topic.latest_available_offsets()
        print("最近的偏移量 offset {}".format(last_offset))
        p = topic.get_producer(sync=False, delivery_reports=True, partitioner=lambda pid, key: pid[0])
        p.produce(str(time.time()).encode())
        s_time = time.time()
        delivery_report = p.get_delivery_report()
        e_time = time.time()
        print ('等待{}s, 递交报告{}'.format(e_time-s_time, delivery_report))
        last_offset = topic.latest_available_offsets()
        print("最近的偏移量 offset {}".format(last_offset))
 
 
if __name__ == '__main__':
    host = '192.168.17.64:9092,192.168.17.65:9092,192.168.17.68:9092'
    kafka_ins = KafkaTest(host)
    topic = 'test'
    # kafka_ins.producer_partition(topic)
    # kafka_ins.producer_designated_partition(topic)
    # kafka_ins.async_produce_message(topic)
    kafka_ins.get_produce_message_report(topic)

注意要点:
多进程使用pykafka共享一个client，会造成只有进程能够正常的写入数据，如果使用了dliver_report（包括同步），会导致子进程彻底阻塞掉不可用
使用producer.produce发送数据出现故障，如下

#!/bin/env python
from pykafka import KafkaClient
host = '192.168.17.64:9092,192.168.17.65:9092,192.168.17.68:9092'
client = KafkaClient(hosts = host)
topic = client.topics["test"]
with topic.get_sync_producer() as producer:
    for i in range(100):
        producer.produce('test message ' + str(i ** 2))

报错：

Traceback (most recent call last):
  File "TaxiKafkaProduce.py", line 15, in <module>
    producer.produce(('test message ' + str(i ** 2)))
  File "/root/anaconda3/lib/python3.6/site-packages/pykafka/producer.py", line 325, in produce
    "got '%s'", type(message))
TypeError: ("Producer.produce accepts a bytes object as message, but it got '%s'", <class 'str'>)

是因为kafka传递的字节，因此在传递字符串处encode()即可，分别是client.topics和producer.produce()，如下：

#!/bin/env python
from pykafka import KafkaClient
host = '192.168.17.64:9092,192.168.17.65:9092,192.168.17.68:9092'
client = KafkaClient(hosts = host)
topic = client.topics["test".encode()]
# 将产生kafka同步消息，这个调用仅仅在我们已经确认消息已经发送到集群之后
with topic.get_sync_producer() as producer:
    for i in range(100):
        producer.produce(('test message ' + str(i ** 2)).encode())
同步与异步
from pykafka import KafkaClient
#可接受多个client
client = KafkaClient(hosts ="192.168.17.64:9092,192.168.17.65:9092,192.168.17.68:9092") 
#查看所有的topic
client.topics
print client.topics
 
topic = client.topics['test_kafka_topic']#选择一个topic
 
message = "test message test message"
# 当有了topic之后呢，可以创建一个producer,来发消息，生产kafka数据,通过字符串形式，
with topic.get_sync_producer() as producer:
    producer.produce(message)
# 以上的例子将产生kafka同步消息，这个调用仅仅在我们已经确认消息已经发送到集群之后
 
#但生产环境，为了达到高吞吐量，要采用异步的方式，通过delivery_reports =True来启用队列接口；
producer = topic.get_producer(sync=False, delivery_reports=True)
producer.produce(message)
try:
    msg, exc = producer.get_delivery_report(block=False)
    if exc is not None:
        print 'Failed to deliver msg {}: {}'.format(msg.partition_key, repr(exc))
    else:
        print 'Successfully delivered msg {}'.format(msg.partition_key)
except Queue.Empty:
    pass

(4) pykafka消费者api
pykafka消费者分为simple和balanced两种

simple适用于需要消费指定分区且不需要自动的重分配(自定义)
balanced自动分配则选择

#coding=utf-8
 
from pykafka import KafkaClient
 
 
class KafkaTest(object):
    def __init__(self, host="192.168.237.129:9092"):
        self.host = host
        self.client = KafkaClient(hosts=self.host)
 
    def simple_consumer(self, topic, offset=0):
        """
        消费者指定消费
        :param offset:
        :return:
        """
 
        topic = self.client.topics[topic.encode()]
        partitions = topic.partitions
        last_offset = topic.latest_available_offsets()
        print("最近可用offset {}".format(last_offset))  # 查看所有分区
        consumer = topic.get_simple_consumer(b"simple_consumer_group", partitions=[partitions[0]])  # 选择一个分区进行消费
        offset_list = consumer.held_offsets
        print("当前消费者分区offset情况{}".format(offset_list))  # 消费者拥有的分区offset的情况
        consumer.reset_offsets([(partitions[0], offset)])  # 设置offset
        msg = consumer.consume()
        print("消费 :{}".format(msg.value.decode()))
        msg = consumer.consume()
        print("消费 :{}".format(msg.value.decode()))
        msg = consumer.consume()
        print("消费 :{}".format(msg.value.decode()))
        offset = consumer.held_offsets
        print("当前消费者分区offset情况{}".format(offset)) # 3
 
    def balance_consumer(self, topic, offset=0):
        """
        使用balance consumer去消费kafka
        :return:
        """
        topic = self.client.topics["kafka_test".encode()]
        # managed=True 设置后，使用新式reblance分区方法，不需要使用zk，而False是通过zk来实现reblance的需要使用zk
        consumer = topic.get_balanced_consumer(b"consumer_group_balanced2", managed=True)
        partitions = topic.partitions
        print("分区 {}".format(partitions))
        earliest_offsets = topic.earliest_available_offsets()
        print("最早可用offset {}".format(earliest_offsets))
        last_offsets = topic.latest_available_offsets()
        print("最近可用offset {}".format(last_offsets))
        offset = consumer.held_offsets
        print("当前消费者分区offset情况{}".format(offset))
        while True:
            msg = consumer.consume()
            offset = consumer.held_offsets
            print("{}, 当前消费者分区offset情况{}".format(msg.value.decode(), offset))
 
if __name__ == '__main__':
    host = '192.168.17.64:9092,192.168.17.65:9092,192.168.17.68:9092'
    kafka_ins = KafkaTest(host)
    topic = 'test'
    # kafka_ins.simple_consumer(topic)
    kafka_ins.balance_consumer(topic)

使用consumber_group和consumer_id

# -* coding:utf8 *-  
from pykafka import KafkaClient
 
host = '192.168.17.64:9092,192.168.17.65:9092,192.168.17.68:9092'
client = KafkaClient(hosts = host)
 
print(client.topics)
 
# 消费者  
topic = client.topics['test'.encode()]
consumer = topic.get_simple_consumer(consumer_group='test_group', 
                            # 设置为False的时候不需要添加consumer_group，直接连接topic即可取到消息
                            auto_commit_enable=True, 
                            auto_commit_interval_ms=1,  
                            #这里就是连接多个zk
                            zookeeper_connect='192.168.17.64:2181,192.168.17.65:2181,192.168.17.68:2181' 
                            consumer_id='test_id')

for message in consumer:
    if message is not None:
        #打印接收到的消息体的偏移个数和值
        print(message.offset, message.value)

报错：AttributeError: 'SimpleConsumer' object has no attribute '_consumer_group'
是因为kafka在传输的时候需要bytes，而不是str，所以在str上加上b标识就可以，如下：

# -* coding:utf8 *-  
from pykafka import KafkaClient
 
host = '192.168.17.64:9092,192.168.17.65:9092,192.168.17.68:9092'
client = KafkaClient(hosts = host)
 
print(client.topics)
 
# 消费者  
topic = client.topics['test'.encode()]
consumer = topic.get_simple_consumer(consumer_group=b'test_group', auto_commit_enable=True, auto_commit_interval_ms=1,  consumer_id=b'test_id')
 
for message in consumer:
    if message is not None:
        print(message.offset, message.value.decode('utf-8'))

不要重复消费，对已经消费过的信息进行舍弃

 consumer = topic.get_simple_consumer(consumer_group=b'test_group', 
                             auto_commit_enable=True, 
                             auto_commit_interval_ms=1, 
                             consumer_id=b'test_id')

不希望得到历史数据的时候，需要使用auto_commit_enable这个参数。
当consumer_group=b'test_group'，运行一次后，能够得到正常数据；再一次后，就数据读取不到了，如下：

{b'kafka_test': None, b'test': None}

如果要在读取一次数据，就需要修改consumber_group的id，例如修改成consumber_group=b'test_group_1'后，再运行一次，就可以正常读取数据了。
因为：是kafka的订阅原理，同一个group下，消费之后已经读取完，如果想得到数据必须修改consumber_group_id。
group是消费者中的概念，按照group（组）对消费者进行区分。对于每个group，需要先指定订阅哪个topic的消息，然后该topic下的partition会平均分配到group下面的consumer上。所以会出现以下这些情况：
1、一个topic被多个group订阅，那么一条消息就会被不同group中的多个consumer处理。
2、同一个group中，每个partition只会被一个consumer处理，这个consumer处理的消息不一定是同一个key的。所以需要在处理的地方判断。

三、kafka-python
(1) kafka-python安装

PyPI安装

pip install kafka-python

conda安装

conda install -c conda-forge kafka-python

anaconda自带pip安装

/root/anaconda3/bin/pip install kafka-python
(2) kafka-python的api
https://kafka-python.readthedocs.io/en/master/apidoc/modules.html
https://kafka-python.readthedocs.io/en/master/index.html
https://pypi.org/project/kafka-python/

(3) kafka-python生产者

import time
from kafka import KafkaProducer
 
 
producer = KafkaProducer(bootstrap_servers = ['192.168.17.64:9092', '192.168.17.65:9092', '192.168.17.68:9092'])
# Assign a topic
topic = 'test'
 
def test():
    print('begin')
    n = 1
    try:
        while (n<=100):
            producer.send(topic, str(n).encode())
            print("send" + str(n))
            n += 1
            time.sleep(0.5)
    except KafkaError as e:
        print(e)
    finally: 
        producer.close()
        print('done')
 
if __name__ == '__main__':
    test()

(4) kafka-python消费者

#!/bin/env python
from kafka import KafkaConsumer
 
#connect to Kafka server and pass the topic we want to consume
consumer = KafkaConsumer('test', group_id = 'test_group', bootstrap_servers = ['192.168.17.64:9092', '192.168.17.65:9092', '192.168.17.68:9092'])
try:
    for msg in consumer:
        print(msg)
        print("%s:%d:%d: key=%s value=%s" % (msg.topic, msg.partition,msg.offset, msg.key, msg.value))
except KeyboardInterrupt, e:
    print(e)

输出结果：

ConsumerRecord(topic='test', partition=0, offset=246, timestamp=1531980887190, timestamp_type=0, key=None, value=b'1', checksum=None, serialized_key_size=-1, serialized_value_size=1)
ConsumerRecord(topic='test', partition=0, offset=247, timestamp=1531980887691, timestamp_type=0, key=None, value=b'2', checksum=None, serialized_key_size=-1, serialized_value_size=1)
ConsumerRecord(topic='test', partition=0, offset=248, timestamp=1531980888192, timestamp_type=0, key=None, value=b'3', checksum=None, serialized_key_size=-1, serialized_value_size=1)
ConsumerRecord(topic='test', partition=0, offset=249, timestamp=1531980888694, timestamp_type=0, key=None, value=b'4', checksum=None, serialized_key_size=-1, serialized_value_size=1)
ConsumerRecord(topic='test', partition=0, offset=250, timestamp=1531980889196, timestamp_type=0, key=None, value=b'5', checksum=None, serialized_key_size=-1, serialized_value_size=1)
ConsumerRecord(topic='test', partition=0, offset=251, timestamp=1531980889697, timestamp_type=0, key=None, value=b'6', checksum=None, serialized_key_size=-1, serialized_value_size=1)
ConsumerRecord(topic='test', partition=0, offset=252, timestamp=1531980890199, timestamp_type=0, key=None, value=b'7', checksum=None, serialized_key_size=-1, serialized_value_size=1)
ConsumerRecord(topic='test', partition=0, offset=253, timestamp=1531980890700, timestamp_type=0, key=None, value=b'8', checksum=None, serialized_key_size=-1, serialized_value_size=1)
ConsumerRecord(topic='test', partition=0, offset=254, timestamp=1531980891202, timestamp_type=0, key=None, value=b'9', checksum=None, serialized_key_size=-1, serialized_value_size=1)
ConsumerRecord(topic='test', partition=0, offset=255, timestamp=1531980891703, timestamp_type=0, key=None, value=b'10', checksum=None, serialized_key_size=-1, serialized_value_size=2)

enable_auto_commit=False
consumer = kafka.KafkaConsumer(bootstrap_servers = ['192.168.17.64:9092','192.168.17.65:9092','192.168.17.68:9092'],
                        group_id ='test_group_id',
                        auto_offset_reset ='latest', 
                        enable_auto_commit = False)

自动提交位移设为flase, 默认为取最新的偏移量，重新建立一个group_id,这样就实现了不影响别的应用程序消费数据，又能消费到最新数据，实现预警（先于用户发现）的目的。

————————————————
版权声明：本文为CSDN博主「learn_tech」的原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接及本声明。
原文链接:https://blog.csdn.net/learn_tech/article/details/81115996