1、导入模块
from pykafka import KafkaClient,common
2、实例化
client = KafkaClient(hosts="host:port") #实例化
print(client.topics)
print(client.brokers)
ps:一个坑
broker打印出来的host如果是域名形式的(kafka自己设置了只能通过域名连接),实例化KafkaClient时传入的是ip,在没有配置host的情况下,连接会报socket连接错误
出现socket错误时,首先检查是否是这种情况,如果是,在host文件中配置域名和ip的映射可以解决
3、生产者demo
from pykafka import KafkaClient,common client = KafkaClient(hosts="ip:port") #实例化 print(client.topics) print(client.brokers) topic = client.topics['test_topic'] #指定topic,没有就新建 producer = topic.get_producer() for i in range(1): producer.produce(('test message ' + str(i ** 2)).encode()) producer.stop()
4、消费者demo
from pykafka import KafkaClient,common import time import json class KafkaTest(object): def __init__(self, host): self.host = host self.client = KafkaClient(hosts=self.host) def balance_consumer(self, topic, offset=0): """ 使用balance consumer去消费kafka :return: """ result=[] topic = self.client.topics[topic.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,必须指定 # zookeeper_connect = "zookeeperIp",consumer_group='test_group', consumer = topic.get_balanced_consumer(consumer_group='test_group', auto_commit_enable=True,managed=True, consumer_timeout_ms=1000) 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() if msg: offset = consumer.held_offsets print("当前位移:{}".format(offset)) result.append(eval(msg.value.decode())) print(msg.value.decode()) consumer.commit_offsets() #commit一下 else: print("没有数据") if __name__ == '__main__': host = 'ip:port' kafka_ins = KafkaTest(host) topic = 'test_topic' kafka_ins.balance_consumer(topic)