下面小编就为大家分享一篇python队列通信:rabbitMQ的使用(实例讲解),具有很好的参考价值,希望对大家有所帮助。一起跟随小编过来看看吧
(一)、前言
为什么引入消息队列?
1.程序解耦
2.提升性能
3.降低多业务逻辑复杂度
(二)、python操作rabbit mq
rabbitmq配置安装基本使用参见上节文章,不再复述。
若想使用python操作rabbitmq,需安装pika模块,直接pip安装:
pip install pika
1.最简单的rabbitmq producer端与consumer端对话:
producer:
#Author :ywq import pika auth=pika.PlainCredentials('ywq','qwe') #save auth indo connection = pika.BlockingConnection(pika.ConnectionParameters( '192.168.0.158',5672,'/',auth)) #connect to rabbit channel = connection.channel() #create channel channel.queue_declare(queue='hello') #declare queue #n RabbitMQ a message can never be sent directly to the queue, it always needs to go through an exchange. channel.basic_publish(exchange='', routing_key='hello', body='Hello World!') #the body is the msg content print(" [x] Sent 'Hello World!'") connection.close()
consumer:
#Author :ywq import pika auth=pika.PlainCredentials('ywq','qwe') #auth info connection = pika.BlockingConnection(pika.ConnectionParameters( '192.168.0.158',5672,'/',auth)) #connect to rabbit channel = connection.channel() #create channel channel.queue_declare(queue='hello') #decalre queue def callback(ch, method, properties, body): print(" [x] Received %r" % body) channel.basic_consume(callback, queue='hello', no_ack=True) print(' [*] Waiting for messages. To exit press CTRL+C') channel.start_consuming()
消息传递消费过程中,可以在rabbit web管理页面实时查看队列消息信息。
2.持久化的消息队列,避免宕机等意外情况造成消息队列丢失。
consumer端无需改变,在producer端代码内加上两个属性,分别使消息持久化、队列持久化,只选其一还是会出现消息丢失,必须同时开启:
delivery_mode=2 #make msg persisdent durable=True
#Author :ywq import pika,sys auth_info=pika.PlainCredentials('ywq','qwe') connection=pika.BlockingConnection(pika.ConnectionParameters( '192.168.0.158',5672,'/',auth_info )) channel=connection.channel() channel.queue_declare(queue='test1',durable=True) #durable=Ture, make queue persistent msg=''.join(sys.argv[1:]) or 'Hello' channel.basic_publish( exchange='', routing_key='test1', body=msg, properties=pika.BasicProperties( delivery_mode=2 #make msg persisdent ) ) print('Send done:',msg) connection.close()
3.公平分发
在多consumer的情况下,默认rabbit是轮询发送消息的,但有的consumer消费速度快,有的消费速度慢,为了资源使用更平衡,引入ack确认机制。consumer消费完消息后会给rabbit发送ack,一旦未ack的消息数量超过指定允许的数量,则不再往该consumer发送,改为发送给其他consumer。
producer端代码不用改变,需要给consumer端代码插入两个属性:
channel.basic_qos(prefetch_count= *) #define the max non_ack_count channel.basic_ack(delivery_tag=deliver.delivery_tag) #send ack to rabbitmq
#Author :ywq import pika,time auth_info=pika.PlainCredentials('ywq','qwe') connection=pika.BlockingConnection(pika.ConnectionParameters( '192.168.0.158',5672,'/',auth_info ) ) channel=connection.channel() channel.queue_declare(queue='test2',durable=True) def callback(chann,deliver,properties,body): print('Recv:',body) time.sleep(5) chann.basic_ack(delivery_tag=deliver.delivery_tag) #send ack to rabbit channel.basic_qos(prefetch_count=1) ''' 注意,no_ack=False 注意,这里的no_ack类型仅仅是告诉rabbit该消费者队列是否返回ack,若要返回ack,需要在callback内定义 prefetch_count=1,未ack的msg数量超过1个,则此consumer不再接受msg,此配置需写在channel.basic_consume上方,否则会造成non_ack情况出现。 ''' channel.basic_consume( callback, queue='test2' ) channel.start_consuming()
三、消息发布/订阅
上方的几种模式都是producer端发送一次,则consumer端接收一次,能不能实现一个producer发送,多个关联的consumer同时接收呢?of course,rabbit支持消息发布订阅,共支持三种模式,通过组件exchange转发器,实现3种模式:
fanout: 所有bind到此exchange的queue都可以接收消息,类似广播。
direct: 通过routingKey和exchange决定的哪个唯一的queue可以接收消息,推送给绑定了该queue的consumer,类似组播。
topic:所有符合routingKey(此时可以是一个表达式)的routingKey所bind的queue可以接收消息,类似前缀列表匹配路由。
1.fanout
publish端(producer):
#Author :ywq import pika,sys,time auth_info=pika.PlainCredentials('ywq','qwe') connection=pika.BlockingConnection(pika.ConnectionParameters( '192.168.0.158',5672,'/',auth_info ) ) channel=connection.channel() channel.exchange_declare(exchange='hello', exchange_type='fanout' ) msg=''.join(sys.argv[1:]) or 'Hello world %s' %time.time() channel.basic_publish( exchange='hello', routing_key='', body=msg, properties=pika.BasicProperties( delivery_mode=2 ) ) print('send done') connection.close()
subscribe端(consumer):
#Author :ywq import pika auth_info=pika.PlainCredentials('ywq','qwe') connection=pika.BlockingConnection(pika.ConnectionParameters( '192.168.0.158',5672,'/',auth_info ) ) channel=connection.channel() channel.exchange_declare( exchange='hello', exchange_type='fanout' ) random_num=channel.queue_declare(exclusive=True) #随机与rabbit建立一个queue,comsumer断开后,该queue立即删除释放 queue_name=random_num.method.queue channel.basic_qos(prefetch_count=1) channel.queue_bind( queue=queue_name, exchange='hello' ) def callback(chann,deliver,properties,body): print('Recv:',body) chann.basic_ack(delivery_tag=deliver.delivery_tag) #send ack to rabbit channel.basic_consume( callback, queue=queue_name, ) channel.start_consuming()
实现producer一次发送,多个关联consumer接收。
使用exchange模式时:
1.producer端不再申明queue,直接申明exchange
2.consumer端仍需绑定队列并指定exchange来接收message
3.consumer最好创建随机queue,使用完后立即释放。
随机队列名在web下可以检测到:
2.direct
使用exchange同时consumer有选择性的接收消息。队列绑定关键字,producer将数据根据关键字发送到消息exchange,exchange根据 关键字 判定应该将数据发送至指定队列,consumer相应接收。即在fanout基础上增加了routing key.
producer:
#Author :ywq import pika,sys auth_info=pika.PlainCredentials('ywq','qwe') connection=pika.BlockingConnection(pika.ConnectionParameters( '192.168.0.158',5672,'/',auth_info ) ) channel=connection.channel() channel.exchange_declare(exchange='direct_log', exchange_type='direct', ) while True: route_key=input('Input routing key:') msg=''.join(sys.argv[1:]) or 'Hello' channel.basic_publish( exchange='direct_log', routing_key=route_key, body=msg, properties=pika.BasicProperties( delivery_mode=2 ) ) connection.close()
consumer:
#Author :ywq import pika,sys auth_info=pika.PlainCredentials('ywq','qwe') connection=pika.BlockingConnection(pika.ConnectionParameters( '192.168.0.158',5672,'/',auth_info )) channel=connection.channel() channel.exchange_declare( exchange='direct_log', exchange_type='direct' ) queue_num=channel.queue_declare(exclusive=True) queue_name=queue_num.method.queue route_key=input('Input routing key:') channel.queue_bind( queue=queue_name, exchange='direct_log', routing_key=route_key ) def callback(chann,deliver,property,body): print('Recv:[level:%s],[msg:%s]' %(route_key,body)) chann.basic_ack(delivery_tag=deliver.delivery_tag) channel.basic_qos(prefetch_count=1) channel.basic_consume( callback, queue=queue_name ) channel.start_consuming()
同时开启多个consumer,其中两个接收notice,两个接收warning,运行效果如下:
3.topic
相较于direct,topic能实现模糊匹配式工作方式(在consumer端指定匹配方式),只要routing key包含指定的关键字,则将该msg发往绑定的queue上。
rabbitmq通配符规则:
符号“#”匹配一个或多个词,符号“”匹配一个词。因此“abc.#”能够匹配到“abc.m.n”,但是“abc.*‘' 只会匹配到“abc.m”。‘.'号为分割符。使用通配符匹配时必须使用‘.'号分割。
producer:
#Author :ywq import pika,sys auth_info=pika.PlainCredentials('ywq','qwe') connection=pika.BlockingConnection(pika.ConnectionParameters( '192.168.0.158',5672,'/',auth_info ) ) channel=connection.channel() channel.exchange_declare(exchange='topic_log', exchange_type='topic', ) while True: route_key=input('Input routing key:') msg=''.join(sys.argv[1:]) or 'Hello' channel.basic_publish( exchange='topic_log', routing_key=route_key, body=msg, properties=pika.BasicProperties( delivery_mode=2 ) ) connection.close()
consumer:
#Author :ywq import pika,sys auth_info=pika.PlainCredentials('ywq','qwe') connection=pika.BlockingConnection(pika.ConnectionParameters( '192.168.0.158',5672,'/',auth_info )) channel=connection.channel() channel.exchange_declare( exchange='topic_log', exchange_type='topic' ) queue_num=channel.queue_declare(exclusive=True) queue_name=queue_num.method.queue route_key=input('Input routing key:') channel.queue_bind( queue=queue_name, exchange='topic_log', routing_key=route_key ) def callback(chann,deliver,property,body): print('Recv:[type:%s],[msg:%s]' %(route_key,body)) chann.basic_ack(delivery_tag=deliver.delivery_tag) channel.basic_qos(prefetch_count=1) channel.basic_consume( callback, queue=queue_name ) channel.start_consuming()
运行效果:
rabbitmq三种publish/subscribe模型简单介绍完毕。
版权声明:本文内容由互联网用户自发贡献,该文观点与技术仅代表作者本人。本站仅提供信息存储空间服务,不拥有所有权,不承担相关法律责任。如发现本站有涉嫌侵权/违法违规的内容, 请发送邮件至 dio@foxmail.com 举报,一经查实,本站将立刻删除。