I setup the rabbitmq clustering according to the official doc and queue mirroring according to this article. Then I configure a HAProxy in front of the rabbitmq cluster.
Problem
My own script using pika starts fine and is able to connect to the HAProxy and consume the messages in the queue. But it will report the following error every 2 mins or so and cause my client to completely disconnected from HAProxy.
Solution
After googling for hours, this article solves my problem.
In short, on a Linux machine, tcp_keepalive is sent every 7200 seconds.
$ cat /proc/sys/net/ipv4/tcp_keepalive_time
7200
By default, HAProxy has a timeout set to 1m.
defaults
timeout client 1m
timeout server 1m
So if your queue is idle for more than 1m, HAProxy will force to drop the connection.
To fix it, increase the HAProxy timeout value to 3h in your rabbitmq listener.
listen rabbitmq
timeout client 3h
timeout server 3h
文章详细介绍了如何配置RabbitMQ集群、队列镜像及HAProxy,解决使用pika库的自定义脚本连接HAProxy后出现的连接断开问题。通过增加HAProxy的超时时间至3小时,成功避免了2分钟内队列空闲导致的连接被强制断开。

8709

被折叠的 条评论
为什么被折叠?



