提高RabbitMQ的File descriptors

1. 问题描述

按默认配置安装完RabbitMQ后，发现其File descriptors（即文件描述符）和Socket descriptors都特别低，分别为924和829。客户端（消费者）保持长连接时很容易就将socket占满。

[root@hadoop1 ~]# rabbitmqctl status
Status of node rabbit@hadoop1 ...
[{pid,18946},
 {running_applications,
     [{rabbitmq_management,"RabbitMQ Management Console","3.7.7"},
 ..........................................................................
 ..........................................................................
 ..........................................................................
 {disk_free_limit,50000000},
 {disk_free,45836664832},
 {file_descriptors,
     [{total_limit,924},
      {total_used,503},
      {sockets_limit,829},
      {sockets_used,501}]},
 {processes,[{limit,1048576},{used,3890}]},
 {run_queue,0},
 {uptime,6841},
 {kernel,{net_ticktime,60}}]

经查该问题是由于系统当前的ulimit -n仅为1024，rabbitmq在启动时会进行如下换算：

file_limit = 1024 - 100; // 924
sockets_limit = trunc((1024 - 100) * 0.9 - 2); //829

2. 解决步骤

RabbitMQ的File descriptors（文件描述符）的数量受到三个参数的约束：

1. 系统级别，配置文件为：/proc/sys/fs/file-max，如果太小需要修改。

[root@hadoop1 ~]# cat /proc/sys/fs/file-max 
778230
[root@hadoop1 ~]# cat /proc/sys/fs/file-nr 
2432	0	778230

2. 用户级别，配置文件为：/etc/security/limits.conf

* - nofile 65536

3. 进程级别，即ulimit -n

#修改
[root@hadoop1 ~]# ulimit -n 65536

以上三个参数都确认>=我们需要的连接数后，重启RabbitMQ，注意需要连Erlang一起重启：

#彻底关闭rabbitmq，包括erlang进程
[root@hadoop1 ~]# rabbitmqctl stop
Stopping and halting node rabbit@hadoop1 ...

#在后台启动rabbitmq服务
[root@hadoop1 ~]# rabbitmq-server -detached
Warning: PID file not written; -detached was passed.

#查询重启后rabbitmq的file_descriptors参数
[root@hadoop1 ~]# rabbitmqctl status
Status of node rabbit@hadoop1 ...
[{pid,5381},
.........................................
 {file_descriptors,
     [{total_limit,65436},
      {total_used,2},
      {sockets_limit,58890},
      {sockets_used,0}]},
 {processes,[{limit,1048576},{used,243}]},
 {run_queue,0},
 {uptime,3},
 {kernel,{net_ticktime,60}}]

重启后即可看到效果。