Question

我有nginx和php-fpm的问题，我已经在这个表单中搜索了但是我找不到解决方案。

服务器： 32核心 64gb内存

设定：

nginx.conf：

worker_processes  32;


events {
    worker_connections 2048;
}

在php-fpm配置

中

pm.max_children = 1500
pm.start_servers = 500
pm.min_spare_servers = 500
pm.max_spare_servers = 500

它可以工作几分钟，但如果我们在服务器上有更多用户，我会收到关注错误消息：

[06-Oct-2014 15:23:20]警告：[池www]似乎很忙（你可能需要增加pm.start_servers，或pm.min / max_spare_servers），产生8个孩子，有483个空闲，和1068名儿童

页面非常慢，很多用户都有502坏网关消息

负载平均值约为20 20 20

我尝试了很多不同的设置，但我不知道是什么原因，我希望sombody可以提供帮助。

祝你好运

Answer 1

你的＆＃34; net.core.somaxconn＆＃34;值？可以通过执行＆＃34; cat / proc / sys / net / core / somaxconn＆＃34;找到它。它应该足够大，可以处理大量的请求。默认值为128，在您的情况下非常少。如果它更少，请将它设为2048。

＆＃34; listen.backlog＆＃34; php-fpm中的指令应该足够大以排队请求。 somaxconn值应大于积压值。

Answer 2

您有32个工作进程，最多可接受2048个连接 (yes, each)。

这意味着您的nginx最多可同时接受65536个连接。

但是你的PHP-FPM最多可以有1500个实例来处理PHP请求。

现在，根据PHP与静态文件请求的比例，您缺少可用的PHP-FPM流程。

假设您的典型网页上的平均PHP /静态比率为1：5，您（理论上）可以提供多达7,500个并行请求。所有其他人都会遇到错误的网关错误。

这只是一个理论数字，有许多因素需要考虑（保持活动，缓存，内核TCP设置等）以确定实际最大值，但这应该可以让您了解问题所在。

Answer 3

感谢您的回答，我现在使用的是unix socket而不是TCP / IP

我现在改为： somaxconn：2048 和listen.backlog：1024

两人都在128岁。

这里再次提供错误日志的示例（这是在我更改之前）

    [07-Oct-2014 11:54:32] NOTICE: [pool www] child 23114 exited with code 0 after 234.275112 seconds from start
    [07-Oct-2014 11:54:32] NOTICE: [pool www] child 25056 started
    [07-Oct-2014 11:54:45] WARNING: [pool www] seems busy (you may need to increase pm.start_servers, or pm.min/max_spare_servers), spawning 8 children, there are 0 idle, and 108 total children
    [07-Oct-2014 11:54:46] WARNING: [pool www] seems busy (you may need to increase pm.start_servers, or pm.min/max_spare_servers), spawning 16 children, there are 0 idle, and 116 total children
    [07-Oct-2014 11:54:47] WARNING: [pool www] seems busy (you may need to increase pm.start_servers, or pm.min/max_spare_servers), spawning 32 children, there are 0 idle, and 132 total children
    [07-Oct-2014 11:54:48] WARNING: [pool www] seems busy (you may need to increase pm.start_servers, or pm.min/max_spare_servers), spawning 32 children, there are 0 idle, and 164 total children
    [07-Oct-2014 11:54:49] WARNING: [pool www] seems busy (you may need to increase pm.start_servers, or pm.min/max_spare_servers), spawning 32 children, there are 0 idle, and 196 total children
    [07-Oct-2014 11:54:50] WARNING: [pool www] seems busy (you may need to increase pm.start_servers, or pm.min/max_spare_servers), spawning 32 children, there are 0 idle, and 228 total children
    [07-Oct-2014 11:54:51] WARNING: [pool www] seems busy (you may need to increase pm.start_servers, or pm.min/max_spare_servers), spawning 32 children, there are 0 idle, and 260 total children
    [07-Oct-2014 11:54:52] WARNING: [pool www] seems busy (you may need to increase pm.start_servers, or pm.min/max_spare_servers), spawning 32 children, there are 11 idle, and 292 total children
    [07-Oct-2014 11:54:53] WARNING: [pool www] seems busy (you may need to increase pm.start_servers, or pm.min/max_spare_servers), spawning 32 children, there are 5 idle, and 324 total children
    [07-Oct-2014 11:54:54] WARNING: [pool www] seems busy (you may need to increase pm.start_servers, or pm.min/max_spare_servers), spawning 32 children, there are 36 idle, and 356 total children
    [07-Oct-2014 11:55:02] WARNING: [pool www] child 25306, script '/srv/www/examplepage.com/public/index.php' (request: "GET /index.php") executing too slow (13.037138 sec), logging
    [07-Oct-2014 11:55:02] WARNING: [pool www] child 24865, script '/srv/www/examplepage.com/public/index.php' (request: "GET /index.php") executing too slow (12.989517 sec), logging

这发生在10秒内，从10个激活的fpm进程到超过400个。

Answer 4

问题很遗憾没有解决，我现在安装了新的遗物，这里举例说明当php-fpm进程突然增加时会发生什么：

http://imgur.com/Zm0BEIz

此时服务器尚未完全占用。

看起来像redis有问题（我没有安装memcached）。

警告：[pool www]似乎很忙（您可能需要增加pm.start_servers或pm.min / max_spare_servers）产生

4 个答案: