设置log_format，添加request_time，$upstream_response_time，位置随意
og_format main '"$request_time" "$upstream_response_time" $remote_addr - $remote_user [$time_local] "$request" '
'$status $body_bytes_sent "$http_referer" '
'"$http_user_agent" "$http_x_forwarded_for"';
日志输出效果：
"0.015" "0.015" 10.1.2.3 - - [20/Mar/2017:04:05:49 +0800] "GET /myApp/servlet/TestServlet HTTP/1.1" 200 52 "-" "Mozilla/4.0 (compatible; MSIE 4.0; Windows NT)" "-"
-------------------------------------------------------------------------------
笔者在根据nginx的accesslog中$request_time进行程序优化时，发现有个接口，直接返回数据，平均的$request_time也比较大。原来$request_time包含了用户数据接收时间，而真正程序的响应时间应该用$upstream_response_time。
下面介绍下2者的差别：
1、request_time
官网描述：request processing time in seconds with a milliseconds resolution; time elapsed between the first bytes were read from the client and the log write after the last bytes were sent to the client 。
指的就是从接受用户请求的第一个字节到发送完响应数据的时间，即包括接收请求数据时间、程序响应时间、输出
响应数据时间。
2、upstream_response_time
官网描述：keeps times of responses obtained from upstream servers; times are kept in seconds with a milliseconds resolution. Several response times are separated by commas and colons like addresses in the $upstream_addr variable
是指从Nginx向后端建立连接开始到接受完数据然后关闭连接为止的时间。
从上面的描述可以看出，$request_time肯定比$upstream_response_time值大，特别是使用POST方式传递参数时，因为Nginx会把request body缓存住，接受完毕后才会把数据一起发给后端。所以如果用户网络较差，或者传递数据较大时，$request_time会比$upstream_response_time大很多。
所以如果使用nginx的accesslog查看php程序中哪些接口比较慢的话，记得在log_format中加入$upstream_response_time。