当nginx用作代理时，为什么Rails中间件开销如此之高？

Question

Thanks to Godfrey Chan for the helpful insights I got from him so far. He pointed me to the fact that there's a Rack middleware that will provide me more accurate timing about the whole Rails request in the X-Runtime HTTP header in the response, when compared to the reported time in the logs (Completed in XXXms...).

Here's what I get from my tests:

1 - Access the direct URL to the Rails action in Chrome:

• X-Runtime: 25ms
• Chrome "Waiting" time: 27ms
• Reported time in the Rails log format: Completed 200 OK in 7ms (Views: 0.9ms | Sequel: 3.0ms)

2 - Access the same URL but using nginx with proxy_pass, also in Chrome:

• X-Runtime: 84ms
• Chrome "Waiting" time: 88ms
• Completed 200 OK in 7ms (Views: 0.8ms | Sequel: 2.9ms)

3 - Copying the Curl address from Chrome's developer tool and running it with curl -I:

• X-Runtime: 105ms (sometimes it goes up to 400ms)
• Completed 200 OK in 88ms (Views: 2.0ms | Sequel: 5.5ms)

Those timings are pretty much consistent when I try them a lot of times.

Any ideas why Rails will take longer to serve the same request if it goes through nginx proxy_pass? I understand that Curl can't take advantage of features like keep-alive, but I believe nginx is able to get advantage of it. But anyway, the time to open a connection shouldn't be taken into account by X-Runtime header, right?

Answer 1

我不能再重现这个问题了，所以不要介意研究它，除非你自己重现它并提供更多细节来调查它。