Big difference between upstream response time and request time

Server NGINX
1

Hi,
I’m using nginx as a reverse proxy to my backends. I see there is a
huge discrepancy between upstream response time and request time.

X.X.X.X [15/Oct/2008:09:32:57 -0700] “GET /js/something.js HTTP/1.1”
upstream_response_time 0.005 request_time 105.026 90732578
10.3.0.134:8000

It’s generally negligible. But sometimes its as high as 5 or more
seconds, and in some cases as high as the example above.

Any idea what’s happening? nginx is running on a dedicated server, and
has about 2000 active connections.

worker_processes 4;
worker_connections 4096;

keepalive_timeout 120;
send_timeout 90;

proxy_read_timeout 120;
proxy_buffer_size 8k;
proxy_buffers 32 32k;

Thanks.

2

Hello!

On Wed, Oct 15, 2008 at 10:25:20AM -0700, Praveen wrote:

Any idea what’s happening? nginx is running on a dedicated server, and
has about 2000 active connections.

The $request_time variable measures wall clock time it took to
fully process request, from first byte got from client to last
byte sent to client. This includes time spent waiting for client
and may really vary depending on what client does / quality of
client’s network connection / etc.

Generally you shouldn’t pay much attention to this number unless
you have control over the client in question.

Maxim D.

3

Thanks! What about the upstream response time? I’m assuming that it is
time from initiating a connection to the backend server till the last
byte of the response is received.

-p

On Wed, Oct 15, 2008 at 11:03 AM, Maxim D. [email protected]
wrote:

10.3.0.134:8000
and may really vary depending on what client does / quality of
client’s network connection / etc.

Generally you shouldn’t pay much attention to this number unless
you have control over the client in question.

Maxim D.

–
Praveen

Can you yell “Movie!” in a crowded firestation?

4

Hello!

On Wed, Oct 15, 2008 at 01:26:26PM -0700, Praveen wrote:

Thanks! What about the upstream response time? I’m assuming that it is
time from initiating a connection to the backend server till the last
byte of the response is received.

Yes. And $upstream_response_time worth monitoring, since increase
here indicate that something goes wrong with your backends.

Note: $upstream_response_time may contain multiple values,
separated by ", ". This may happen if for some reason request
wasn’t processed by first backend server and went to other one
according to proxy_next_upstream setting.

Maxim D.