Problem request_timeout not working with proxy_next_upstream on proxy_connect_timeout but proxy_read

Server NGINX
1

Hello
In our setup we have an NGNX as front-end and several back-end.
The problem is our load profile, we have a lot of simple and fast http
requests, and very few but very heavy in terms of time and BE cpu
requests.

So my idea is to use proxy_next_upstream for simple request as usual and
it
works perfectly.
And for heavy requests based on URL I want to passthrough it to BE with
lowest CPU load by specifying small proxy_connect_timeout and using
proxy_next_upstream timeout.
But in case of all system overload with heavy requests i don’t want them
to
travel through all BEs because of proxy_read_timeout is about 1 minute.
I was hoping to set a request_timeout to the same value as
proxy_read_timeout and from my point of view this should prevent heavy
requests to travel all upstreams based on proxy_read_timeout, but they
do.

I ve found a similar topic but the proposition was to make two new
options
to proxy_next_upstream such as timeout_tcp timeout_http or something
similar.

Thanks for your future advices and comments.

Posted at Nginx Forum:

2

For beter understanding here is my config sniper

upstream super_upstream {
keepalive 128;
server be1 max_fails=45 fail_timeout=3s;
server be2 max_fails=45 fail_timeout=3s;
server be3 max_fails=45 fail_timeout=3s;
}

server {
server_name pytn.ru;
location ^~ /simple_requests {
proxy_read_timeout 2s;
proxy_send_timeout 2s;
proxy_connect_timeout 10ms;
proxy_next_upstream error timeout invalid_header http_500 http_502
http_503 http_504;
proxy_pass http://super_upstream;
}

location ^~ /very_heavy_requests {
send_timeout 60s;
proxy_read_timeout 60s;
proxy_send_timeout 60s;
proxy_connect_timeout 5ms;
proxy_next_upstream timeout;
proxy_pass http://super_upstream;
}

Posted at Nginx Forum:

3

Hello!

On Wed, Mar 06, 2013 at 05:28:12AM -0500, yvlasov wrote:

But in case of all system overload with heavy requests i don’t want them to
travel through all BEs because of proxy_read_timeout is about 1 minute.
I was hoping to set a request_timeout to the same value as
proxy_read_timeout and from my point of view this should prevent heavy
requests to travel all upstreams based on proxy_read_timeout, but they do.

I ve found a similar topic but the proposition was to make two new options
to proxy_next_upstream such as timeout_tcp timeout_http or something
similar.

Thanks for your future advices and comments.

I think that some aggregate upstream timeout, which will prevent
switch to a next upstream server if passed, whould be better
solution to this problem.

–
Maxim D.
http://nginx.org/en/donation.html

4

Good idea but we have to keep in mind it should depend on location
context.
THX

Posted at Nginx Forum: