Proxy pass confusion

Server NGINX
1

Here is my server setup:

Amazon load balancer (loadbalancer.aws.com) → nginx servers in
reverse-proxy/caching mode → back end PHP servers

If I visit the nginx server directly in my browser, everything works
perfectly.

If I visit loadbalancer.aws.com, the nginx server redirects me to
http://decupstream, which is what I named the upstream block in my
configuration. The nginx.conf looks like:

upstream decupstream {
server 10.167.1.50:8080;
server 10.160.242.232:8080;
server 10.222.218.126:8080;
}

proxy_cache_path /var/cache/nginx levels=1:2 keys_zone=gop2012:10m;

server {
    listen       80;
    server_name  gop-proxy.mycompany.com;

    location / {
      proxy_pass            http://decupstream;
      proxy_cache           gop;
      proxy_cache_valid     10m;
      proxy_cache_use_stale error timeout invalid_header http_500;
      proxy_next_upstream   error timeout invalid_header http_500;
      proxy_cache_lock      on;
      proxy_ignore_headers Cache-Control Expires;
      proxy_set_header Host "gop.mycompany.com";
    }
}

How do I get nginx to forward requests to the Apache/PHP servers even if
they don’t come in directly to the nginx box?

-jsd-

2

On Fri, Aug 24, 2012 at 01:40:27PM -0700, Jon Drukman wrote:

Hi there,

Amazon load balancer (loadbalancer.aws.com) → nginx servers in
reverse-proxy/caching mode → back end PHP servers

If I visit the nginx server directly in my browser, everything works
perfectly.

If I visit loadbalancer.aws.com, the nginx server redirects me to
http://decupstream, which is what I named the upstream block in my
configuration.

Are you sure that it is nginx redirecting you, and not the back-end
servers? Look at the headers from a “curl -i” of a request that fails,
and see is there any indication that it came from the back-end. Maybe
compare that with the headers from the same request that succeeds when
you visit the nginx server directly.

The only way I can think this would happen would be if you didn’t have
the “proxy_set_header Host” line in the actually-running configuration.

The nginx.conf fragment you’ve included fails to load. (“proxy_cache”
zone “gop” is unknown.) Can you confirm that “proxy_set_header Host”
is included in the working one, and provide a minimal fragment that
demonstrates the problem?

For example, if you omit all of the proxy_cache-related directives, do
you still see the problem? If so, then you’ve found a simpler test
case. (And if not, then maybe the cache configuration should be examined
more closely.)

f

Francis D. [email protected]