Reverse proxy configuration help

jims nginx-forum at
Tue Feb 19 12:27:08 UTC 2013

Jonathan Matthews Wrote:
> On 18 February 2013 15:06, jims <nginx-forum at> wrote:
> > I am new to nginx, it being recommended to solve a problem.
> [ Having read your mail, this kind of reverse proxying is exactly what
> nginx is very good at; I think you're just trying to do too much, too
> quickly, and need to step back from the problem for a moment to
> identify what your first steps should be; then iterate from simple to
> complex behaviours, only moving forward once each behaviour works
> successfully. ]
Point taken.  Going straight for the desired end result doesn't always save

Thanks for your response, Jonathan.  It has been helpful.  Read on for
responses to your comments...

> > The problem:  I have a VPS hosting a website and an application
> server in my
> > DMZ.  I have a test and prod version of each.  I want both DMZ'ed
> servers
> > reverse-proxied such that requests where the referrer is the test
> web server
> > always go to the test app server and requests where the referrer is
> anything
> > but the test web server always go to the production app server.
> When you say "referrer", do you really mean the referrer as
> distinguished by client-originated HTTP headers? I wouldn't do that,
> personally ...
When I say "referrer" I mean the site where the link is presented to the end
user.  If that is what is "distinguished by client-originated HTTP headers"
then yes.
The desired result is that if a person is in our pool of testers and is
testing the development website, any app server link (although pointing
putatively to the production app server) would be sent to the reverse-proxy
that's front-ending the test app server.  The idea is to minimize
unauthorized traffic to the test server.  By using only links that get to
the production app server, if someone saves the link and tries again later,
they will hit the production app server's reverse-proxy front-end.  They
would only hit our test app server if they are actively testing for us. 
Once testing is complete, the proven code can be promoted to te production
webste without having to deal with changing test links to prod links in the
process  Those who will be maintaining the links ongoing should not be
expected either to change links as part of a move-to-production or to have
to learn how to put variables into all the links, and we would not have to
modify the CMS to handle links with variables - they should be able to copy
and paste to create links, which resulting content should be able to be
promoted to production without change, or it defeats the purpose of using a
modern content-management system.
> > The app servers can only be accessed over https, and the proxy will
> > eventually but not quite yet.
> That last part may be more of an issue for you, as you'll discover you
> need an IP address per SSL site you want to host.
Normally, yes, and each of the app server hostnames has its own registered
IP address now, with trusted certs associated.  We are working on obtaining
a wildcard cert which we'd use for the proxy as well as the website, and
will add IP addresses to the proxy if necessary.  I would hope that, since
we want the proxy to choose between two back-end app servers for the same
front-end uri, depending on whether or not there is a referrer of the
development website, one IP should be all that's needed on the front-end,
> > Question:  What is the best way to accomplish this?  I am trying to
> use two
> > different registered host names which are registered to the
> secondary IP on
> > the VPS, as the proxied names for the app servers, but that's not
> working
> > too well.  I wonder if it would be better to have a single server
> name for
> > the proxy with the two proxied servers selected based on referrer,
> rather
> > than trying to redirect to another server name, with one server name
> > servicing one proxied server and the other, the other proxied
> server.
> Goodness, no. I wouldn't /touch/ referer headers for HTTP routing. So
> unreliable!
OK.  How would you recommend ensuring that if you click on a link on our dev
site, it goes to the proxied test app server but if you access that same URL
in any other way, whether by way of a link on the prod website, a bookmark,
someone emailing you the link - the request goes to the proxied prod app
server?  As I said, I'm an nginx newb, so monosyllabic responses are
appreciated... ;)
> > Regardless, I can't seem to get past the connection to the backend
> server.
> > I keep getting a 110 connection failure.  I have tried several
> > configurations but none seem to work.
> What does a connection, via telnet/netcat, from the server, show you?
I get a connection.  I haven't figured out the right HTTP command to send to
get a valid response yet, but I get a response - not a timeout.
> > The problem I'm running into may be related to use of the
> valid_referers
> > directive.  It doesn't seem to do what I need, which is to use one
> back-end
> > for requests referred from one web server host but use the other for
> all
> > other requests.
> I may be repeating a single tune here, but I would really force your
> business to re-examine your requirements if you think this is
> desirable behaviour.
See my earlier response explaining the business requirement, to understand
why this is a desireable behavior. 
> > If I have two server directives with the same IP but two different
> server
> > names, it seems I can't have two location directives, one within
> each server
> > name.
> Each server may have zero or more location directives.
> Each location belongs to exactly one server stanza.
> I don't understand exactly what you think doesn't work, but if it
> contradicts the above 2 lines, then it's not legal nginx config.
If you look at the example conf I posted, that configuration - two separate
server stanzas, each with a location directive, and I get that message.   I
probably have something else misconfigured.  Again, newb...
> >  If I could get that to work, it seems to me it should allow me to
> > redirect to the default app server using the valid_referers
> directive within
> > the referrer-specific app server's server directive, but that
> doesn't seem
> > to work the way I expect, either.
> When you say "redirect" here, you really mean "reverse proxy", don't
> you?
> "Redirecting" is a very specific, unrelated thing in HTTP-server-speak
> .
The redirect is a redirect - telling nginx to use a different reverse-proxy
"upstream" server from what it would normally use based on the URL in the
request.  However, if there is a better way to get the same result I am all
for it.  For example, a method whereby the same front-end url chooses an
upstream server based on the valid_referer criterion, or whatever it is you
would recommend other than the referrer,.
> > I don't have a config file to post because it has gone through a
> dozen
> > iterations already, none of which have been saved.
> apt-get install git-core :-P
I don't want to install apt on my centos server :/  How 'bout 'yum install
> > A generic example of
> > one that doesn't work would be :
> > server {
> >    listen;
> >    server_name;
> >    valid_referers;
> >      if ($invalid_referer) {
> >         return   301$request_uri;
> >       }
> >   proxy_bind;
> >   access_log /var/log/nginx/devpxyaccess.log main;
> >   error_log /var/log/nginx/devpxyerror.log debug;
> >   location / {
> >       proxy_pass;
> >       proxy_redirect / ;
> >    }
> > }
> > server {
> >   listen ;
> >   server_name ;
> >   proxy_bind ;
> >   access_log /var/log/nginx/pxyaccess.log main ;
> >   error_log /var/log/nginx/pxyerror.log debug ;
> >   location / {
> >       proxy_pass ;
> >       proxy_redirect / ;
> >   }
> > }
> >
> The only real problem I can see is that you don't have a resolver
> specified, so nginx doesn't know how to resolve the app FQDNs.
> Irrespective of this, there are much nicer ways to achieve this, which
> might use:
> * Nginx maps to translate from client Host header to backend FQDN.
Would that work if the goal is to direct traffic based on where you're
coming from?  I will explore...
> * Access/error logs specified using variables, but DRY them out at a
> higher level than per-server (i.e. state them once, globally, at the
> http level.
The logs are specified at per-server to quickly identify where the failure
lies.  They will be only at the nginx.conf http level when I have a
suceessful configuration.
> * A single server stanza, switching between backends.
I like the idea - I'm just stuck on how to get it to switch based on where
the client is coming from...
> I could write a version that uses these concepts for you, but I'd be
> depriving you of the educational and life-affirming journey of Getting
> There Yourself if I did ;-)
> If you want to get the best possible help with this, reduce the
> clutter in your example/failing config (i.e. make the smallest
> possible config that doesn't do what you think it /should/ do), and
> re-engage with the list.
> > When I do that it says "location" directive isn't allowed here...
> When you do what?
When I set up my included config file to use the two-server-stanza
configuration I posted (with hostnames/addresses pointing to real-life
stuff, of course) that's what I get when issuing the service restart.
> Jonathan
> -- 
> Jonathan Matthews // Oxford, London, UK
> _______________________________________________
> nginx mailing list
> nginx at

Thanks again - you've been quite helpful. 


Posted at Nginx Forum:,236278,236312#msg-236312

More information about the nginx mailing list