Serve *only* from cache for particular user-agents

[Maxim Dounin]

Hello!

On Fri, Feb 21, 2014 at 11:46:02AM -0500, rge3 wrote:

> Maxim Dounin Wrote:
> -------------------------------------------------------
> > Hello!
> > 
> > On Fri, Feb 21, 2014 at 10:25:58AM -0500, rge3 wrote:
> > 
> > > I havne't found any ideas for this and thought I might ask here.  We
> > have a
> > > fairly straightforward proxy_cache setup with a proxy_pass backend. 
> > We
> > > cache documents for different lengths of time or go the backend for
> > what's
> > > missing.  My problem is we're getting overrun with bot and spider
> > requests. 
> > > MSN in particular started hitting us exceptionally hard yesterday
> > and
> > > started bringing our backend servers down.  Because they're crawling
> > the
> > > site from end to end our cache is missing a lot of those pages and
> > nginx has
> > > to pass the request on through.
> > > 
> > > I'm looking for a way to match on User-Agent and say that if it
> > matches
> > > certain bots to *only* serve out of proxy_cache.  Ideally I'd like
> > the logic
> > > to be:  if it's in the cache, serve it.  If it's not, then return
> > some 4xx
> > > error.  But in the case of those user-agents, *don't* go to the
> > backend. 
> > > Only give them cache.  My first thought was something like...
> > > 
> > > if ($http_user_agent ~* msn-bot) {
> > >       proxy_pass http://devnull;
> > >  }
> > > 
> > > by making a bogus backend.  But in nginx 1.4.3 (that's what we're
> > running) I
> > > get
> > > nginx: [emerg] "proxy_pass" directive is not allowed here
> > > 
> > > Does anyone have another idea?
> > 
> > The message suggests you are trying to write the snippet above at 
> > server{} level.  Moving things into a location should do the 
> > trick.
> > 
> > Please make sure to read http://wiki.nginx.org/IfIsEvil though.
> 
> That seems to have done it!  With a location block I now have...
> 
>                        location / {
>                                 proxy_cache_valid  200 301 302  30m;
> 
>                                 if ($http_user_agent ~* msn-bot) {
>                                         proxy_pass http://devnull;
>                                 }
> 
>                                 if ($http_user_agent !~* msn-bot) {
>                                         proxy_pass http://productionrupal;
>                                 }
>                         }

Second condition can be removed, it's surplus. Just a

     location / {
         if (...) {
            proxy_pass ...
         }

         proxy_pass ...
     }

should be enough.

> That seems to work perfectly.  But is it a safe use of "if"?  Is there a
> safer way to do it without an if?

As long as it's full configuration, there should be no problems.

-- 
Maxim Dounin
http://nginx.org/