$memcached_key doesn't fetch unicode url

[Maxim Dounin]

Hello!

On Sun, May 04, 2014 at 06:42:39PM +0300, kirpit wrote:

> Hi,
> 
> I'm doing some sort of downstream cache that I save every entire page into
> memcached from the application with urls such as:
> 
> www.example.com/file/path/?query=1
> 
> then I'm fetching them from nginx if available with the config:
>     location / {
>         # try to fetch from memcached
>         set                 $memcached_key "$host$request_uri";
>         memcached_pass      localhost:11211;
>         expires             10m;
>         # not fetched from memcached, fallback
>         error_page          404 405 502 = @fallback;
>     }
> 
> 
> This works perfectly fine for latin char urls. However, it fails to catch
> unicode urls such as:
> 
> www.example.com/flights/istc-lonc-11062014/lonc-istc-19062014/Cheap-return-tickets-from-%C4%B0stanbul-to-Londra-in-Haziran.html?class=economy&adult=1
> 
> The unicode char "%C4%B0" appears same in the nginx logs, application cache
> setting key (that is actually taken from raw REQUEST_URI what nginx gives).
> 
> The example url content also exist in the memcached itself when if I try:
> "get
> www.example.com/flights/istc-lonc-11062014/lonc-istc-19062014/Cheap-return-tickets-from-%C4%B0stanbul-to-Londra-in-Haziran.html?class=economy&adult=1
> "
> 
> However, nginx cannot fetch anything from memcached, @fallbacks every time.
> 
> I'm using version 1.6.0. Any help is much appreciated.

Value of $memcached_key is escaped to make sure memcached protocol 
constraints are not violated - most notably, space and "%" 
character are escaped into "%20" and "%25", respectively (space is 
escaped as it's not allowed in memcached keys, and "%" to make the 
escaping reversible).

Given the

www.example.com/flights/istc-lonc-11062014/lonc-istc-19062014/Cheap-return-tickets-from-%C4%B0stanbul-to-Londra-in-Haziran.html?class=economy&adult=1

$memcached_key, nginx will try to fetch

www.example.com/flights/istc-lonc-11062014/lonc-istc-19062014/Cheap-return-tickets-from-%25C4%25B0stanbul-to-Londra-in-Haziran.html?class=economy&adult=1

from memcached.

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