Proxy_cache_key generation issues (read: globbing cache invalidation)

[Johan Bergström]

Hey!

At my place, we have a pretty image api written in Python. The generated images are then cached using proxy_cache (which is awesome) and optionally purged by either a) calculating the hash and deleting in file system or b) using the excellent (self proclamation intended :) http://labs.frickle.com/nginx_cache_purge/

So, onto the issue: Our image api accepts a wide range of customization, filters and whatnot, which we therefore chose to cache by unique uri: proxy_cache_key mysecrethash$host$request_uri;

This creates a dilemma when all representations of an image needs to be removed (e.g: a user wants to delete the image). We now have an unknown number of cached representation of this image.

My first approach was to patch proxy_cache even further by storing the cache_key and hash in a table, but after speaking some with mdounin @ irc I prefer the approach of storing to disk using a a plain cache key instead of a calculated one. This would allow me do do something like rm -rf a/b/c/abc.jpg* to remove all representations (e.g: abc.jpg?zoom=3 and abc.jpg?bw=true) [just an example, I know nginx uses inverted catalogue naming]

So, what I'm asking is if there's an better idea than using a plain proxy_cache_key for cache creation to solve my issue. Guess i should note that scanning access logs to find "used" filenames is unavailable to me.


Thanks for reading,
Johan Bergström