Here are some details about the module that we want to sponsor development for: 1. We want to use nginx as a reverse proxy for a Domain Forwarding Service 2. This would require nginx to cache responses that contain only headers and no body (301 HTTP responses) 3. We also need a way to invalidate caches. This should ideally be a command line tool. Regards, Latesh Galia Software Engineer, Directi http://www.directi.com
on 2009-03-18 14:06
on 2009-03-18 14:48
Hi Latesh, Thanks for the information. I have some further questions, mostly to give me a clearer idea of your aims. 1) Are you only looking at caching redirection (i.e. 3xx) responses? 2) Are you looking to cache the responses in memory, on disk, or both? 3) When you say 'invalidate', do you mean purge (i.e. remove the cached entry from the cache)? 4) How are you intending on purging the caches? Here are some possible ways with pros/cons - using ttl's (expiry times) for each cached object individually Pros : provides the most granularity and will probably give the smoothest load to backends Cons : adds slight overhead to the request, and complexity in writing it (if the caching part is done from scratch) - purging all objects at once whenever you wanted Pros : very simple to implement Cons : might cause a temporary rush on the backends when objects are purged - purging a subsection of the objects on a regular basis (e.g. 1/32 of the objects every 30 mins) Pros : very simple to implement, much less of a rush on backends after purges Cons : limited granularity 5) How many objects are we looking at caching roughly? thousands, millions, tens/hundreds of millions? 6) How many worker processes do you/will you use on the Nginx installation (if you use more than 1, then this adds complexity to the caching process, and makes memory caching a pain - to the extent that you might as well use memcached for it). 7) Are you definitely looking at having an internal cache, or would you be happy with using memcached? One idea I thought of is to check for cached pages memcached first for a cached header, then serving from the backend if no header-only object is found. A header-only response would then put the object in memcached. Pros : you could put objects/control the cache separately to the Nginx installation, perhaps saving some overhead Cons : for non-header-only objects, there would be the slight extra overhead of checking memcached on each request (which would obviously be slower than using an in-memory cache) 8) What kind of load does your server get (average req/s etc)? I will probably have a few more questions later, but the answers to the above should give me a better idea of what you want, and give me a better idea of a good way to implement your task. Regards, Marcus.
on 2009-03-18 15:06
Hi Latesh, One more question : what operating system do you run on the server(s) that will have Nginx on? Thanks, Marcus.