I think that is what is going on here… but it seems to be missing a
step. Has anyone else tried this? Basically using PHP (or whatever
language) very quickly to do the lookup of the domain/key (the two
things you need in MogileFS) and then sending the request on to the
mogileFS engine and freeing up the PHP layer?
http://d.hatena.ne.jp/perezvon/20080418/1208531594
From what I can tell, the request comes in:
- http://foo.com/images/foo.gif
- nginx has /images set to proxy_pass to php or something else
- php does a lookup based on that URL and returns a valid mogileFS
storage node (this is where I am not sure it is bulletproof) - it says “X-Accel-Redirect: /reproxy” and throws in the standard
perlbal header “x-reproxy-url” - nginx takes that header and dynamically proxy_pass’es to that URL
Does that make sense? This was my next step once I get my normal web
serving solid, to look into integrating MogileFS-based storage nodes
as close to nginx as possible, and take advantage of X-Accel-Redirect
(or try to get a module created that worked like it with MogileFS
directly?)
Extra bonus: nginx can be used instead of mogstored for the webdav as
well… 
In theory every webserver machine can have PHP/whatever + memcached +
mogstored + nginx. Nginx could do all 3 functions: web serving +
fastcgi proxy + mogstored all in one!
I believe that solution would allow for almost near infinite
shared-nothing horizontal scalability (until your MogileFS database
becomes too big and you have to scale that)
Thoughts?