So you do use caching? Is it RAM caching or disk caching? If RAM caching, it could be that running under Cherokee and uWSGI is deleting the environment that web2py is run in after a certain number of requests. This would reduce the usefulness of a RAM cache but would also produce the results you're seeing. Whereas a Python-based server (Rocket or Cherrypy) would not do this and a RAM cache would persist as long as the server process (or until otherwise deleted).
Please let us know when you look into the possibilities of caching being the issue. @ron_m: Awesome. Great suggestions! -tim On Dec 25, 3:20 pm, David Zejda <[email protected]> wrote: > Thanks for the all replies! > > I do not directly cache DAL objects, but yes, caching may be behind > the leak > through some references between objects. Maybe I'll examine it with > objgraph, > and I agree, it would be nice to have the function at hand in the > admin interface. > > Cherrypy & anyserver brings no significant change. > > With Cherokee & uWSGI the problem has gone. Moreover the new setup is > subjectively > faster to the previous one (Rocket behind Apache proxy). :) > > David
