I'm porting my Quixote/Durus application to Pylons, which has a
multithreaded server.  It's kind of a pain to make a connection
threadlocal and initialize it in each thread, and I'd also like a
shared object cache rather than one in each thread. The database is
prebuilt and readonly, so there's no modifications going on.  Right
now with three SCGI subprocesses, each process is 110 MB containing
mostly the memory cache.  We do a lot of complex searches so search
speed is more important than memory use, still it would be nice to
have one 80MB cache rather than multiple.  Is this possible in Durus?

An alternative would be going to ZODB, which is thread safe.  However,
Durus has been working well for us and I have only a few months to
upgrade the application, so I'd rather not switch to a new library.

Another issue is this will have to run both as a web app and as a
standalone (a local web app under wxPython).  So far this is working
with Quixote and Durus.  We'll have to see how Pylons does with it.

--
Mike Orr