The Galaxy application does store quite a bit in memory (not as globals though). This doesn't preclude running under mod_wsgi, but it will work best in a configuration that uses a small number of long running processes with multiple threads.
Basically, we run nginx proxying paste on the main Galaxy, so that is the configuration that is most well tested, and that is what we recommend (and can help support most easily). Other configurations might work just fine. I've used Galaxy with various other wsgi servers with issues.
The performance of Paste's http server isn't really an issue since all requests that actually to paste require database access, which is typically orders of magnitude slower than Paste#http's overhead.
-- jt
On Wed, Oct 24, 2012 at 12:23 PM, Paul Boddie paul.boddie@biotek.uio.no wrote:
Having a "persistent" Galaxy-only Web server process, as the Paste-based server seems to be, might inadvertently encourage people to store things in memory (class and module globals, for example) that would mysteriously go away (and cause errors) under different deployment mechanisms, although I would imagine that doing load balancing of several Paste-based servers might provoke similar problems and so they would be known to the community.