I am using Tornado web framework but one of my deployment environments is mod_wsgi on prefork Apache 2 (weird hosting requirement).
Is processes=n threads=1 the correct way to set up my WSGIDaemonProcess for a tornado.wsgi.WSGIAdapter? My own code should be threadsafe however, I am binding a SQLAlchemy Session() to an instance of tornado.web.RequestHandler in the handler's .prepare() as here (I'm obviously not using coroutines in WSGI-mode). I'm also using blocking tornado.httpclient.HTTPClient in some request handlers.
Also, does WSGIDaemonProcess process fork() comparably to tornado.httpserver.HTTPServer().start(n)?
(I am posting this question here since I have no idea how to ask both @Ben Darnell and @Graham Dumpleton at the same time without spamming both tornado and mod_wsgi lists).
WSGIAdapter is thread-safe, and so is the subset of Tornado that works with it (RequestHandler.write, for example, is thread-safe when used with WSGIAdapter, but not when used with IOLoop). So when running with WSGIAdapter, you can configure your WSGI server in the same way you would for any other WSGI framework (typically something like 1 process per CPU core (this is comparable to what you get with HTTPServer.start(n)) and 5-10 threads per process depending on how much of your time you spend in I/O).
Note that tornado.httpclient.HTTPClient objects are not thread-safe, so you'll need to put a lock around it or use a separate object per thread.