mike_mp at zzzcomputing.com
Wed Sep 8 21:31:27 EDT 2004
hi - this is a slightly tricky one to demonstrate. Running apache 2.0 with the worker MPM on linux, it appears that a page that requests a Session, and then takes a really long time to finish, blocks certain other session-enabled pages from completing, regardless of which session ID they have. though not every page, just maybe ones in the same child process ? not totally sure. I am assuming this behavior is due to the global locking mechanism used within Session (to which I cant readily find documentation). I have a feeling this is "expected" behavior as opposed to a bug....though it sort of sucks if thats the case. I suppose I am looking for a description of why the below example does what it does. first, run apache with the worker MPM with a pretty small process/thread size, say 5 ThreadsPerChild and 2 StartServers. then, mptest.py, configured just like mptest.py in the mod python documentation: from mod_python import apache, util, Session from time import sleep def handler(req): req.content_type = 'text/html' session = Session.Session(req) fields = util.FieldStorage(req) if fields.has_key('pause'): req.write("hi, this is a very slow page\n") sleep(5) req.write("really, a very slow page\n") sleep(5) req.write("ok\n") else: req.write(""" hello world ! reloading in 100ms <script>setTimeout('window.location.href="mptest.py"', 100)</script> """) return apache.OK Then, set your browser to call the fast version over and over again: http://myserver.com/mptest.py then to watch it hang due to a totally unrelated access by someone else, go to a shell window somewhere and type: wget --output-document=- http://myserver.com/mptest.py?pause=1 to get the slow version. at least on my server, the fast-reloading page will hit a certain process/thread that will just hang, and there it will stay until your slow command-line one completes. i am gathering this is either totally new news, or utterly obvious to everyone but me. comments ?