|
Jack Diederich
jack_diederich at email.com
Mon Jul 29 11:20:21 EST 2002
I tried sending this before I was a member of the list, and I guess it wasn't approved... I have a project running under mod_python with about 10k of python source. On a pIII-500 it takes 2 seconds for the first request to an httpd process, and 0.2 seconds for each request after that to run. This isn't a show stopper for now, but it really bugs me (the software will be production in about a month) I've thought of switching to WebWare, which shouldn't have this problem, but is otherwise more than I need and the switching costs would be non-zero. The problem seems to be that the imports, even with the PythonImport directive listing all of them aren't done until ChildInitHanlder(), which is called the first time a child process gets its first hit, apparently. I say apparently because the difference between using PythonImport or not is zero time-wise. Using .pyo files instead of .py cuts the first-hit time in half, but is still five times larger than the time to handle a second hit. Why can't the work (whatever it is) be done _before_ the fork of the children? This is what mod_perl does. mod_php doesn't, it has some possibly intentional design problems that require a fresh interperator init and byte compile between requests (but there are hooks in the code for the $ work-aournd module). But looking through the python internals I couldn't figure out why this isn't possible. I couldn't figure out where to patch it myself, but I didn't see any showstoppers like with php. So ... anyone know how to work around this, why its done this way, or where I should look to patch it? thanks in advance, -jackdied -- __________________________________________________________ Sign-up for your own FREE Personalized E-mail at Mail.com http://www.mail.com/?sr=signup Get 4 DVDs for $.49 cents! plus shipping & processing. Click to join. http://adfarm.mediaplex.com/ad/ck/990-1736-3566-59
|