Hello, Dieter Maurer wrote:

Maybe, I have a much simpler solution:

Something in ZServer makes all file fields seekable by delivering them through some temporary (either a StringIO or a temporary file).

Maybe, you could do the same for your requests?

Yes, I could. But trying to avoid it was the very reason I tried to patch HTTPRequest for. What you propose is what the HTTPServer does for instance. I have at least two objections: 1) For big requests (big means some hardcoded threshold) HTTPServer buffers the request on disk. Then ZPublisher creates a FieldStorage instance which will write the request once again to disk. And that also means the request is read from disk several times. That's a bit too much blocking disk I/O in a production webserver for my liking. And that's not counting retries; each retry will add another pair of blocking disk I/O in the current implementation. What I'm trying to do is: - avoid unneeded disk I/O in retries - make it possible that self.stdin in HTTPRequest can be non-seekable such that it doesn't have to be buffered on disk before FieldStorage is created (and does exactly that buffering). 2) I'm convinced that the one and only ZServer thread should do without any blocking (I/O) call (besides select() that is) and I want AJPServer also be an experiment to see if this improves performance. cheers, andreas