On 9/20/05, David Binger  wrote:
> On Sep 20, 2005, at 5:50 PM, Mike Orr wrote:
>
> >>
> >> PROPFIND is a webdav method.  Webdav is described in rfc 2518.
> >> Maybe this is just a browser trying to mount your server.
> >>
> >
> > But should it be?  If the application is not meant to be WebDAV
> > friendly, is there any reason not to send an error and force them to
> > use GET?
>
> I agree that the unimplemented method response is appropriate here.
> I just meant that these requests might be not be attacks.
> Should it be Quixote or an application that restricts the request
> method?
> What if your application really wants to implement WebDAV?

It looks like a job for ._q_traverse().  How about something like this?

def get_methods(self, path):
    """Return a list of HTTP methods allowed for this traversal path."""
    return ['GET', 'POST']

def _q_traverse(self, path):
    ...
    method = quixote.get_request().get_method().upper()
   if method == 'OPTIONS':
       return self.get_methods()    # Properly formatted per RFC, of course..
    elif method == 'TRACE':
        return quixote.get_request().get_headers()  # Properly formatted.
        # request.get_headers() does not exist yet.
        # I'm not really sure what this method is supposed to do.
    elif method not in self.get_methods(path):
        quixote.get_response().set_status(501)
        return "ERROR MESSAGE"
    ...


The user can override .get_methods() for WebDAV.  They'll have to
override ._q_traverse() for the action behavior.  If it's reasonable
we can add a WebDAV mixin class as an example.  The behavior is so
application-specific though I'm not sure there's much we could put in
a mixin.  Maybe just a note in the module docstring suggesting a
pattern.

> >> It will be interesting to see how well that works.
> >> It seems relatively expensive.
> >>
> >
> > It won't work for huge logs.  The last organization I worked for had
> > logs that were 1+ GB compressed.  But normalizing the user agent field
> > (or leaving it out) and using a datetime for the date and an int for
> > the size should partly offset the overhead.  I've also set an
> > expiration of one week for images, and that cut the number of requests
> > significantly.
>
> I was thinking that there might be contention at the database level
> since this would, I assume, mean frequent writes to the same table.
> Maybe that's not a problem, though.

MySQL should multiplex them fine; it's designed to do that.  If MySQL
gets overwhelmed and aborts, we'll have to think of another strategy,
but it doesn't seem close to that now.

Actually, the company I last worked for *did* keep their
multi-gigabyte access log in both MySQL and flat files.  They
normalized every string field including the URL, so any interesting
query required joining several tables.  Doing a query took twenty
minutes each!

--
Mike Orr  or