Interesting approach Malcom, I will read up on a few of the tools to
see if I can catch the error in the act. I have been doing something
similar with strace and some sort of segfault that is very hard to
catch.

Graham, I haven't researched mod_wsgi much, is it an alternative to
mod_python or does it just relate to the development server (as that's
the only place I've seen it mentioned)

trey

On Oct 6, 7:01 pm, Graham Dumpleton <[EMAIL PROTECTED]>
wrote:
> The way mod_python reads request content is a bit broken and can screw
> up in various ways, albeit rarely. This may be another example to add
> to the existing ones. The other known examples are:
>
>  http://issues.apache.org/jira/browse/MODPYTHON-212
>  http://issues.apache.org/jira/browse/MODPYTHON-234
>
> I haven't sat down to try and work out how to fix the code in
> mod_python, but being mindful that it was doing the wrong thing, I
> completely wrote from scratch in mod_wsgi the input reading code and
> didn't rely on the mod_python code in any way. Thus, as far as I know
> these problems shouldn't occur with mod_wsgi.
>
> Graham
>
> On Oct 7, 2:06 am, Malcolm Tredinnick <[EMAIL PROTECTED]>
> wrote:
>
> > On Sat, 2007-10-06 at 15:20 +0000, Trey wrote:
> > > There are other people that have brought this up a little bit some
> > > time ago. I run a small to medium sized web application that takes
> > > profile pictures. By far my largest customer service issue is people
> > > not being able to upload their photos.
>
> > > For the most part I have played it down as their connection sucking or
> > > perhaps doing something stupid with the browser, but there are a
> > > couple of things that I am running into that are causing an issue.
>
> > I internally wince a bit every time this problem comes up. There's
> > something going on that makes it occur regularly for certain
> > combinations of browsers and servers (there's a report in Edgewall's own
> > Trac installation about a repeatable upload problem for one group of
> > people that is the same problem -- see [1]). However, it's not easily
> > reproducible for other people.
>
> > [1]http://trac.edgewall.org/ticket/5226
>
> > > 1. I can't replicate this, no matter what I do with my browser in the
> > > middle of an upload.
> > > 2. Judging by the django code near the problem, this is working on
> > > information that has already been received.
> > > 3. I get this a few times a day at least, different people every time.
>
> > > If someone with some knowledge of the core or http areas of django
> > > could speak up on this I would be very grateful.
>
> > The error itself appears to be coming from inside mod_python. The
> > self._req attribute is mod_python's request object and we're in the
> > process of reading data from there. Unfortunately, that doesn't help
> > very much with the problem diagnosis, but if you Google around for the
> > exact error message, you'll see it appearing in other places as well.
> > Just not commonly enough to help work out what's really going on. I
> > don't think there's a lot we can do on the Django side to mitigate this.
>
> > If it was me running a site that had this problem and I was really
> > dedicated to solving it, I'd try to put a rolling tcpdump on the
> > connection to try and capture the problem live. It will be a real bear
> > to try and interpret the results, but if you have accurate
> > (synchronised) clocks on all the machines in question and are capturing
> > the exception when it happens (with a timestamp), you should be able to
> > get close. Tcpdump has the ability to roll the logs after a certain size
> > and to create no more than a certain number of log files before reusing
> > the first ones (so it can use a ring of files, essentially). And
> > remember that it doesn't capture all the packets, only the first few
> > bytes -- enough to determine the protocol and see the start of the
> > message body. You can tweak this, of course, but the point is you won't
> > be capturing *all* of the traffic to your site, so storage requirements
> > won't be unmanageably huge (but they will be large). If you can set
> > aside a few gigabytes of disk space to capture logs and are able to
> > respond reasonably fast when the error occurs (so you can grab the log
> > that will contain the error before it gets overwritten next time
> > around), this is a viable technique.
>
> > Unfortunately, it may not tell you a lot (it's always a good idea to
> > know what you hope to get out of a major debugging thing like this). You
> > might see that either there is a slight hold-up in the client sending
> > data and maybe you'll be able to see the browser type in the HTTP
> > headers. Or you might see no break in the packet timings at all, which
> > says there was a hold up inside application space (Apache, I guess). I'm
> > not sure how either piece of information will help much, although
> > knowing where to look further will always be a benefit.
>
> > I guess I'd also try to ensure I was running the latest mod_python (and
> > Apache) in cases like this. The code is getting better all the time --
> > you just don't see any huge "I upgrade mod_python and the wheels
> > completely fell off my site" problems -- so taking advantage of all the
> > bug fixes you can can't hurt here, just in case it is an
> > application-layer problem.
>
> > Please realise that everything I've written here is along the lines of
> > how I would go about trying to gather more information. I have no
> > special insight into what is causing the problem, although from reading
> > a bunch of problem reports on the web, it sounds like it is repeatable
> > for certain client setups, so might not be just a sporadic connection
> > issue.
>
> > Regards,
> > Malcolm


--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups 
"Django users" group.
To post to this group, send email to django-users@googlegroups.com
To unsubscribe from this group, send email to [EMAIL PROTECTED]
For more options, visit this group at 
http://groups.google.com/group/django-users?hl=en
-~----------~----~----~----~------~----~------~--~---

Reply via email to