[mod_python] Req object shared?

Jamieson Becker wrote ..
> btw, Graham, what's the best way to handle error management with
> mod_python? log it to a db? catch all exceptions and throw them in a
> file? Is the best pythonic way to throw errors to throw exceptions, or
> should you return an error code from your functions? Any good tips or
> hints here? I've actually been using Jabber to throw errors and debug
> info in real time at me while coding, but Jabber 1x screws up the order
> of the messages. Jabber 2x works great but there's a noticeable delay
> while the page is generated.

Error management isn't something I have gone into really. So far
I have only been using mod_python for simple static based web site
construction and simple forms based utilities talking to some live
applications using XML-RPC where I am the primary user. In the
first instance, no errors to report, in the latter I am the only real user
and so I have PythonDebug still On.

It is interesting to see you using Jabber though. Do you have to
create a seperate socket connection to Jabber each time or does the
connection to it persist? Does the message delivery aspects of the
Jabber Python client run in a separate thread or when you publish
a log message does it block until it has been delivered to the Jabber
server and that is why you get a delay?

If you are interested in this ability to get real time information out
of Apache, you might be interested in one of my other projects.
This project gives some similar capabilties to Jabber, ie., request
and reply messaging, plus presence and publish and subscribe
features. This package though is targeted specifically at application
development which needs to be spread across multiple processes
and platforms. This is different to Jabber where its main purpose
is user focused instant messaging and using it in applications has
come along later.

With my package I already have it working to the point where each
Apache subprocess can connect into the messaging system and
thus request handlers can initiate remote calls against a backend
application. One can even have services running in the Apache
processes themselves and services running within distinct Python
interpreters within the same Apache process can even talk amongst
themselves at the same time as Apache is handling HTTP requests.
The underlying message delivery runs in a separate thread and all
socket communications is event driven, so true blocking of the
system doesn't occur.

Anyway, getting back to what you are doing with logging, with this
system any error which you want logged could be published by
a service associated with that Python interpreter in the Apache
process. Another service running in the backend application could
be subscribed to all these published log messages and it could
collect them and log them to a file or even show them in realtime
via a GUI if desired. Probably not much different to what you are
doing, but using a different system.

For reference, the package I am talking about is OSE. Information
can be found at:

  http://ose.sourceforge.net

The documentation for the Python wrappers is at:

  http://ose.sourceforge.net/browse.php?group=python-manual&entry=manual.htm

Documentation doesn't as such explain how to hook it into Apache
though as ability to do that is only in beta version of 8.0 whereas
documentation describes version 7.0.

I can give further details and finally get around to incorporating an
example into Vampire if there is interest.


Graham