[Spambayes] Guidance re pickles versus DB for Outlook

Moore, Paul Paul.Moore@atosorigin.com
Tue Nov 26 10:37:52 2002


From: Mark Hammond [mailto:mhammond@skippinet.com.au]
> I've been thinking about the "database" to use for the Outlook plugin. =
 I
> see two reasonable choices today: pickles and whatever anydbm picks up =
on
> Windows.

I agree this needs some consideration. I've also noticed the Outlook startup
and shutdown slowness. If we ever move to a "train assuming the classifier
got it right" model, we get lots more training, and therefore even less
chance we can avoid the shutdown slowup.

One potential problem with anydbm is that it picks up Berkleley DB on Windows,
and the version shipped with Python 2.2 is very old (1.85, I believe) and
has known bugs. I got some peculiar behaviour (hanging) with pop3proxy using
anydbm. There's no reason to believe that problems will be common, but any
problems that *do* occur will be awfully hard for the user to locate, let
alone diagnose.

> So, I see two basic routes I can take:
> * Move to a DB, but stick with a fully synchronous model.
> * Stick with pickles, but move to a threaded asynchronous model.

Pickles and async sounds a *lot* harder, but I suspect it's a more robust
option, at least until we get a newer DB shipped with Python.

We could use another persistence mechanism (bsddb3, or even Zodb) but that
would have to be shipped with the code, which starts to raise nasty
packaging issues...

Paul.



More information about the Spambayes mailing list