Comment 2 for bug 649247

I've reproduced this issue locally, and I think I know what's causing it
(and why it doesn't occur more often, which is basically luck.)

I think I'm going to ditch the separate process for now, and return to a
separate thread that interacts with the database for both reads and
writes. I thought of using a twisted server, but frankly I don't want
the extra dependency especially when it's not clear that it would reduce
complexity or help debuggability. Watch this space for news, hopefully soon.

eric casteleijn
Canonical Ltd.