Following up on my question regarding locking...
>Well... I'm less and less inclined to think that the update/insert failure
>is due to a lock problem. Weird! Perhaps the python/psycopg interface?
>I've never had a case where the python/psycopg invocation worked differently
>than with the psql interface. It may yet be time to build the psycopg from
>source, as the Debian version is pretty old...
The problem has been resolved... ok, I've been dense. A "minor" change resulted
in going down a slightly different pathway in a python function (that I wrote)
that handles certain python-database exchanges. This alternate pathway had
a bug (now expunged) that caused it to begin a transaction, but failed to
rollback or commit. Doh! This family of functions has been used for years
without finding the particular combination that excited this annoying behavior,
but it got me this time! Added another test to the regression set...
My apologies to Federico (psycopg), which was blameless, and my continued
thanks to the Postgresql community (particularly Michael and Tom) for their
help!
-frank