Paul Jones <pbj(at)cmicdo(dot)com> writes:
> On Mon, Jan 11, 2016 at 10:04:16AM -0500, Tom Lane wrote:
>> It looks like pg_upgrade tries to load all libraries from functions in
>> any database in the old cluster into a single session in the new cluster,
>> which will fail in a scenario like this even if you keep python2 and
>> python3 functions rigorously separated into distinct databases. I'm
>> not sure if we could weaken that test enough to work.
> I guess the thing to do is to manually pg_dump the databases that have
> python, drop them, upgrade, and then manually restore.
If you'd rather build a patched version of PG, I have posted a
work-in-progress patch to address this issue:
http://www.postgresql.org/message-id/31659.1452538805@sss.pgh.pa.us
regards, tom lane