On Sun, Oct 23, 2016 at 3:42 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> The business about not throwing a serialization failure due to actions
> of our own xact does seem worth fixing now, but if I understand correctly,
> we can deal with that by adding a test for xmin-is-our-own-xact into
> the existing code structure. I propose doing that much (and adding
> Munro's regression test case) and calling it good for today.
Thanks. This is the only part of it that I consider an actual
*bug* (since you can retry the serialization failure forever and
never move on because there is no other transaction involved which
can finish to clear the problem) as opposed to an opportunity to
optimize (by reducing false positive serialization failures
actually involving other transactions). I never saw anything in
the literature addressing the intersection of UPSERT and SSI, and I
think we do need to think about and discuss this a bit more before
we can be sure of the best fix. This is probably not thread on
which to have that discussion.
--
Kevin Grittner
EDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company