On Thu, Nov 17, 2011 at 11:56 PM, Maxim Boguk <maxim.boguk at gmail.com> wrote:
> PGRES_FATAL_ERROR ERROR:  could not serialize access due to read/write
> dependencies among transactions
> DETAIL:  Reason code: Canceled on identification as a pivot, during write.
> HINT:  The transaction might succeed if retried.
> In that case slony just terminate without commiting batch (and
> returned to the life only by watchdog).

That seems like one of the sorts of cases that we could expect there
being with the new "full serialization" support in 9.1.

Terminating the batch is not notably disastrous; a retry should work fine.

The case that would be troublesome would be if you were trying to set
up a subscription, and get serialization interruptions often enough
that you can't get the subscription to "take."

But this seems like the sort of thing that you have to be prepared for
when you adopt full serialization support.

My inclination is actually to do a bit of a "push back"; I don't think
it's quite right that Postgres is responding, in this case, by
indicating that this is a FATAL error.  If we can expect a retry to
work, that surely isn't FATAL, is it?