"Kevin Grittner" <Kevin.Grittner@wicourts.gov> wrote:
> (1) A read write transaction might need to be canceled to
> prevent the view of the data a committed read only transaction has
> already seen from becoming inconsistent. (Dan's example)
And this one seems entirely a theoretical possibility. I spent a
little time looking it over, and I don't see how it could be made to
work from hot standbys without an unbounded flow of predicate lock
information from all standbys to the master *plus* blocking commits
on the master for the duration of the longest round trip latency to
any standby. I think we can call this one dead on arrival.
-Kevin