On 9 March 2017 at 11:17, Craig Ringer <craig@2ndquadrant.com> wrote:
> Hi all
>
> I've found a minor off-by-one error in the resource retention logic
> for logical slots, where we treat confirmed_flush as meaning "flushed
> up to and including this LSN". Seems reasonable, but the rest of the
> code treats it as "flushed up to but excluding this LSN". In
> particular, we treat confirmed_flush as the inclusive startpoint for a
> new logical decoding session if no startpoint is provided by the
> client and will replay a change whose commit record begins at exactly
> confirmed_flush to the client.
This should be backpatched to 9.4.
The others do not need backpatching.
-- Craig Ringer http://www.2ndQuadrant.com/PostgreSQL Development, 24x7 Support, Training & Services