Hi,
On 2018-09-21 13:55:36 +0900, Michael Paquier wrote:
> On Thu, Sep 20, 2018 at 07:40:40PM -0400, Bruce Momjian wrote:
> > This surprised me since I expected a new timestamp after commit. Is
> > this something we want to change or document? Are there other
> > per-transaction behaviors we should adjust?
>
> I don't quite follow your argument here. clock_timestamp is known to be
> volatile, while the two others are stable, so its value can change
> within a transaction.
Isn't the point that transaction_timestamp() does *not* currently change
its value, even though the transaction (although not the outermost
statement) has finished?
I think Bruce has quite the point here.
Greetings,
Andres Freund