Hi,
On 2023-11-09 10:25:18 +0900, Michael Paquier wrote:
> On Thu, Nov 09, 2023 at 10:10:39AM +0900, torikoshia wrote:
> > I am a little concerned about that the reset time is not the same and that
> > GetCurrentTimestamp() is called multiple times, but I think it would be
> > acceptable because the function is probably not used that often and the
> > reset time is not atomic in practice.
>
> Arf, right. I misremembered that this is just a clock_timestamp() so
> that's not transaction-resilient. Anyway, my take is that this is not
> a big deal in practice compared to the usability of the wrapper.
It seems inconsequential cost-wise. Resetting stats is way more expensive that
a few timestamp determinations. Correctness wise it actually seems *better* to
record the timestamps more granularly, after all, that moves them closer to
the time the individual kind of stats is reset.
Greetings,
Andres Freund