On Tue, Jul 5, 2016 at 8:51 PM, Robert Haas <robertmhaas@gmail.com> wrote:
>
> So I'm a bit confused about what you are saying here. If the page is
> marked all-frozen but actually isn't all-frozen, then we should clear
> the all-frozen bit in the VM.
>
Agreed.
> The easiest way to do that is to clear
> both bits in the VM plus the page-level bit, as done here, because we
> don't presently have a way of clearing just one of the visibility map
> bits.
>
Okay, but due to that we are clearing the visibility information
(all-visible) even though we should not clear it based on all-frozen.
I don't know if there is much harm even if we do the way it is
proposed in patch, but why not improve it if we can.
> Now, since the heap_lock_tuple issue requires us to introduce a new
> method to clear all-visible without clearing all-frozen, we could
> possibly use that here too, once we have it.
>
makes sense.
--
With Regards,
Amit Kapila.
EnterpriseDB: http://www.enterprisedb.com