On Wed, Dec 04, 2024 at 09:55:43AM +0200, Heikki Linnakangas wrote:
> On 04/12/2024 03:24, Tom Lane wrote:
>> Andres Freund <andres@anarazel.de> writes:
>> > Most of this I agree with. But I'm not sure just removing the toast snapshot
>> > stuff is good - we've had a bunch of bugs where we don't hold a snapshot for
>> > long enough to actually ensure that toast tuples stay alive.
>>
>> Yeah, the stuff concerned with toast snapshots has nothing to do
>> with that old user-visible feature. It's to keep us from writing
>> incorrect code, and it's still (very) needed.
>
> Right. Here's a new attempt that keeps that check.
Looks reasonable to me. One idea I had is to make SnapshotToast a macro
that first does an AssertMacro(HaveRegisteredOrActiveSnapshot()), but the
approach in the patch seems fine, too.
--
nathan
