On Mon, Nov 11, 2019 at 2:12 PM Andres Freund wrote:
> Hm. I feel like there's plenty reasons to get a snapshot in extensions -
> there's plenty APIs one cannot really call without doing so?
Sure, I don't disagree with that.
> What I'm
> worried about is not primarily that GetSnapshotData() is b
Hi,
On 2019-09-06 10:25:22 -0400, Robert Haas wrote:
> On Thu, Sep 5, 2019 at 5:32 PM Andres Freund wrote:
> > On 2019-09-05 14:50:50 -0400, Robert Haas wrote:
> > > The best way that I've been able to come up with to enforce this rule
> > > after a little bit of thought is to add Assert(IsTransa
On Thu, Sep 5, 2019 at 5:32 PM Andres Freund wrote:
> On 2019-09-05 14:50:50 -0400, Robert Haas wrote:
> > The best way that I've been able to come up with to enforce this rule
> > after a little bit of thought is to add Assert(IsTransactionState())
> > to a bunch of functions in snapmgr.c, most n
Hi,
On 2019-09-05 14:50:50 -0400, Robert Haas wrote:
> The best way that I've been able to come up with to enforce this rule
> after a little bit of thought is to add Assert(IsTransactionState())
> to a bunch of functions in snapmgr.c, most notably
> GetTransactionSnapshot and GetCatalogSnapshot.
Way back in 2011, commit 57eb009092684e6e1788dd0dae641ccee1668b10
moved AbortTransaction's AtEOXact_Snapshot call to CleanupTransaction
to fix a problem when a ROLLBACK statement was prepared at the
protocol level and executed in a transaction with REPEATABLE READ or
higher isolation. RevalidateCac
