On Mon, 2007-03-19 at 16:06 +0530, Pavan Deolasee wrote:
> Simon Riggs wrote:
> >
> > This problem is solved by moving the wait (for all transactions in
> > reference snapshot to finish) so that it is now between the first and
> > second scans, as described.
> >
> > During the second Vscan we would prune each block, so the only remaining
> > tuple in the block when the second scan sees it would be (10,30) and it
> > would no longer be a HOT tuple - the index would have a pointer to it,
> > so no new index pointer would be added. The pointer to (10,30) is the
> > same pointer that was added in the first phase for the tuple (10,20).
> >
>
> The problem is that in the first phase, the pointer was inserted
> with key=20 whereas now its changed to 30. So we need to delete the old
> index entry and add a new one.
So don't index HOT tuples in the first phase, wait until the second.
That should be just a single if() test in IndexBuildHeapScan().
-- Simon Riggs EnterpriseDB http://www.enterprisedb.com