Re: Page Scan Mode in Hash Index

On Tue, Aug 22, 2017 at 7:24 PM, Ashutosh Sharma <ashu(dot)coek88(at)gmail(dot)com> wrote:
> On Tue, Aug 22, 2017 at 3:55 PM, Amit Kapila <amit(dot)kapila16(at)gmail(dot)com> wrote:
>
> Okay, I got your point now. I think, currently in _hash_kill_items(),
> if an overflow page is pinned we do not check if it got modified since
> the last read or
> not. Hence, if the vacuum runs on an overflow page that is pinned and
> also has some dead tuples in it then it could create a problem for
> scan basically,
> when scan would attempt to mark the killed items as dead. To get rid
> of such problem, i think, even if an overflow page is pinned we should
> check if it got
> modified or not since the last read was performed on the page. If yes,
> then do not allow scan to mark killed items as dead. Attached is the
> newer version with these changes along with some other cosmetic
> changes mentioned in your earlier email. Thanks.
>

Thanks for the new version. I again looked at the patches and fixed
quite a few comments in the code and ReadMe. You have forgotten to
update README for the changes in vacuum patch
(0003-Improve-locking-startegy-during-VACUUM-in-Hash-Index_v7). I
don't have anything more to add. If you are okay with changes, then
we can move it to Ready For Committer unless someone else has some
more comments.

--
With Regards,
Amit Kapila.
EnterpriseDB: http://www.enterprisedb.com