Re: Page Scan Mode in Hash Index

On Wed, Sep 20, 2017 at 4:04 PM, Robert Haas <robertmhaas(at)gmail(dot)com> wrote:
> On Tue, Sep 19, 2017 at 11:34 PM, Amit Kapila <amit(dot)kapila16(at)gmail(dot)com> wrote:
>> This point has been discussed above [1] and to avoid this problem we
>> are keeping the scan always behind vacuum for unlogged and temporary
>> tables as we are doing without this patch. That will ensure vacuum
>> won't be able to remove the TIDs which we are going to mark as dead.
>> This has been taken care in patch 0003. I understand that this is
>> slightly ugly, but the other alternative (as mentioned in the email
>> [1]) is much worse.
>
> Hmm. So if I understand correctly, you're saying that the LSN check
> in patch 0001 is actually completely unnecessary if we only apply
> 0001, but is needed in preparation for 0003, after which it will
> really be doing something?
>

Right.

> In more detail, I suppose the idea is: a TID cannot be reused until a
> VACUUM has intervened; VACUUM always visits every data page in the
> index; we won't allow a scan to pass VACUUM (and thus possibly have
> one of its TIDs get reused) except when the LSN check is actually
> sufficient to guarantee no TID reuse (i.e. table is not unlogged).

Right.

> Page-at-a-time index vacuum as in _hash_vacuum_one_page doesn't matter
> because such an operation doesn't allow TIDs to be reused.
>

Page-at-a-time index vacuum also allows TIDs to be reused but this is
done only for already marked dead items whereas vacuum can make the
non-dead entries to be removed. We don't have a problem with
page-at-a-time vacuum as it won't remove any items which the scan is
going to mark as dead.

> Did I get it right?
>

I think so.

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