Re: Write Ahead Logging for Hash Indexes

On Thu, Dec 1, 2016 at 6:51 PM, Amit Kapila <amit.kapila16@gmail.com> wrote:
>> Thanks.  I am thinking that it might make sense to try to get the
>> "microvacuum support for hash index" and "cache hash index meta page"
>> patches committed before this one, because I'm guessing they are much
>> simpler than this one, and I think therefore that the review of those
>> patches can probably move fairly quickly.
>
> I think it makes sense to move "cache hash index meta page" first,
> however "microvacuum support for hash index" is based on WAL patch as
> the action in this patch (delete op) also needs to be logged.  One
> idea could be that we can try to split the patch so that WAL logging
> can be done as a separate patch, but I am not sure if it is worth.

The thing is, there's a fair amount locking stupidity in what just got
committed because of the requirement that the TID doesn't decrease
within a page.  I'd really like to get that fixed.

>>  Of course, ideally I can
>> also start reviewing this one in the meantime.  Does that make sense
>> to you?
>>
>
> You can start reviewing some of the operations like "Create Index",
> "Insert".  However, some changes are required because of change in
> locking strategy for Vacuum.  I am planning to work on rebasing it and
> making required changes in next week.

I'll review after that, since I have other things to review meanwhile.

-- 
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company