On Tue, Nov 22, 2016 at 10:41 PM, Peter Geoghegan <pg@heroku.com> wrote:
> On Tue, Nov 22, 2016 at 7:39 PM, Robert Haas <robertmhaas@gmail.com> wrote:
>>> Heikki's been fooling with some ideas that I think have more promise.
>>> I wish he'd get to the point of presenting them publicly rather than
>>> just over beers at conferences.
>>
>> That would be good, too!
>
> I was told that TED was kind of formally proposed in the "MultiXact
> hindsight design" thread.
I read that, but:
1. Nothing's really come of that, AFAICS, and
2. It doesn't allow for in-place update, and
3. It doesn't eliminate the need for freezing.
Implementing TED would probably have been better than doing fkeylocks
the way we did, but now that we've mostly stabilized the multixact
work that was actually done, I have a hard time imagining that we
really want to revamp the whole thing without fixing any of the more
fundamental problems with our on-disk format. In my mind, the bloat
created by storing both the old and new versions of an updated tuple
in the heap, and the need to rewrite pages multiple times to set hint
bits, set all-visible, and ultimately freeze are the three-ton
gorillas in the corner. I am not arguing that what I've outlined here
is the only way to fix those problems or even the best way, just that
it isn't really worth the trouble of doing major surgery if we aren't
going to make a dent in those problems.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company