Re: [HACKERS] Hint Bits and Write I/O

On Sat, 2008-08-02 at 00:24 -0400, Tom Lane wrote:
> Alvaro Herrera <alvherre(at)commandprompt(dot)com> writes:
> > I think it makes sense to commit this patch now, per previous
> > discussions on which we have agreed to make incremental changes.
>
> Yeah, but at the same time there is merit in the argument that the
> proposed patch hasn't actually been proven to be usable for anything.
> I would be a lot happier if there were even a trivial proof-of-concept
> plugin example submitted with it, just to prove that there were no
> showstopper problems in the plugin design, like failure to pass
> essential information or not getting the locking straight.

Plugins were my other patch. I did originally submit a version with
changes, but this patch was specifically a version with *no* external
behaviour changes, to form a base from which various people's ideas
might be explored.

> > I'm just wondering if the change of usage_count from 16 to 8 bits was
> > discussed and agreed?
>
> Umm ... it was not, but given that we have logic in there to limit the
> usage_count to 5 or so, it's hard to argue that there's a big problem.

It was discussed and it was Tom's suggestion to do this. I agreed!

> I confess to not having read the patch in detail --- where did the other
> 8 bits go to?

Keeping track of the number of hints set on a block since last write.

--
Simon Riggs www.2ndQuadrant.com
PostgreSQL Training, Services and Support