Re: TB-sized databases

On Thu, 6 Dec 2007, Tom Lane wrote:
> Matthew <matthew(at)flymine(dot)org> writes:
> > ... For this query, Postgres would perform a nested loop,
> > iterating over all rows in the small table, and doing a hundred index
> > lookups in the big table. This completed very quickly. However, adding the
> > LIMIT meant that suddenly a merge join was very attractive to the planner,
> > as it estimated the first row to be returned within milliseconds, without
> > needing to sort either table.
>
> > The problem is that Postgres didn't know that the first hit in the big
> > table would be about half-way through, after doing a index sequential scan
> > for half a bazillion rows.
>
> Hmm. IIRC, there are smarts in there about whether a mergejoin can
> terminate early because of disparate ranges of the two join variables.
> Seems like it should be straightforward to fix it to also consider
> whether the time-to-return-first-row will be bloated because of
> disparate ranges. I'll take a look --- but it's probably too late
> to consider this for 8.3.

Very cool. Would that be a planner cost estimate fix (so it avoids the
merge join), or a query execution fix (so it does the merge join on the
table subset)?

Matthew

--
I've run DOOM more in the last few days than I have the last few
months. I just love debugging ;-) -- Linus Torvalds