Re: Functional dependency in GROUP BY through JOINs

On 6 December 2012 17:21, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> Simon Riggs <simon(at)2ndQuadrant(dot)com> writes:
>> On 5 December 2012 23:37, David Rowley <dgrowleyml(at)gmail(dot)com> wrote:
>>> Though this plan might not be quite as optimal as it could be as it performs
>>> the grouping after the join.
>
>> PostgreSQL always calculates aggregation as the last step.
>
>> It's a well known optimisation to push-down GROUP BY clauses to the
>> lowest level, but we don't do that, yet.
>
>> You're right that it can make a massive difference to many queries.
>
> In the case being presented here, it's not apparent to me that there's
> any advantage to be had at all. You still need to aggregate over the
> rows joining to each uniquely-keyed row. So how exactly are you going
> to "push down the GROUP BY", and where does the savings come from?

David presents SQL that shows how that is possible.

In terms of operators, after push down we aggregate 1 million rows and
then join 450. Which seems cheaper than join 1 million rows and
aggregate 1 million. So we're passing nearly 1 million fewer rows into
the join.

--
Simon Riggs http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services