Re: Push down Aggregates below joins

From: Antonin Houska <ah(at)cybertec(dot)at>
To: Heikki Linnakangas <hlinnaka(at)iki(dot)fi>
Cc: Tomas Vondra <tomas(dot)vondra(at)2ndquadrant(dot)com>, pgsql-hackers <pgsql-hackers(at)postgresql(dot)org>
Subject: Re: Push down Aggregates below joins
Date: 2018-06-22 07:16:51
Message-ID: 8281.1529651811@localhost
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

Heikki Linnakangas <hlinnaka(at)iki(dot)fi> wrote:

> Ah, cool! I missed that thread earlier. Yes, seems like we've been hacking on
> the same feature. Let's compare!
>
> I've been using this paper as a guide:
>
> "Including Group-By in Query Optimization", by Surajit Chaudhuri and Kyuseok
> Shim:
> https://pdfs.semanticscholar.org/3079/5447cec18753254edbbd7839f0afa58b2a39.pdf

> Using the terms from that paper, my patch does only "Invariant Grouping
> transormation", while yours can do "Simple Coalescing Grouping", which is more
> general. In layman terms, my patch can push the Aggregate below a join, while
> your patch can also split an Aggregate so that you do a partial aggregate
> below the join, and a final stage above it.

Thanks for the link. I've just checked the two approaches briefly so far.

> My thinking was to start with the simpler Invariant Grouping transformation
> first, and do the more advanced splitting into partial aggregates later, as
> a separate patch.

I think for this you need to make sure that no join duplicates already
processed groups. I tried to implement PoC of "unique keys" in v5 of my patch
[1], see 09_avoid_agg_finalization.diff. The point is that the final
aggregation can be replaced by calling pg_aggregate(aggfinalfn) function if
the final join generates only unique values of the GROUP BY expression.

Eventually I considered this an additional optimization of my approach and
postponed work on this to later, but I think something like this would be
necessary for your approach. As soon as you find out that a grouped relation
is joined to another relation in a way that duplicates the grouping
expression, you cannot proceed in creating grouped path.

> There's some difference in the way we're dealing with the grouped
> RelOptInfos. You're storing them completely separately, in PlannerInfo's new
> 'simple_grouped_rel_array' array and 'join_grouped_rel_list/hash'. I'm
> attaching each grouped RelOptInfo to its parent.

In the first version of my patch I added several fields to RelOptInfo
(reltarget for the grouped paths, separate pathlist, etc.) but some people
didn't like it. In the later versions I introduced a separate RelOptInfo for
grouped relations, but stored it in a separate list. Your approach might make
the patch a bit less invasive, i.e. we don't have to add those RelOptInfo
lists / arrays to standard_join_search() and subroutines.

> I got away with much less code churn in allpaths.c, indxpath.c,
> joinpath.c. You're adding new 'do_aggregate' flags to many functions. I'm not
> sure if you needed that because you do partial aggregation and I don't, but it
> would be nice to avoid it.

IIRC, the do_aggregate argument determines how the grouped join should be
created. If it's false, the planner joins a grouped relation (if it exists) to
non-grouped one. If it's true, it joins two non-grouped relations and applies
(partial) aggregation to the result.

> You're introducing a new GroupedVar expression to represent Aggrefs between
> the partial and final aggregates, while I'm just using Aggref directly, above
> the aggregate node. I'm not thrilled about introducing an new Var-like
> expression. We already have PlaceHolderVars, and I'm always confused on how
> those work. But maybe that's necessary to supprot partial aggregation?

The similarity of GroupedVar and PlaceHolderVar is that they are evaluated at
some place in the join tree and the result is only passed to the joins above
and eventually to the query target, w/o being evaluated again. In contrast,
generic expressions are evaluated in the query target (only the input Vars get
propagated from lower nodes), but that's not what we want for 2-stage
aggregation.

In my patch GroupedVar represents either the result of partial aggregation
(the value of the transient state) or a grouping expression which is more
complex than a plain column reference (Var expression).

> In the other thread, Robert Haas wrote:
> > Concretely, in your test query "SELECT p.i, avg(c1.v) FROM
> > agg_pushdown_parent AS p JOIN agg_pushdown_child1 AS c1 ON c1.parent =
> > p.i GROUP BY p.i" you assume that it's OK to do a Partial
> > HashAggregate over c1.parent rather than p.i. This will be false if,
> > say, c1.parent is of type citext and p.i is of type text; this will
> > get grouped together that shouldn't. It will also be false if the
> > grouping expression is something like GROUP BY length(p.i::text),
> > because one value could be '0'::numeric and the other '0.00'::numeric.
> > I can't think of a reason why it would be false if the grouping
> > expressions are both simple Vars of the same underlying data type, but
> > I'm a little nervous that I might be wrong even about that case.
> > Maybe you've handled all of this somehow, but it's not obvious to me
> > that it has been considered.
>
> Ah, I made the same mistake. I did consider the "GROUP BY length(o.i::text)",
> but not the cross-datatype case. I think we should punt on that for now, and
> only do the substitution for simple Vars of the same datatype. That seems safe
> to me.

Yes, I reached the same conclusion. I'll add this restriction to the next
version of the patch.

> Overall, your patch is in a more polished state than my prototype.

Probably I spent much more time on it.

> For easier review, though, I think we should try to get something smaller
> committed first, and build on that. Let's punt on the Var substitution.

As mentioned above, I think we can only live without the Var substitution (in
other words without 2-stage aggregation) if we can check the uniqueness of
grouping keys of any path. So the question is how much effort this check
requires.

> And I'd suggest adopting my approach of attaching the grouped RelOptInfos to
> the parent RelOptInfo, that seems simpler.

o.k., I'll try this in the next version.

> And if we punt on the partial aggregation, and only allow pushing down the
> whole Aggregate, how much smaller would your patch get?

I can't tell now, need to spend some time looking at the code.

> (I'll be on vacation for the next two weeks, but I'll be following this
> thread. After that, I plan to focus on this feature, as time from reviewing
> patches in the commitfest permits.)

Likewise, I'll be off from July 5th to 22nd. I'll post what I have before I
leave and will see what you could make out of it :-)

It's cool that you intend to work on this feature too!

[1] https://www.postgresql.org/message-id/18007.1513957437%40localhost

--
Antonin Houska
Cybertec Schönig & Schönig GmbH
Gröhrmühlgasse 26, A-2700 Wiener Neustadt
Web: https://www.cybertec-postgresql.com

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Konstantin Knizhnik 2018-06-22 07:18:12 Re: libpq compression
Previous Message Konstantin Knizhnik 2018-06-22 07:11:58 Re: libpq compression