Skip to content

Commit af3deff

Browse files
anarazelanarazel
committed
Fix slot type handling for Agg nodes performing internal sorts.
Since 15d8f83 we assert that - and since 7ef04e4, 4da597e rely on - the slot type for an expression's ecxt_{outer,inner,scan}tuple not changing, unless explicitly flagged as such. That allows to either skip deforming (for a virtual tuple slot) or optimize the code for JIT accelerated deforming appropriately (for other known slot types). This assumption was sometimes violated for grouping sets, when nodeAgg.c internally uses tuplesorts, and the child node doesn't return a TTSOpsMinimalTuple type slot. Detect that case, and flag that the outer slot might not be "fixed". It's probably worthwhile to optimize this further in the future, and more granularly determine whether the slot is fixed. As we already instantiate per-phase transition and equal expressions, we could cheaply set the slot type appropriately for each phase. But that's a separate change from this bugfix. This commit does include a very minor optimization by avoiding to create a slot for handling tuplesorts, if no such sorts are performed. Previously we created that slot unnecessarily in the common case of computing all grouping sets via hashing. The code looked too confusing without that, as the conditions for needing a sort slot and flagging that the slot type isn't fixed, are the same. Reported-By: Ashutosh Sharma Author: Andres Freund Discussion: https://postgr.es/m/CAE9k0PmNaMD2oHTEAhRyxnxpaDaYkuBYkLa1dpOpn=RS0iS2AQ@mail.gmail.com Backpatch: 12-, where the bug was introduced in 15d8f83
1 parent cb9bb15 commit af3deff

File tree

3 files changed

+99
-2
lines changed

3 files changed

+99
-2
lines changed

‎src/backend/executor/nodeAgg.c

Lines changed: 32 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -2231,13 +2231,43 @@ ExecInitAgg(Agg *node, EState *estate, int eflags)
22312231
/*
22322232
* initialize source tuple type.
22332233
*/
2234+
aggstate->ss.ps.outerops =
2235+
ExecGetResultSlotOps(outerPlanState(&aggstate->ss),
2236+
&aggstate->ss.ps.outeropsfixed);
2237+
aggstate->ss.ps.outeropsset = true;
2238+
22342239
ExecCreateScanSlotFromOuterPlan(estate, &aggstate->ss,
2235-
ExecGetResultSlotOps(outerPlanState(&aggstate->ss), NULL));
2240+
aggstate->ss.ps.outerops);
22362241
scanDesc = aggstate->ss.ss_ScanTupleSlot->tts_tupleDescriptor;
2237-
if (node->chain)
2242+
2243+
/*
2244+
* If there are more than two phases (including a potential dummy phase
2245+
* 0), input will be resorted using tuplesort. Need a slot for that.
2246+
*/
2247+
if (numPhases > 2)
2248+
{
22382249
aggstate->sort_slot = ExecInitExtraTupleSlot(estate, scanDesc,
22392250
&TTSOpsMinimalTuple);
22402251

2252+
/*
2253+
* The output of the tuplesort, and the output from the outer child
2254+
* might not use the same type of slot. In most cases the child will
2255+
* be a Sort, and thus return a TTSOpsMinimalTuple type slot - but the
2256+
* input can also be be presorted due an index, in which case it could
2257+
* be a different type of slot.
2258+
*
2259+
* XXX: For efficiency it would be good to instead/additionally
2260+
* generate expressions with corresponding settings of outerops* for
2261+
* the individual phases - deforming is often a bottleneck for
2262+
* aggregations with lots of rows per group. If there's multiple
2263+
* sorts, we know that all but the first use TTSOpsMinimalTuple (via
2264+
* the nodeAgg.c internal tuplesort).
2265+
*/
2266+
if (aggstate->ss.ps.outeropsfixed &&
2267+
aggstate->ss.ps.outerops != &TTSOpsMinimalTuple)
2268+
aggstate->ss.ps.outeropsfixed = false;
2269+
}
2270+
22412271
/*
22422272
* Initialize result type, slot and projection.
22432273
*/

‎src/test/regress/expected/groupingsets.out

Lines changed: 55 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1360,6 +1360,61 @@ explain (costs off)
13601360
-> Function Scan on gstest_data
13611361
(10 rows)
13621362

1363+
-- Verify that we correctly handle the child node returning a
1364+
-- non-minimal slot, which happens if the input is pre-sorted,
1365+
-- e.g. due to an index scan.
1366+
BEGIN;
1367+
SET LOCAL enable_hashagg = false;
1368+
EXPLAIN SELECT a, b, count(*), max(a), max(b) FROM gstest3 GROUP BY GROUPING SETS(a, b,()) ORDER BY a, b;
1369+
QUERY PLAN
1370+
-------------------------------------------------------------------------
1371+
Sort (cost=1.20..1.21 rows=5 width=24)
1372+
Sort Key: a, b
1373+
-> GroupAggregate (cost=1.03..1.14 rows=5 width=24)
1374+
Group Key: a
1375+
Group Key: ()
1376+
Sort Key: b
1377+
Group Key: b
1378+
-> Sort (cost=1.03..1.03 rows=2 width=8)
1379+
Sort Key: a
1380+
-> Seq Scan on gstest3 (cost=0.00..1.02 rows=2 width=8)
1381+
(10 rows)
1382+
1383+
SELECT a, b, count(*), max(a), max(b) FROM gstest3 GROUP BY GROUPING SETS(a, b,()) ORDER BY a, b;
1384+
a | b | count | max | max
1385+
---+---+-------+-----+-----
1386+
1 | | 1 | 1 | 1
1387+
2 | | 1 | 2 | 2
1388+
| 1 | 1 | 1 | 1
1389+
| 2 | 1 | 2 | 2
1390+
| | 2 | 2 | 2
1391+
(5 rows)
1392+
1393+
SET LOCAL enable_seqscan = false;
1394+
EXPLAIN SELECT a, b, count(*), max(a), max(b) FROM gstest3 GROUP BY GROUPING SETS(a, b,()) ORDER BY a, b;
1395+
QUERY PLAN
1396+
-----------------------------------------------------------------------------------------
1397+
Sort (cost=12.32..12.33 rows=5 width=24)
1398+
Sort Key: a, b
1399+
-> GroupAggregate (cost=0.13..12.26 rows=5 width=24)
1400+
Group Key: a
1401+
Group Key: ()
1402+
Sort Key: b
1403+
Group Key: b
1404+
-> Index Scan using gstest3_pkey on gstest3 (cost=0.13..12.16 rows=2 width=8)
1405+
(8 rows)
1406+
1407+
SELECT a, b, count(*), max(a), max(b) FROM gstest3 GROUP BY GROUPING SETS(a, b,()) ORDER BY a, b;
1408+
a | b | count | max | max
1409+
---+---+-------+-----+-----
1410+
1 | | 1 | 1 | 1
1411+
2 | | 1 | 2 | 2
1412+
| 1 | 1 | 1 | 1
1413+
| 2 | 1 | 2 | 2
1414+
| | 2 | 2 | 2
1415+
(5 rows)
1416+
1417+
COMMIT;
13631418
-- More rescan tests
13641419
select * from (values (1),(2)) v(a) left join lateral (select v.a, four, ten, count(*) from onek group by cube(four,ten)) s on true order by v.a,four,ten;
13651420
a | a | four | ten | count

‎src/test/regress/sql/groupingsets.sql

Lines changed: 12 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -384,6 +384,18 @@ explain (costs off)
384384
from (values (1),(2)) v(x), gstest_data(v.x)
385385
group by cube (a,b) order by a,b;
386386

387+
-- Verify that we correctly handle the child node returning a
388+
-- non-minimal slot, which happens if the input is pre-sorted,
389+
-- e.g. due to an index scan.
390+
BEGIN;
391+
SET LOCAL enable_hashagg = false;
392+
EXPLAIN SELECT a, b, count(*), max(a), max(b) FROM gstest3 GROUP BY GROUPING SETS(a, b,()) ORDER BY a, b;
393+
SELECT a, b, count(*), max(a), max(b) FROM gstest3 GROUP BY GROUPING SETS(a, b,()) ORDER BY a, b;
394+
SET LOCAL enable_seqscan = false;
395+
EXPLAIN SELECT a, b, count(*), max(a), max(b) FROM gstest3 GROUP BY GROUPING SETS(a, b,()) ORDER BY a, b;
396+
SELECT a, b, count(*), max(a), max(b) FROM gstest3 GROUP BY GROUPING SETS(a, b,()) ORDER BY a, b;
397+
COMMIT;
398+
387399
-- More rescan tests
388400
select * from (values (1),(2)) v(a) left join lateral (select v.a, four, ten, count(*) from onek group by cube(four,ten)) s on true order by v.a,four,ten;
389401
select array(select row(v.a,s1.*) from (select two,four, count(*) from onek group by cube(two,four) order by two,four) s1) from (values (1),(2)) v(a);

0 commit comments

Comments
 (0)