Use processed group clauses in PG16

Latest version of Postgres introduced an optimization
which removes redundant grouping and DISTINCT columns.
This optimization needs to be taken into account when
generating pushdown aggregation plans so we can create
valid plans with correct grouping columns.

(cherry picked from commit c81ae6f)

.unreleased/fix_6377

@@ -0,0 +1 @@
+Fixes: #6377 Use processed group clauses in PG16

src/planner/partialize.c

@@ -352,7 +352,11 @@ create_sorted_partial_agg_path(PlannerInfo *root, Path *path, PathTarget *target
 											   target,
 											   parse->groupClause ? AGG_SORTED : AGG_PLAIN,
 											   AGGSPLIT_INITIAL_SERIAL,
+#if PG16_LT
 											   parse->groupClause,
+#else
+											   root->processed_groupClause,
+#endif
 											   NIL,
 											   agg_partial_costs,
 											   d_num_groups);
@@ -367,8 +371,6 @@ static AggPath *
 create_hashed_partial_agg_path(PlannerInfo *root, Path *path, PathTarget *target,
 							   double d_num_groups, GroupPathExtraData *extra_data)
 {
-	Query *parse = root->parse;
-
 	/* Determine costs for aggregations */
 	AggClauseCosts *agg_partial_costs = &extra_data->agg_partial_costs;
 
@@ -378,7 +380,11 @@ create_hashed_partial_agg_path(PlannerInfo *root, Path *path, PathTarget *target
 										 target,
 										 AGG_HASHED,
 										 AGGSPLIT_INITIAL_SERIAL,
-										 parse->groupClause,
+#if PG16_LT
+										 root->parse->groupClause,
+#else
+										 root->processed_groupClause,
+#endif
 										 NIL,
 										 agg_partial_costs,
 										 d_num_groups);
@@ -890,7 +896,11 @@ ts_pushdown_partial_agg(PlannerInfo *root, Hypertable *ht, RelOptInfo *input_rel
 											  grouping_target,
 											  parse->groupClause ? AGG_SORTED : AGG_PLAIN,
 											  AGGSPLIT_FINAL_DESERIAL,
+#if PG16_LT
 											  parse->groupClause,
+#else
+											  root->processed_groupClause,
+#endif
 											  (List *) parse->havingQual,
 											  agg_final_costs,
 											  d_num_groups));
@@ -904,7 +914,11 @@ ts_pushdown_partial_agg(PlannerInfo *root, Hypertable *ht, RelOptInfo *input_rel
 											  grouping_target,
 											  AGG_HASHED,
 											  AGGSPLIT_FINAL_DESERIAL,
+#if PG16_LT
 											  parse->groupClause,
+#else
+											  root->processed_groupClause,
+#endif
 											  (List *) parse->havingQual,
 											  agg_final_costs,
 											  d_num_groups));

test/sql/.gitignore

@@ -9,6 +9,7 @@
 /insert_many-*.sql
 /parallel-*.sql
 /partitioning-*.sql
+/partitionwise-*.sql
 /plan_expand_hypertable-*.sql
 /plan_hashagg-*.sql
 /plan_hashagg_optimized-*.sql

test/sql/CMakeLists.txt

@@ -35,7 +35,6 @@ set(TEST_FILES
     null_exclusion.sql
     partition.sql
     partitioning.sql
-    partitionwise.sql
     pg_dump_unprivileged.sql
     pg_join.sql
     plain.sql
@@ -68,6 +67,7 @@ set(TEST_TEMPLATES
     rowsecurity.sql.in
     update.sql.in
     parallel.sql.in
+    partitionwise.sql.in
     plan_expand_hypertable.sql.in
     plan_ordered_append.sql.in
     query.sql.in

test/sql/partitionwise.sql → test/sql/partitionwise.sql.in

@@ -273,3 +273,12 @@ FROM hyper_timepart
 GROUP BY 1, 2
 ORDER BY 1, 2
 LIMIT 10;
+
+-- Test removal of redundant group key optimization in PG16
+-- All lower versions include the redundant key on device column
+:PREFIX
+SELECT device, avg(temp)
+FROM hyper_timepart
+WHERE device = 1
+GROUP BY 1
+LIMIT 10;

tsl/test/expected/merge_append_partially_compressed-16.out

@@ -702,7 +702,7 @@ SELECT x1, x2, max(time) FROM test1 GROUP BY x1, x2, time ORDER BY time limit 10
 ---------------------------------------------------------------------------------------------------------------
  Limit (actual rows=5 loops=1)
    ->  Finalize GroupAggregate (actual rows=5 loops=1)
-         Group Key: test1.x1, test1.x2, test1."time"
+         Group Key: test1."time", test1.x1, test1.x2
          ->  Custom Scan (ChunkAppend) on test1 (actual rows=5 loops=1)
                Order: test1."time", test1.x1, test1.x2
                ->  Merge Append (actual rows=5 loops=1)
@@ -711,15 +711,15 @@ SELECT x1, x2, max(time) FROM test1 GROUP BY x1, x2, time ORDER BY time limit 10
                            Sort Key: _hyper_3_7_chunk."time", _hyper_3_7_chunk.x1, _hyper_3_7_chunk.x2
                            Sort Method: quicksort 
                            ->  Partial HashAggregate (actual rows=4 loops=1)
-                                 Group Key: _hyper_3_7_chunk.x1, _hyper_3_7_chunk.x2, _hyper_3_7_chunk."time"
+                                 Group Key: _hyper_3_7_chunk."time", _hyper_3_7_chunk.x1, _hyper_3_7_chunk.x2
                                  Batches: 1 
                                  ->  Custom Scan (DecompressChunk) on _hyper_3_7_chunk (actual rows=4 loops=1)
                                        ->  Seq Scan on compress_hyper_4_8_chunk (actual rows=3 loops=1)
                      ->  Sort (actual rows=1 loops=1)
                            Sort Key: _hyper_3_7_chunk."time", _hyper_3_7_chunk.x1, _hyper_3_7_chunk.x2
                            Sort Method: quicksort 
                            ->  Partial HashAggregate (actual rows=1 loops=1)
-                                 Group Key: _hyper_3_7_chunk.x1, _hyper_3_7_chunk.x2, _hyper_3_7_chunk."time"
+                                 Group Key: _hyper_3_7_chunk."time", _hyper_3_7_chunk.x1, _hyper_3_7_chunk.x2
                                  Batches: 1 
                                  ->  Seq Scan on _hyper_3_7_chunk (actual rows=1 loops=1)
 (22 rows)