ORCA now mimics planner when it comes to empty stats

When there is no stats available for any table, ORCA was treating it as an
empty table while planning. On the other hand planner is utilizing a guc
`gp_enable_relsize_collection` to obtain the estimated size of the table, but
no other statistics. This commit enables ORCA to have the same behavior as
planner when the guc is set.
Signed-off-by: NSambitesh Dash <sdash@pivotal.io>

src/backend/gpopt/gpdbwrappers.cpp

@@ -2449,6 +2449,24 @@ gpdb::EstimateRelationSize
 	GP_WRAP_END;
 }
 
+void
+gpdb::CdbEstimateRelationSize
+	(
+	RelOptInfo   *relOptInfo,
+	Relation rel,
+	int32 *attr_widths,
+	BlockNumber *pages,
+	double *tuples
+	)
+{
+	GP_WRAP_START;
+	{
+		cdb_estimate_rel_size(relOptInfo, rel, rel, attr_widths, pages, tuples);
+		return;
+	}
+	GP_WRAP_END;
+}
+
 void
 gpdb::CloseRelation
 	(

src/backend/gpopt/translate/CTranslatorRelcacheToDXL.cpp

@@ -2288,6 +2288,16 @@ CTranslatorRelcacheToDXL::PimdobjRelStats
 		else
 		{
 			rows = rel->rd_rel->reltuples;
+
+			if (rows == 0 && gp_enable_relsize_collection)
+			{
+				RelOptInfo *relOptInfo = makeNode(RelOptInfo);
+				relOptInfo->cdbpolicy = gpdb::Pdistrpolicy(rel);
+
+				gpdb::CdbEstimateRelationSize(relOptInfo, rel, NULL, &pages, &rows);
+				pfree(relOptInfo);
+
+			}
 		}
 
 		pmdidRelStats->AddRef();

src/backend/optimizer/util/plancat.c

@@ -59,14 +59,6 @@ static List *get_relation_constraints(PlannerInfo *root,
 						 Oid relationObjectId, RelOptInfo *rel,
 						 bool include_notnull);
 
-static void
-cdb_estimate_rel_size(RelOptInfo   *relOptInfo,
-                      Relation      baserel,
-                      Relation      rel,
-                      int32        *attr_widths,
-				      BlockNumber  *pages,
-                      double       *tuples);
-
 static void get_external_relation_info(Relation relation, RelOptInfo *rel);
 
 

src/include/gpopt/gpdbwrappers.h

@@ -526,6 +526,7 @@ namespace gpdb {
 
 	// estimate the relation size using the real number of blocks and tuple density
 	void EstimateRelationSize(Relation rel,	int32 *attr_widths,	BlockNumber *pages,	double *tuples);
+	void CdbEstimateRelationSize (RelOptInfo *relOptInfo, Relation rel, int32 *attr_widths, BlockNumber *pages, double *tuples);
 
 	// close the given relation
 	void CloseRelation(Relation rel);

src/include/optimizer/plancat.h

@@ -33,6 +33,13 @@ extern void get_relation_info(PlannerInfo *root, Oid relationObjectId,
 extern void estimate_rel_size(Relation rel, int32 *attr_widths,
 				  BlockNumber *pages, double *tuples);
 
+extern void cdb_estimate_rel_size(RelOptInfo   *relOptInfo,
+								  Relation      baserel,
+								  Relation      rel,
+								  int32        *attr_widths,
+								  BlockNumber  *pages,
+								  double       *tuples);
+
 extern int32 get_relation_data_width(Oid relid, int32 *attr_widths);
 
 extern bool relation_excluded_by_constraints(PlannerInfo *root,

src/test/regress/expected/qp_misc_jiras.out

@@ -4569,6 +4569,42 @@ select test();
  
 (1 row)
 
+--
+-- Test gp_enable_relsize_collection's effect on ORCA plan generation
+--
+create table tbl_z(x int) distributed by (x);
+set optimizer_metadata_caching to off;
+insert into tbl_z select i from generate_series(1,100) i;
+-- plan with no relsize collection
+explain select 1 as t1 where 1 <= ALL (select x from tbl_z);
+                                                    QUERY PLAN                                                    
+------------------------------------------------------------------------------------------------------------------
+ Result  (cost=10000000651.88..10000000651.89 rows=1 width=0)
+   One-Time Filter: (SubPlan 1)
+   SubPlan 1  (slice0)
+     ->  Materialize  (cost=10000000000.00..10000001544.50 rows=32100 width=4)
+           ->  Gather Motion 3:1  (slice1; segments: 3)  (cost=10000000000.00..10000001063.00 rows=96300 width=4)
+                 ->  Seq Scan on tbl_z  (cost=10000000000.00..10000001063.00 rows=32100 width=4)
+ Optimizer: legacy query optimizer
+(7 rows)
+
+set gp_enable_relsize_collection = on;
+-- plan with relsize collection
+explain select 1 as t1 where 1 <= ALL (select x from tbl_z);
+                                                   QUERY PLAN                                                   
+----------------------------------------------------------------------------------------------------------------
+ Result  (cost=10000000006.52..10000000006.53 rows=1 width=0)
+   One-Time Filter: (SubPlan 1)
+   SubPlan 1  (slice0)
+     ->  Materialize  (cost=10000000000.00..10000000015.44 rows=321 width=4)
+           ->  Gather Motion 3:1  (slice1; segments: 3)  (cost=10000000000.00..10000000010.63 rows=963 width=4)
+                 ->  Seq Scan on tbl_z  (cost=10000000000.00..10000000010.63 rows=321 width=4)
+ Optimizer: legacy query optimizer
+(7 rows)
+
+drop table if exists tbl_z;
+reset optimizer_metadata_caching;
+reset gp_enable_relsize_collection;
 -- start_ignore
 drop schema qp_misc_jiras cascade;
 NOTICE:  drop cascades to table qp_misc_jiras._tbl10050_test

src/test/regress/expected/qp_misc_jiras_optimizer.out

@@ -4607,6 +4607,50 @@ select test();
  
 (1 row)
 
+--
+-- Test gp_enable_relsize_collection's effect on ORCA plan generation
+--
+create table tbl_z(x int) distributed by (x);
+set optimizer_metadata_caching to off;
+insert into tbl_z select i from generate_series(1,100) i;
+-- plan with no relsize collection
+explain select 1 as t1 where 1 <= ALL (select x from tbl_z);
+                                           QUERY PLAN                                           
+------------------------------------------------------------------------------------------------
+ Result  (cost=0.00..882688.07 rows=1 width=4)
+   ->  Nested Loop Left Anti Semi (Not-In) Join  (cost=0.00..882688.07 rows=1 width=1)
+         Join Filter: true
+         ->  Result  (cost=0.00..0.00 rows=1 width=1)
+         ->  Materialize  (cost=0.00..431.00 rows=1 width=1)
+               ->  Gather Motion 3:1  (slice1; segments: 3)  (cost=0.00..431.00 rows=1 width=1)
+                     ->  Table Scan on tbl_z  (cost=0.00..431.00 rows=1 width=1)
+                           Filter: 1 > x
+ Optimizer: PQO version 2.65.1
+(9 rows)
+
+set gp_enable_relsize_collection = on;
+-- plan with relsize collection
+explain select 1 as t1 where 1 <= ALL (select x from tbl_z);
+                                                                                                                                                                              QUERY PLAN                                                                                                                                                                               
+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
+ Result  (cost=0.00..882700.51 rows=1 width=4)
+   ->  Nested Loop  (cost=0.00..882700.51 rows=1 width=1)
+         Join Filter: true
+         ->  Result  (cost=0.00..431.01 rows=1 width=1)
+               Filter: (CASE WHEN (pg_catalog.sum((sum((CASE WHEN 1 > x THEN 1 ELSE 0 END))))) IS NULL THEN true WHEN (pg_catalog.sum((sum((CASE WHEN x IS NULL THEN 1 ELSE 0 END))))) > 0::bigint THEN NULL::boolean WHEN 1 IS NULL THEN NULL::boolean WHEN (pg_catalog.sum((sum((CASE WHEN 1 > x THEN 1 ELSE 0 END))))) = 0::bigint THEN true ELSE false END) = true
+               ->  Result  (cost=0.00..431.01 rows=1 width=1)
+                     ->  Aggregate  (cost=0.00..431.01 rows=1 width=16)
+                           ->  Gather Motion 3:1  (slice1; segments: 3)  (cost=0.00..431.01 rows=1 width=16)
+                                 ->  Aggregate  (cost=0.00..431.01 rows=1 width=16)
+                                       ->  Result  (cost=0.00..431.01 rows=321 width=8)
+                                             ->  Table Scan on tbl_z  (cost=0.00..431.01 rows=321 width=4)
+         ->  Result  (cost=0.00..0.00 rows=1 width=1)
+ Optimizer: PQO version 2.65.1
+(13 rows)
+
+drop table if exists tbl_z;
+reset optimizer_metadata_caching;
+reset gp_enable_relsize_collection;
 -- start_ignore
 drop schema qp_misc_jiras cascade;
 NOTICE:  drop cascades to table qp_misc_jiras._tbl10050_test

src/test/regress/sql/qp_misc_jiras.sql

@@ -2562,6 +2562,25 @@ select test();
 select test();
 select test();
 select test();
+
+
+--
+-- Test gp_enable_relsize_collection's effect on ORCA plan generation
+--
+create table tbl_z(x int) distributed by (x);
+set optimizer_metadata_caching to off;
+insert into tbl_z select i from generate_series(1,100) i;
+
+-- plan with no relsize collection
+explain select 1 as t1 where 1 <= ALL (select x from tbl_z);
+set gp_enable_relsize_collection = on;
+-- plan with relsize collection
+explain select 1 as t1 where 1 <= ALL (select x from tbl_z);
+
+drop table if exists tbl_z;
+reset optimizer_metadata_caching;
+reset gp_enable_relsize_collection;
+
 -- start_ignore
 drop schema qp_misc_jiras cascade;
 -- end_ignore