[#147774653] Implemented ValuesScan Operator in ORCA

This commit introduces a new operator for ValuesScan, earlier we
generated `UNION ALL` for cases where VALUES lists passed are all
constants, but now a new Operator CLogicalConstTable with an array of
const tuples will be generated

Once the plan is generated by ORCA, it will be translated to valuesscan
node in GPDB.

This enhancement helps significantly in improving the total run time for the queries
involving values scan in ORCA with const values.
Signed-off-by: NEkta Khanna <ekhanna@pivotal.io>

src/backend/gpopt/translate/CTranslatorDXLToPlStmt.cpp

@@ -151,6 +151,7 @@ CTranslatorDXLToPlStmt::InitTranslators()
 			{EdxlopPhysicalDynamicBitmapTableScan,	&gpopt::CTranslatorDXLToPlStmt::PplanBitmapTableScan},
 			{EdxlopPhysicalCTAS, 					&gpopt::CTranslatorDXLToPlStmt::PplanCTAS},
 			{EdxlopPhysicalPartitionSelector,		&gpopt::CTranslatorDXLToPlStmt::PplanPartitionSelector},
+			{EdxlopPhysicalValuesScan,				&gpopt::CTranslatorDXLToPlStmt::PplanValueScan},
 	};
 
 	const ULONG ulTranslators = GPOS_ARRAY_SIZE(rgTranslators);
@@ -2001,6 +2002,77 @@ CTranslatorDXLToPlStmt::PrteFromDXLTVF
 	return prte;
 }
 
+
+// create a range table entry from a CDXLPhysicalValuesScan node
+RangeTblEntry *
+CTranslatorDXLToPlStmt::PrteFromDXLValueScan
+	(
+	const CDXLNode *pdxlnValueScan,
+	CDXLTranslateContext *pdxltrctxOut,
+	CDXLTranslateContextBaseTable *pdxltrctxbt,
+	Plan *pplanParent
+	)
+{
+	CDXLPhysicalValuesScan *pdxlop = CDXLPhysicalValuesScan::PdxlopConvert(pdxlnValueScan->Pdxlop());
+
+	RangeTblEntry *prte = MakeNode(RangeTblEntry);
+
+	prte->relid = InvalidOid;
+	prte->subquery = NULL;
+	prte->rtekind = RTE_VALUES;
+	prte->inh = false;			/* never true for values RTEs */
+	prte->inFromCl = true;
+	prte->requiredPerms = 0;
+	prte->checkAsUser = InvalidOid;
+
+	Alias *palias = MakeNode(Alias);
+	palias->colnames = NIL;
+
+	// get value alias
+	palias->aliasname = CTranslatorUtils::SzFromWsz(pdxlop->PstrOpName()->Wsz());
+
+	// project list
+	CDXLNode *pdxlnPrL = (*pdxlnValueScan)[EdxltsIndexProjList];
+
+	// get column names
+	const ULONG ulCols = pdxlnPrL->UlArity();
+	for (ULONG ul = 0; ul < ulCols; ul++)
+	{
+		CDXLNode *pdxlnPrElem = (*pdxlnPrL)[ul];
+		CDXLScalarProjElem *pdxlopPrEl = CDXLScalarProjElem::PdxlopConvert(pdxlnPrElem->Pdxlop());
+
+		CHAR *szColName = CTranslatorUtils::SzFromWsz(pdxlopPrEl->PmdnameAlias()->Pstr()->Wsz());
+
+		Value *pvalColName = gpdb::PvalMakeString(szColName);
+		palias->colnames = gpdb::PlAppendElement(palias->colnames, pvalColName);
+
+		// save mapping col id -> index in translate context
+		(void) pdxltrctxbt->FInsertMapping(pdxlopPrEl->UlId(), ul+1 /*iAttno*/);
+	}
+
+	CMappingColIdVarPlStmt mapcidvarplstmt = CMappingColIdVarPlStmt(m_pmp, pdxltrctxbt, NULL, pdxltrctxOut, m_pctxdxltoplstmt, pplanParent);
+	const ULONG ulChildren = pdxlnValueScan->UlArity();
+	List *values_lists = NIL;
+	for (ULONG ulValue = EdxlValIndexConstStart; ulValue < ulChildren; ulValue++)
+	{
+		CDXLNode *pdxlnValueList = (*pdxlnValueScan)[ulValue];
+		const ULONG ulCols = pdxlnValueList->UlArity();
+		List *value = NIL;
+		for (ULONG ulCol = 0; ulCol < ulCols ; ulCol++)
+		{
+			Expr *pconst = m_pdxlsctranslator->PexprFromDXLNodeScalar((*pdxlnValueList)[ulCol], &mapcidvarplstmt);
+			value = gpdb::PlAppendElement(value, pconst);
+		}
+		values_lists = gpdb::PlAppendElement(values_lists, value);
+	}
+
+	prte->values_lists = (List *) values_lists;
+	prte->eref = palias;
+
+	return prte;
+}
+
+
 //---------------------------------------------------------------------------
 //	@function:
 //		CTranslatorDXLToPlStmt::PnljFromDXLNLJ
@@ -6115,4 +6187,69 @@ CTranslatorDXLToPlStmt::PplanBitmapIndexProbe
 	return pplan;
 }
 
+// translates a DXL Value Scan node into a GPDB Value scan node
+Plan *
+CTranslatorDXLToPlStmt::PplanValueScan
+	(
+	const CDXLNode *pdxlnValueScan,
+	CDXLTranslateContext *pdxltrctxOut,
+	Plan *pplanParent,
+	DrgPdxltrctx *pdrgpdxltrctxPrevSiblings
+	)
+{
+	// translation context for column mappings
+	CDXLTranslateContextBaseTable dxltrctxbt(m_pmp);
+
+	// we will add the new range table entry as the last element of the range table
+	Index iRel = gpdb::UlListLength(m_pctxdxltoplstmt->PlPrte()) + 1;
+
+	dxltrctxbt.SetIdx(iRel);
+
+	// create value scan node
+	ValuesScan *pvaluescan = MakeNode(ValuesScan);
+	pvaluescan->scan.scanrelid = iRel;
+	Plan *pplan = &(pvaluescan->scan.plan);
+
+	RangeTblEntry *prte = PrteFromDXLValueScan(pdxlnValueScan, pdxltrctxOut, &dxltrctxbt, pplan);
+	GPOS_ASSERT(NULL != prte);
+
+	pvaluescan->values_lists = (List *)gpdb::PvCopyObject(prte->values_lists);
+
+	m_pctxdxltoplstmt->AddRTE(prte);
+
+	pplan->plan_node_id = m_pctxdxltoplstmt->UlNextPlanId();
+	pplan->plan_parent_node_id = IPlanId(pplanParent);
+	pplan->nMotionNodes = 0;
+
+	// translate operator costs
+	TranslatePlanCosts
+	(
+		CDXLPhysicalProperties::PdxlpropConvert(pdxlnValueScan->Pdxlprop())->Pdxlopcost(),
+		&(pplan->startup_cost),
+		&(pplan->total_cost),
+		&(pplan->plan_rows),
+		&(pplan->plan_width)
+	);
+
+	// a table scan node must have at least 2 children: projection list and at least 1 value list
+	GPOS_ASSERT(2 <= pdxlnValueScan->UlArity());
+
+	CDXLNode *pdxlnPrL = (*pdxlnValueScan)[EdxltsIndexProjList];
+
+	// translate proj list
+	List *plTargetList = PlTargetListFromProjList
+							(
+							pdxlnPrL,
+							&dxltrctxbt,
+							NULL,
+							pdxltrctxOut,
+							pplan
+							);
+
+	pplan->targetlist = plTargetList;
+
+	return (Plan *) pvaluescan;
+}
+
+
 // EOF

src/backend/gpopt/translate/CTranslatorQueryToDXL.cpp

@@ -3012,23 +3012,29 @@ CTranslatorQueryToDXL::PdxlnFromValues
 {
 	List *plTuples = prte->values_lists;
 	GPOS_ASSERT(NULL != plTuples);
-	
+
 	const ULONG ulValues = gpdb::UlListLength(plTuples);
 	GPOS_ASSERT(0 < ulValues);
 
 	// children of the UNION ALL
 	DrgPdxln *pdrgpdxln = GPOS_NEW(m_pmp) DrgPdxln(m_pmp);
-	
+
+	// array of datum arrays for Values
+	DrgPdrgPdxldatum *pdrgpdrgpdxldatumValues = GPOS_NEW(m_pmp) DrgPdrgPdxldatum(m_pmp);
+
 	// array of input colid arrays
 	DrgPdrgPul *pdrgpdrgulInputColIds = GPOS_NEW(m_pmp) DrgPdrgPul(m_pmp);
 
 	// array of column descriptor for the UNION ALL operator
 	DrgPdxlcd *pdrgpdxlcd = GPOS_NEW(m_pmp) DrgPdxlcd(m_pmp);
-	
+
 	// translate the tuples in the value scan
 	ULONG ulTuplePos = 0;
 	ListCell *plcTuple = NULL;
 	GPOS_ASSERT(NULL != prte->eref);
+
+	// flag for checking value list has only constants. For all constants --> VALUESCAN operator else retain UnionAll
+	BOOL fAllConstant = true;
 	ForEach (plcTuple, plTuples)
 	{
 		List *plTuple = (List *) lfirst(plcTuple);
@@ -3039,24 +3045,24 @@ CTranslatorQueryToDXL::PdxlnFromValues
 
 		// array of project elements (for expression elements)
 		DrgPdxln *pdrgpdxlnPrEl = GPOS_NEW(m_pmp) DrgPdxln(m_pmp);
-		
+
 		// array of datum (for datum constant values)
 		DrgPdxldatum *pdrgpdxldatum = GPOS_NEW(m_pmp) DrgPdxldatum(m_pmp);
-		
+
 		// array of column descriptors for the CTG containing the datum array
 		DrgPdxlcd *pdrgpdxlcdCTG = GPOS_NEW(m_pmp) DrgPdxlcd(m_pmp);
-		
+
 		List *plColnames = prte->eref->colnames;
 		GPOS_ASSERT(NULL != plColnames);
 		GPOS_ASSERT(gpdb::UlListLength(plTuple) == gpdb::UlListLength(plColnames));
-		
+
 		// translate the columns
 		ULONG ulColPos = 0;
 		ListCell *plcColumn = NULL;
 		ForEach (plcColumn, plTuple)
 		{
 			Expr *pexpr = (Expr *) lfirst(plcColumn);
-			
+
 			CHAR *szColName = (CHAR *) strVal(gpdb::PvListNth(plColnames, ulColPos));
 			ULONG ulColId = ULONG_MAX;	
 			if (IsA(pexpr, Const))
@@ -3065,13 +3071,13 @@ CTranslatorQueryToDXL::PdxlnFromValues
 				Const *pconst = (Const *) pexpr;
 				CDXLDatum *pdxldatum = m_psctranslator->Pdxldatum(pconst);
 				pdrgpdxldatum->Append(pdxldatum);
-				
+
 				ulColId = m_pidgtorCol->UlNextId();
-				
+
 				CWStringDynamic *pstrAlias = CDXLUtils::PstrFromSz(m_pmp, szColName);
 				CMDName *pmdname = GPOS_NEW(m_pmp) CMDName(m_pmp, pstrAlias);
 				GPOS_DELETE(pstrAlias);
-				
+
 				CDXLColDescr *pdxlcd = GPOS_NEW(m_pmp) CDXLColDescr
 													(
 													m_pmp,
@@ -3091,17 +3097,18 @@ CTranslatorQueryToDXL::PdxlnFromValues
 			}
 			else
 			{
+				fAllConstant = false;
 				// translate the scalar expression into a project element
 				CDXLNode *pdxlnPrE = PdxlnPrEFromGPDBExpr(pexpr, szColName, true /* fInsistNewColIds */ );
 				pdrgpdxlnPrEl->Append(pdxlnPrE);
 				ulColId = CDXLScalarProjElem::PdxlopConvert(pdxlnPrE->Pdxlop())->UlId();
-				
+
 				if (0 == ulTuplePos)
 				{
 					CWStringDynamic *pstrAlias = CDXLUtils::PstrFromSz(m_pmp, szColName);
 					CMDName *pmdname = GPOS_NEW(m_pmp) CMDName(m_pmp, pstrAlias);
 					GPOS_DELETE(pstrAlias);
-					
+
 					CDXLColDescr *pdxlcd = GPOS_NEW(m_pmp) CDXLColDescr
 														(
 														m_pmp,
@@ -3113,52 +3120,74 @@ CTranslatorQueryToDXL::PdxlnFromValues
 														);
 					pdrgpdxlcd->Append(pdxlcd);
 				}
-			} 
+			}
 
 			GPOS_ASSERT(ULONG_MAX != ulColId);
 
 			pdrgpulColIds->Append(GPOS_NEW(m_pmp) ULONG(ulColId));
 			ulColPos++;
 		}
-		
+
 		pdrgpdxln->Append(PdxlnFromColumnValues(pdrgpdxldatum, pdrgpdxlcdCTG, pdrgpdxlnPrEl));
+		if (fAllConstant)
+		{
+			pdrgpdxldatum->AddRef();
+			pdrgpdrgpdxldatumValues->Append(pdrgpdxldatum);
+		}
+
 		pdrgpdrgulInputColIds->Append(pdrgpulColIds);
 		ulTuplePos++;
-		
+
 		// cleanup
 		pdrgpdxldatum->Release();
 		pdrgpdxlnPrEl->Release();
 		pdrgpdxlcdCTG->Release();
 	}
-	
+
 	GPOS_ASSERT(NULL != pdrgpdxlcd);
 
-	if (1 < ulValues)
+	if (fAllConstant)
+	{
+		// create Const Table DXL Node
+		CDXLLogicalConstTable *pdxlop = GPOS_NEW(m_pmp) CDXLLogicalConstTable(m_pmp, pdrgpdxlcd, pdrgpdrgpdxldatumValues);
+		CDXLNode *pdxln = GPOS_NEW(m_pmp) CDXLNode(m_pmp, pdxlop);
+
+		// make note of new columns from Value Scan
+		m_pmapvarcolid->LoadColumns(m_ulQueryLevel, ulRTIndex, pdxlop->Pdrgpdxlcd());
+
+		// cleanup
+		pdrgpdxln->Release();
+		pdrgpdrgulInputColIds->Release();
+
+		return pdxln;
+	}
+	else if (1 < ulValues)
 	{
 		// create a UNION ALL operator
 		CDXLLogicalSetOp *pdxlop = GPOS_NEW(m_pmp) CDXLLogicalSetOp(m_pmp, EdxlsetopUnionAll, pdrgpdxlcd, pdrgpdrgulInputColIds, false);
 		CDXLNode *pdxln = GPOS_NEW(m_pmp) CDXLNode(m_pmp, pdxlop, pdrgpdxln);
-		
+
 		// make note of new columns from UNION ALL
 		m_pmapvarcolid->LoadColumns(m_ulQueryLevel, ulRTIndex, pdxlop->Pdrgpdxlcd());
-		
+		pdrgpdrgpdxldatumValues->Release();
+
 		return pdxln;
 	}
 
-	
 	GPOS_ASSERT(1 == pdrgpdxln->UlLength());
-			
+
 	CDXLNode *pdxln = (*pdrgpdxln)[0];
 	pdxln->AddRef();
-		
+
 	// make note of new columns
 	m_pmapvarcolid->LoadColumns(m_ulQueryLevel, ulRTIndex, pdrgpdxlcd);	
-	
+
 	//cleanup
+	pdrgpdrgpdxldatumValues->Release();
 	pdrgpdxln->Release();
 	pdrgpdrgulInputColIds->Release();
 	pdrgpdxlcd->Release();
-	
+
 	return pdxln;
 }
 

src/include/gpopt/translate/CTranslatorDXLToPlStmt.h

@@ -581,6 +581,15 @@ namespace gpdxl
 				DrgPdxltrctx *pdrgpdxltrctxPrevSiblings // translation contexts of previous siblings
 				);
 
+			// translate a DXL Value Scan into GPDB Value Scan
+			Plan *PplanValueScan
+				(
+				const CDXLNode *pdxlnValueScan,
+				CDXLTranslateContext *pdxltrctxOut,
+				Plan *pplanParent,
+				DrgPdxltrctx *pdrgpdxltrctxPrevSiblings
+				);
+
 			// translate DXL filter list into GPDB filter list
 			List *PlFilterList
 				(
@@ -600,6 +609,15 @@ namespace gpdxl
 				Plan *pplanParent
 				);
 
+			// create range table entry from a CDXLPhysicalValueScan node
+			RangeTblEntry *PrteFromDXLValueScan
+				(
+				const CDXLNode *pdxlnValueScan,
+				CDXLTranslateContext *pdxltrctxOut,
+				CDXLTranslateContextBaseTable *pdxltrctxbt,
+				Plan *pplanParent
+				);
+
 			// create range table entry from a table descriptor
 			RangeTblEntry *PrteFromTblDescr
 				(
@@ -736,7 +754,7 @@ namespace gpdxl
 
 			// check if the given operator is a DML operator on a distributed table
 			BOOL FTargetTableDistributed(CDXLOperator *pdxlop);
-			
+
 			// add a target entry for the given colid to the given target list
 			ULONG UlAddTargetEntryForColId
 				(

src/test/regress/expected/bfv_index_optimizer.out

@@ -27,17 +27,13 @@ explain select * from bfv_tab1, (values(147, 'RFAAAA'), (931, 'VJAAAA')) as v (i
  Gather Motion 3:1  (slice1; segments: 3)  (cost=0.00..2.00 rows=1 width=256)
    ->  Nested Loop  (cost=0.00..2.00 rows=1 width=256)
          Join Filter: true
-         ->  Append  (cost=0.00..0.00 rows=1 width=12)
-               ->  Result  (cost=0.00..0.00 rows=1 width=12)
-                     ->  Result  (cost=0.00..0.00 rows=1 width=12)
-               ->  Result  (cost=0.00..0.00 rows=1 width=12)
-                     ->  Result  (cost=0.00..0.00 rows=1 width=12)
-         ->  Index Scan using bfv_tab1_idx1 on bfv_tab1  (cost=0.00..2.00 rows=1 width=244)
-               Index Cond: bfv_tab1.unique1 = "outer".column1
-               Filter: bfv_tab1.stringu1::text = "outer".column2
- Settings:  optimizer=on
- Optimizer status: PQO version 1.621
-(13 rows)
+         ->  Result  (cost=0.00..0.00 rows=2 width=12)
+               ->  Values Scan on "Values"  (cost=0.00..0.00 rows=2 width=12)
+         ->  Index Scan using bfv_tab1_idx1 on bfv_tab1  (cost=0.00..6.00 rows=1 width=244)
+               Index Cond: bfv_tab1.unique1 = "Values".column1
+               Filter: bfv_tab1.stringu1::text = "Values".column2
+ Optimizer status: PQO version 2.35.1
+(9 rows)
 
 set gp_enable_relsize_collection=on;
 explain select * from bfv_tab1, (values(147, 'RFAAAA'), (931, 'VJAAAA')) as v (i, j)
@@ -47,17 +43,14 @@ explain select * from bfv_tab1, (values(147, 'RFAAAA'), (931, 'VJAAAA')) as v (i
  Gather Motion 3:1  (slice1; segments: 3)  (cost=0.00..2.00 rows=1 width=256)
    ->  Nested Loop  (cost=0.00..2.00 rows=1 width=256)
          Join Filter: true
-         ->  Append  (cost=0.00..0.00 rows=1 width=12)
-               ->  Result  (cost=0.00..0.00 rows=1 width=12)
-                     ->  Result  (cost=0.00..0.00 rows=1 width=12)
-               ->  Result  (cost=0.00..0.00 rows=1 width=12)
-                     ->  Result  (cost=0.00..0.00 rows=1 width=12)
-         ->  Index Scan using bfv_tab1_idx1 on bfv_tab1  (cost=0.00..2.00 rows=1 width=244)
-               Index Cond: bfv_tab1.unique1 = (147)
-               Filter: bfv_tab1.stringu1::text = ('RFAAAA'::text)
- Settings:  gp_enable_relsize_collection=on; optimizer=on
- Optimizer status: PQO version 1.667
-(13 rows)
+         ->  Result  (cost=0.00..0.00 rows=2 width=12)
+               ->  Values Scan on "Values"  (cost=0.00..0.00 rows=2 width=12)
+         ->  Index Scan using bfv_tab1_idx1 on bfv_tab1  (cost=0.00..6.00 rows=1 width=244)
+               Index Cond: bfv_tab1.unique1 = "Values".column1
+               Filter: bfv_tab1.stringu1::text = "Values".column2
+ Settings:  gp_enable_relsize_collection=on
+ Optimizer status: PQO version 2.35.1
+(10 rows)
 
 reset gp_enable_relsize_collection;
 --start_ignore

src/test/regress/expected/gp_explain_optimizer.out

@@ -94,17 +94,13 @@ WHERE mpp22263.unique1 = v.i and mpp22263.stringu1 = v.j;
  Gather Motion 3:1  (slice1; segments: 3)  (cost=0.00..2.00 rows=1 width=256)
    ->  Nested Loop  (cost=0.00..2.00 rows=1 width=256)
          Join Filter: true
-         ->  Append  (cost=0.00..0.00 rows=1 width=12)
-               ->  Result  (cost=0.00..0.00 rows=1 width=12)
-                     ->  Result  (cost=0.00..0.00 rows=1 width=12)
-               ->  Result  (cost=0.00..0.00 rows=1 width=12)
-                     ->  Result  (cost=0.00..0.00 rows=1 width=12)
+         ->  Result  (cost=0.00..0.00 rows=1 width=12)
+               ->  Values Scan on "Values"  (cost=0.00..0.00 rows=1 width=12)
          ->  Index Scan using mpp22263_idx1 on mpp22263  (cost=0.00..2.00 rows=1 width=244)
-               Index Cond: mpp22263.unique1 = "outer".column1
-               Filter: mpp22263.stringu1::text = "outer".column2
- Settings:  optimizer=on
- Optimizer status: PQO version 2.7.0
-(13 rows)
+               Index Cond: mpp22263.unique1 = "Values".column1
+               Filter: mpp22263.stringu1::text = "Values".column2
+ Optimizer status: PQO version 2.35.1
+(9 rows)
 
 -- atmsort.pm masks out differences in the Filter line, so just memorizing
 -- the output of the above EXPLAIN isn't enough to catch a faulty Filter line.
@@ -115,10 +111,10 @@ select * from mpp22263, (values(147, 'RFAAAA'), (931, 'VJAAAA')) as v (i, j)
 WHERE mpp22263.unique1 = v.i and mpp22263.stringu1 = v.j;
   $$) as et
 WHERE et like '%Filter: %';
-                               et                                
------------------------------------------------------------------
+                                et                                
+------------------------------------------------------------------
          Join Filter: true
-               Filter: mpp22263.stringu1::text = "outer".column2
+               Filter: mpp22263.stringu1::text = "Values".column2
 (2 rows)
 
 --