Merge branch 'master' into drop-local-temp-table

planner/core/exhaust_physical_plans.go

@@ -959,7 +959,7 @@ func (p *LogicalJoin) constructInnerTableScanTask(
 		// TableScan as inner child of IndexJoin can return at most 1 tuple for each outer row.
 		RowCount:     math.Min(1.0, countAfterAccess),
 		StatsVersion: ds.stats.StatsVersion,
-		// Cardinality would not be used in cost computation of IndexJoin, set leave it as default nil.
+		// NDV would not be used in cost computation of IndexJoin, set leave it as default nil.
 	}
 	rowSize := ds.TblColHists.GetTableAvgRowSize(p.ctx, ds.TblCols, ts.StoreType, true)
 	sessVars := ds.ctx.GetSessionVars()
@@ -1432,7 +1432,7 @@ func (ijHelper *indexJoinBuildHelper) updateBestChoice(ranges []*ranger.Range, p
 	}
 	var innerNDV float64
 	if stats := ijHelper.innerPlan.statsInfo(); stats != nil && stats.StatsVersion != statistics.PseudoVersion {
-		innerNDV = getCardinality(path.IdxCols[:usedColsLen], ijHelper.innerPlan.Schema(), stats)
+		innerNDV = getColsNDV(path.IdxCols[:usedColsLen], ijHelper.innerPlan.Schema(), stats)
 	}
 	// We choose the index by the NDV of the used columns, the larger the better.
 	// If NDVs are same, we choose index which uses more columns.
@@ -2141,7 +2141,7 @@ func (la *LogicalApply) exhaustPhysicalPlans(prop *property.PhysicalProperty) ([
 	}
 	cacheHitRatio := 0.0
 	if la.stats.RowCount != 0 {
-		ndv := getCardinality(columns, la.schema, la.stats)
+		ndv := getColsNDV(columns, la.schema, la.stats)
 		// for example, if there are 100 rows and the number of distinct values of these correlated columns
 		// are 70, then we can assume 30 rows can hit the cache so the cache hit ratio is 1 - (70/100) = 0.3
 		cacheHitRatio = 1 - (ndv / la.stats.RowCount)

planner/core/plan.go

@@ -126,7 +126,7 @@ func optimizeByShuffle4Window(pp *PhysicalWindow, ctx sessionctx.Context) *Physi
 	for _, item := range pp.PartitionBy {
 		partitionBy = append(partitionBy, item.Col)
 	}
-	NDV := int(getCardinality(partitionBy, dataSource.Schema(), dataSource.statsInfo()))
+	NDV := int(getColsNDV(partitionBy, dataSource.Schema(), dataSource.statsInfo()))
 	if NDV <= 1 {
 		return nil
 	}
@@ -167,7 +167,7 @@ func optimizeByShuffle4StreamAgg(pp *PhysicalStreamAgg, ctx sessionctx.Context)
 			partitionBy = append(partitionBy, col)
 		}
 	}
-	NDV := int(getCardinality(partitionBy, dataSource.Schema(), dataSource.statsInfo()))
+	NDV := int(getColsNDV(partitionBy, dataSource.Schema(), dataSource.statsInfo()))
 	if NDV <= 1 {
 		return nil
 	}

planner/core/stats_test.go

@@ -139,7 +139,7 @@ func (s *testStatsSuite) TestGroupNDVs(c *C) {
 	}
 }
 
-func (s *testStatsSuite) TestCardinalityGroupCols(c *C) {
+func (s *testStatsSuite) TestNDVGroupCols(c *C) {
 	store, dom, err := newStoreWithBootstrap()
 	c.Assert(err, IsNil)
 	defer func() {

planner/core/task.go

@@ -849,7 +849,7 @@ func (p *PhysicalMergeJoin) GetCost(lCnt, rCnt float64) float64 {
 	cpuCost += probeCost
 	// For merge join, only one group of rows with same join key(not null) are cached,
 	// we compute average memory cost using estimated group size.
-	NDV := getCardinality(innerKeys, innerSchema, innerStats)
+	NDV := getColsNDV(innerKeys, innerSchema, innerStats)
 	memoryCost := (innerStats.RowCount / NDV) * sessVars.MemoryFactor
 	return cpuCost + memoryCost
 }

planner/core/testdata/stats_suite_in.json

@@ -55,7 +55,7 @@
     ]
   },
   {
-    "name": "TestCardinalityGroupCols",
+    "name": "TestNDVGroupCols",
     "cases": [
       // DataSource -> Aggregation.
       "select count(1) from t1 group by a, b",

planner/core/testdata/stats_suite_out.json

@@ -130,7 +130,7 @@
     ]
   },
   {
-    "Name": "TestCardinalityGroupCols",
+    "Name": "TestNDVGroupCols",
     "Cases": [
       {
         "SQL": "select count(1) from t1 group by a, b",

planner/property/stats_info.go

@@ -19,7 +19,7 @@ import (
 	"github.com/pingcap/tidb/statistics"
 )
 
-// GroupNDV stores the cardinality of a group of columns.
+// GroupNDV stores the NDV of a group of columns.
 type GroupNDV struct {
 	// Cols are the UniqueIDs of columns.
 	Cols []int64
@@ -35,39 +35,39 @@ func ToString(ndvs []GroupNDV) string {
 type StatsInfo struct {
 	RowCount float64
 
-	// Column.UniqueID -> Cardinality
-	Cardinality map[int64]float64
+	// Column.UniqueID -> NDV
+	ColNDVs map[int64]float64
 
 	HistColl *statistics.HistColl
 	// StatsVersion indicates the statistics version of a table.
 	// If the StatsInfo is calculated using the pseudo statistics on a table, StatsVersion will be PseudoVersion.
 	StatsVersion uint64
 
-	// GroupNDVs stores the cardinality of column groups.
+	// GroupNDVs stores the NDV of column groups.
 	GroupNDVs []GroupNDV
 }
 
 // String implements fmt.Stringer interface.
 func (s *StatsInfo) String() string {
-	return fmt.Sprintf("count %v, Cardinality %v", s.RowCount, s.Cardinality)
+	return fmt.Sprintf("count %v, ColNDVs %v", s.RowCount, s.ColNDVs)
 }
 
 // Count gets the RowCount in the StatsInfo.
 func (s *StatsInfo) Count() int64 {
 	return int64(s.RowCount)
 }
 
-// Scale receives a selectivity and multiplies it with RowCount and Cardinality.
+// Scale receives a selectivity and multiplies it with RowCount and NDV.
 func (s *StatsInfo) Scale(factor float64) *StatsInfo {
 	profile := &StatsInfo{
 		RowCount:     s.RowCount * factor,
-		Cardinality:  make(map[int64]float64, len(s.Cardinality)),
+		ColNDVs:      make(map[int64]float64, len(s.ColNDVs)),
 		HistColl:     s.HistColl,
 		StatsVersion: s.StatsVersion,
 		GroupNDVs:    make([]GroupNDV, len(s.GroupNDVs)),
 	}
-	for id, c := range s.Cardinality {
-		profile.Cardinality[id] = c * factor
+	for id, c := range s.ColNDVs {
+		profile.ColNDVs[id] = c * factor
 	}
 	for i, g := range s.GroupNDVs {
 		profile.GroupNDVs[i] = g