func TestCompareSelectivityWithStatsAndQuantiles(t *testing.T) {
sixtyValue := 60.0
numberSixty := NewLiteralNumber(sixtyValue)
fortyValue := 40.0
numberForty := NewLiteralNumber(fortyValue)
abvStats := stats.DefaultPathStats(0, 100)
abvStats.Rows = 50
abvStats.DistinctValues = 10
abvStats.MostFrequentValues.Consider(sixtyValue, 6)
abvStats.Quantiles = []stats.QuantileRange{
stats.QuantileRange{Start: 0.0, End: 20.0, Count: 10},
stats.QuantileRange{Start: 21.0, End: 40.0, Count: 10},
stats.QuantileRange{Start: 41.0, End: 60.0, Count: 10},
stats.QuantileRange{Start: 61.0, End: 80.0, Count: 10},
stats.QuantileRange{Start: 81.0, End: 100.0, Count: 10},
}
pathStatistics := map[string]stats.PathStatistics{
"doc.abv": abvStats,
}
tests := []struct {
input BooleanExpression
output interface{}
}{
// when value is in freq vals, we should get precise estimate
{NewEqualToOperator(NewProperty("doc.abv"), numberSixty), 6.0 / 50.0},
{NewNotEqualToOperator(NewProperty("doc.abv"), numberSixty), 44.0 / 50.0},
// this time it should fall back to the quantiles
{NewEqualToOperator(NewProperty("doc.abv"), numberForty), 10.0 / 50.0},
{NewNotEqualToOperator(NewProperty("doc.abv"), numberForty), 40.0 / 50.0},
// now test the ranges
{NewLessThanOperator(NewProperty("doc.abv"), numberSixty), 30.0 / 50.0},
{NewGreaterThanOperator(NewProperty("doc.abv"), numberSixty), 30.0 / 50.0},
// note these two find exact values for the equals portion
{NewLessThanOrEqualOperator(NewProperty("doc.abv"), numberSixty), 36.0 / 50.0},
{NewGreaterThanOrEqualOperator(NewProperty("doc.abv"), numberSixty), 36.0 / 50.0},
// these two should not
{NewLessThanOrEqualOperator(NewProperty("doc.abv"), numberForty), 30.0 / 50.0},
{NewGreaterThanOrEqualOperator(NewProperty("doc.abv"), numberForty), 50.0 / 50.0},
}
for _, x := range tests {
result := x.input.GetSelectivity(pathStatistics)
if !reflect.DeepEqual(result, x.output) {
t.Errorf("Expected %v for %v, got %v", x.output, x.input, result)
}
}
}
func TestCompareSelectivityWithStatsNoQuantiles(t *testing.T) {
sixtyValue := 60.0
numberSixty := NewLiteralNumber(sixtyValue)
fortyValue := 40.0
numberForty := NewLiteralNumber(fortyValue)
abvStats := stats.DefaultPathStats(0, 100)
abvStats.Rows = 50
abvStats.DistinctValues = 10
abvStats.MostFrequentValues.Consider(sixtyValue, 6)
pathStatistics := map[string]stats.PathStatistics{
"doc.abv": abvStats,
}
tests := []struct {
input BooleanExpression
output interface{}
}{
// when value is in freq vals, we should get precise estimate
{NewEqualToOperator(NewProperty("doc.abv"), numberSixty), 6.0 / 50.0},
{NewNotEqualToOperator(NewProperty("doc.abv"), numberSixty), 44.0 / 50.0},
// if value is not in freq vals, and we dont have quantiles
// it should fall back to defaults
{NewEqualToOperator(NewProperty("doc.abv"), numberForty), 1.0 / 10.0},
{NewNotEqualToOperator(NewProperty("doc.abv"), numberForty), 9.0 / 10.0},
}
for _, x := range tests {
result := x.input.GetSelectivity(pathStatistics)
if !reflect.DeepEqual(result, x.output) {
t.Errorf("Expected %v for %v, got %v", x.output, x.input, result)
}
}
}
func (this *CouchbaseViewAccessPath) UpdateStats() {
log.Printf("Starting UpdateStats for %v", this)
// only try to address single column stats here
if len(this.keys) == 1 {
pathStat := stats.DefaultPathStats(MIN_KEY, MAX_KEY)
vres, err := this.dataSource.bucket.View(this.ddoc, this.view, map[string]interface{}{"reduce": false, "limit": 0})
if err != nil {
log.Printf("Unable to determine cardinality of view, defaulting to MAX")
} else {
pathStat.Rows = vres.TotalRows
pathStat.DistinctValues = vres.TotalRows
}
// try to gather deeper stats
targetCountPerQuantile := pathStat.Rows / pathStat.NumQuantiles()
options := map[string]interface{}{"group_level": 1}
viewRowsChannel := make(chan couchbase.ViewRow)
go WalkViewInBatches(viewRowsChannel, this.dataSource.bucket, this.ddoc, this.view, options, BATCH_SIZE)
distinctRows := 0
currentQuantile := stats.QuantileRange{}
runningCount := 0
numQuantilesBuilt := 0
for row := range viewRowsChannel {
if distinctRows == 0 {
pathStat.MinValue = row.Key
}
if currentQuantile.Count == 0 {
currentQuantile.Start = row.Key
}
pathStat.MaxValue = row.Key
currentQuantile.End = row.Key
distinctRows++
// expect result to be _stats reduce
switch stats_reduce := row.Value.(type) {
case map[string]interface{}:
switch stats_count := stats_reduce["count"].(type) {
case float64:
pathStat.MostFrequentValues.Consider(row.Key, stats_count)
currentQuantile.Count = currentQuantile.Count + int(stats_count)
runningCount = runningCount + int(stats_count)
}
}
if currentQuantile.Count > targetCountPerQuantile {
//close out the quantile
pathStat.Quantiles = append(pathStat.Quantiles, currentQuantile)
numQuantilesBuilt = numQuantilesBuilt + 1
// update the target counts (we may have overshot because of a large bin)
targetCountPerQuantile = (pathStat.Rows - runningCount) / (pathStat.NumQuantiles() - numQuantilesBuilt)
//empty out a new quantile
currentQuantile = stats.QuantileRange{}
}
}
// close out the last quantile
pathStat.Quantiles = append(pathStat.Quantiles, currentQuantile)
numQuantilesBuilt = numQuantilesBuilt + 1
pathStat.DistinctValues = distinctRows
this.dataSource.pathStats[this.keys[0]] = pathStat
log.Printf("%v", pathStat)
}
log.Printf("Finished UpdateStats for %v", this)
}
