VictoriaMetrics/lib/logstorage/stats_count.go
Aliaksandr Valialkin 246c339e3d
lib/logstorage: read timestamps column when it is really needed during query execution
Previously timestamps column was read unconditionally on every query.
This could significantly slow down queries, which do not need reading this column
like in https://github.com/VictoriaMetrics/VictoriaMetrics/issues/7070 .
2024-09-25 19:18:37 +02:00

210 lines
4.9 KiB
Go

package logstorage
import (
"slices"
"strconv"
"unsafe"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
)
type statsCount struct {
fields []string
}
func (sc *statsCount) String() string {
return "count(" + statsFuncFieldsToString(sc.fields) + ")"
}
func (sc *statsCount) updateNeededFields(neededFields fieldsSet) {
if len(sc.fields) == 0 {
// There is no need in fetching any columns for count(*) - the number of matching rows can be calculated as blockResult.rowsLen
return
}
neededFields.addFields(sc.fields)
}
func (sc *statsCount) newStatsProcessor() (statsProcessor, int) {
scp := &statsCountProcessor{
sc: sc,
}
return scp, int(unsafe.Sizeof(*scp))
}
type statsCountProcessor struct {
sc *statsCount
rowsCount uint64
}
func (scp *statsCountProcessor) updateStatsForAllRows(br *blockResult) int {
fields := scp.sc.fields
if len(fields) == 0 {
// Fast path - unconditionally count all the columns.
scp.rowsCount += uint64(br.rowsLen)
return 0
}
if len(fields) == 1 {
// Fast path for count(single_column)
c := br.getColumnByName(fields[0])
if c.isConst {
if c.valuesEncoded[0] != "" {
scp.rowsCount += uint64(br.rowsLen)
}
return 0
}
if c.isTime {
scp.rowsCount += uint64(br.rowsLen)
return 0
}
switch c.valueType {
case valueTypeString:
for _, v := range c.getValuesEncoded(br) {
if v != "" {
scp.rowsCount++
}
}
return 0
case valueTypeDict:
zeroDictIdx := slices.Index(c.dictValues, "")
if zeroDictIdx < 0 {
scp.rowsCount += uint64(br.rowsLen)
return 0
}
for _, v := range c.getValuesEncoded(br) {
if int(v[0]) != zeroDictIdx {
scp.rowsCount++
}
}
return 0
case valueTypeUint8, valueTypeUint16, valueTypeUint32, valueTypeUint64, valueTypeFloat64, valueTypeIPv4, valueTypeTimestampISO8601:
scp.rowsCount += uint64(br.rowsLen)
return 0
default:
logger.Panicf("BUG: unknown valueType=%d", c.valueType)
return 0
}
}
// Slow path - count rows containing at least a single non-empty value for the fields enumerated inside count().
bm := getBitmap(br.rowsLen)
defer putBitmap(bm)
bm.setBits()
for _, f := range fields {
c := br.getColumnByName(f)
if c.isConst {
if c.valuesEncoded[0] != "" {
scp.rowsCount += uint64(br.rowsLen)
return 0
}
continue
}
if c.isTime {
scp.rowsCount += uint64(br.rowsLen)
return 0
}
switch c.valueType {
case valueTypeString:
valuesEncoded := c.getValuesEncoded(br)
bm.forEachSetBit(func(i int) bool {
return valuesEncoded[i] == ""
})
case valueTypeDict:
if !slices.Contains(c.dictValues, "") {
scp.rowsCount += uint64(br.rowsLen)
return 0
}
valuesEncoded := c.getValuesEncoded(br)
bm.forEachSetBit(func(i int) bool {
dictIdx := valuesEncoded[i][0]
return c.dictValues[dictIdx] == ""
})
case valueTypeUint8, valueTypeUint16, valueTypeUint32, valueTypeUint64, valueTypeFloat64, valueTypeIPv4, valueTypeTimestampISO8601:
scp.rowsCount += uint64(br.rowsLen)
return 0
default:
logger.Panicf("BUG: unknown valueType=%d", c.valueType)
return 0
}
}
scp.rowsCount += uint64(br.rowsLen)
scp.rowsCount -= uint64(bm.onesCount())
return 0
}
func (scp *statsCountProcessor) updateStatsForRow(br *blockResult, rowIdx int) int {
fields := scp.sc.fields
if len(fields) == 0 {
// Fast path - unconditionally count the given column
scp.rowsCount++
return 0
}
if len(fields) == 1 {
// Fast path for count(single_column)
c := br.getColumnByName(fields[0])
if c.isConst {
if c.valuesEncoded[0] != "" {
scp.rowsCount++
}
return 0
}
if c.isTime {
scp.rowsCount++
return 0
}
switch c.valueType {
case valueTypeString:
valuesEncoded := c.getValuesEncoded(br)
if v := valuesEncoded[rowIdx]; v != "" {
scp.rowsCount++
}
return 0
case valueTypeDict:
valuesEncoded := c.getValuesEncoded(br)
dictIdx := valuesEncoded[rowIdx][0]
if v := c.dictValues[dictIdx]; v != "" {
scp.rowsCount++
}
return 0
case valueTypeUint8, valueTypeUint16, valueTypeUint32, valueTypeUint64, valueTypeFloat64, valueTypeIPv4, valueTypeTimestampISO8601:
scp.rowsCount++
return 0
default:
logger.Panicf("BUG: unknown valueType=%d", c.valueType)
return 0
}
}
// Slow path - count the row at rowIdx if at least a single field enumerated inside count() is non-empty
for _, f := range fields {
c := br.getColumnByName(f)
if v := c.getValueAtRow(br, rowIdx); v != "" {
scp.rowsCount++
return 0
}
}
return 0
}
func (scp *statsCountProcessor) mergeState(sfp statsProcessor) {
src := sfp.(*statsCountProcessor)
scp.rowsCount += src.rowsCount
}
func (scp *statsCountProcessor) finalizeStats() string {
return strconv.FormatUint(scp.rowsCount, 10)
}
func parseStatsCount(lex *lexer) (*statsCount, error) {
fields, err := parseStatsFuncFields(lex, "count")
if err != nil {
return nil, err
}
sc := &statsCount{
fields: fields,
}
return sc, nil
}