VictoriaMetrics/lib/logstorage/stats_values.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

209 lines
4.8 KiB
Go

package logstorage
import (
"fmt"
"strings"
"unsafe"
)
type statsValues struct {
fields []string
limit uint64
}
func (sv *statsValues) String() string {
s := "values(" + statsFuncFieldsToString(sv.fields) + ")"
if sv.limit > 0 {
s += fmt.Sprintf(" limit %d", sv.limit)
}
return s
}
func (sv *statsValues) updateNeededFields(neededFields fieldsSet) {
updateNeededFieldsForStatsFunc(neededFields, sv.fields)
}
func (sv *statsValues) newStatsProcessor() (statsProcessor, int) {
svp := &statsValuesProcessor{
sv: sv,
}
return svp, int(unsafe.Sizeof(*svp))
}
type statsValuesProcessor struct {
sv *statsValues
values []string
}
func (svp *statsValuesProcessor) updateStatsForAllRows(br *blockResult) int {
if svp.limitReached() {
// Limit on the number of unique values has been reached
return 0
}
stateSizeIncrease := 0
fields := svp.sv.fields
if len(fields) == 0 {
for _, c := range br.getColumns() {
stateSizeIncrease += svp.updateStatsForAllRowsColumn(c, br)
}
} else {
for _, field := range fields {
c := br.getColumnByName(field)
stateSizeIncrease += svp.updateStatsForAllRowsColumn(c, br)
}
}
return stateSizeIncrease
}
func (svp *statsValuesProcessor) updateStatsForAllRowsColumn(c *blockResultColumn, br *blockResult) int {
stateSizeIncrease := 0
if c.isConst {
v := strings.Clone(c.valuesEncoded[0])
stateSizeIncrease += len(v)
values := svp.values
for i := 0; i < br.rowsLen; i++ {
values = append(values, v)
}
svp.values = values
stateSizeIncrease += br.rowsLen * int(unsafe.Sizeof(values[0]))
return stateSizeIncrease
}
if c.valueType == valueTypeDict {
dictValues := make([]string, len(c.dictValues))
for i, v := range c.dictValues {
dictValues[i] = strings.Clone(v)
stateSizeIncrease += len(v)
}
values := svp.values
for _, encodedValue := range c.getValuesEncoded(br) {
idx := encodedValue[0]
values = append(values, dictValues[idx])
}
svp.values = values
stateSizeIncrease += br.rowsLen * int(unsafe.Sizeof(values[0]))
return stateSizeIncrease
}
values := svp.values
for _, v := range c.getValues(br) {
if len(values) == 0 || values[len(values)-1] != v {
v = strings.Clone(v)
stateSizeIncrease += len(v)
}
values = append(values, v)
}
svp.values = values
stateSizeIncrease += br.rowsLen * int(unsafe.Sizeof(values[0]))
return stateSizeIncrease
}
func (svp *statsValuesProcessor) updateStatsForRow(br *blockResult, rowIdx int) int {
if svp.limitReached() {
// Limit on the number of unique values has been reached
return 0
}
stateSizeIncrease := 0
fields := svp.sv.fields
if len(fields) == 0 {
for _, c := range br.getColumns() {
stateSizeIncrease += svp.updateStatsForRowColumn(c, br, rowIdx)
}
} else {
for _, field := range fields {
c := br.getColumnByName(field)
stateSizeIncrease += svp.updateStatsForRowColumn(c, br, rowIdx)
}
}
return stateSizeIncrease
}
func (svp *statsValuesProcessor) updateStatsForRowColumn(c *blockResultColumn, br *blockResult, rowIdx int) int {
stateSizeIncrease := 0
if c.isConst {
v := strings.Clone(c.valuesEncoded[0])
stateSizeIncrease += len(v)
svp.values = append(svp.values, v)
stateSizeIncrease += int(unsafe.Sizeof(svp.values[0]))
return stateSizeIncrease
}
if c.valueType == valueTypeDict {
// collect unique non-zero c.dictValues
valuesEncoded := c.getValuesEncoded(br)
dictIdx := valuesEncoded[rowIdx][0]
v := strings.Clone(c.dictValues[dictIdx])
stateSizeIncrease += len(v)
svp.values = append(svp.values, v)
stateSizeIncrease += int(unsafe.Sizeof(svp.values[0]))
return stateSizeIncrease
}
// collect unique values for the given rowIdx.
v := c.getValueAtRow(br, rowIdx)
v = strings.Clone(v)
stateSizeIncrease += len(v)
svp.values = append(svp.values, v)
stateSizeIncrease += int(unsafe.Sizeof(svp.values[0]))
return stateSizeIncrease
}
func (svp *statsValuesProcessor) mergeState(sfp statsProcessor) {
if svp.limitReached() {
return
}
src := sfp.(*statsValuesProcessor)
svp.values = append(svp.values, src.values...)
}
func (svp *statsValuesProcessor) finalizeStats() string {
items := svp.values
if len(items) == 0 {
return "[]"
}
if limit := svp.sv.limit; limit > 0 && uint64(len(items)) > limit {
items = items[:limit]
}
return marshalJSONArray(items)
}
func (svp *statsValuesProcessor) limitReached() bool {
limit := svp.sv.limit
return limit > 0 && uint64(len(svp.values)) >= limit
}
func parseStatsValues(lex *lexer) (*statsValues, error) {
fields, err := parseStatsFuncFields(lex, "values")
if err != nil {
return nil, err
}
sv := &statsValues{
fields: fields,
}
if lex.isKeyword("limit") {
lex.nextToken()
n, ok := tryParseUint64(lex.token)
if !ok {
return nil, fmt.Errorf("cannot parse 'limit %s' for 'values': %w", lex.token, err)
}
lex.nextToken()
sv.limit = n
}
return sv, nil
}