VictoriaMetrics/lib/streamaggr/histogram_bucket.go
Aliaksandr Valialkin 32aa0751a1
lib/streamaggr: follow-up for 7cb894a777
- Use bytesutil.InternString() instead of strings.Clone() for inputKey and outputKey in aggregatorpushSamples().
  This should reduce string allocation rate, since strings can be re-used between aggrState flushes.
- Reduce memory allocations at dedupAggrShard by storing dedupAggrSample by value in the active series map.
- Remove duplicate call to bytesutil.InternBytes() at Deduplicator, since it is already called inside dedupAggr.pushSamples().
- Add missing string interning at rateAggrState.pushSamples().

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/pull/6402
2024-06-07 16:35:53 +02:00

118 lines
2.8 KiB
Go

package streamaggr
import (
"math"
"sync"
"time"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
"github.com/VictoriaMetrics/metrics"
)
// histogramBucketAggrState calculates output=histogram_bucket, e.g. VictoriaMetrics histogram over input samples.
type histogramBucketAggrState struct {
m sync.Map
stalenessSecs uint64
}
type histogramBucketStateValue struct {
mu sync.Mutex
h metrics.Histogram
deleteDeadline uint64
deleted bool
}
func newHistogramBucketAggrState(stalenessInterval time.Duration) *histogramBucketAggrState {
stalenessSecs := roundDurationToSecs(stalenessInterval)
return &histogramBucketAggrState{
stalenessSecs: stalenessSecs,
}
}
func (as *histogramBucketAggrState) pushSamples(samples []pushSample) {
currentTime := fasttime.UnixTimestamp()
deleteDeadline := currentTime + as.stalenessSecs
for i := range samples {
s := &samples[i]
outputKey := getOutputKey(s.key)
again:
v, ok := as.m.Load(outputKey)
if !ok {
// The entry is missing in the map. Try creating it.
v = &histogramBucketStateValue{}
outputKey = bytesutil.InternString(outputKey)
vNew, loaded := as.m.LoadOrStore(outputKey, v)
if loaded {
// Use the entry created by a concurrent goroutine.
v = vNew
}
}
sv := v.(*histogramBucketStateValue)
sv.mu.Lock()
deleted := sv.deleted
if !deleted {
sv.h.Update(s.value)
sv.deleteDeadline = deleteDeadline
}
sv.mu.Unlock()
if deleted {
// The entry has been deleted by the concurrent call to flushState
// Try obtaining and updating the entry again.
goto again
}
}
}
func (as *histogramBucketAggrState) removeOldEntries(currentTime uint64) {
m := &as.m
m.Range(func(k, v interface{}) bool {
sv := v.(*histogramBucketStateValue)
sv.mu.Lock()
deleted := currentTime > sv.deleteDeadline
if deleted {
// Mark the current entry as deleted
sv.deleted = deleted
}
sv.mu.Unlock()
if deleted {
m.Delete(k)
}
return true
})
}
func (as *histogramBucketAggrState) flushState(ctx *flushCtx, resetState bool) {
_ = resetState // it isn't used here
currentTime := fasttime.UnixTimestamp()
currentTimeMsec := int64(currentTime) * 1000
as.removeOldEntries(currentTime)
m := &as.m
m.Range(func(k, v interface{}) bool {
sv := v.(*histogramBucketStateValue)
sv.mu.Lock()
if !sv.deleted {
key := k.(string)
sv.h.VisitNonZeroBuckets(func(vmrange string, count uint64) {
ctx.appendSeriesWithExtraLabel(key, "histogram_bucket", currentTimeMsec, float64(count), "vmrange", vmrange)
})
}
sv.mu.Unlock()
return true
})
}
func roundDurationToSecs(d time.Duration) uint64 {
if d < 0 {
return 0
}
secs := d.Seconds()
return uint64(math.Ceil(secs))
}