review comments

This commit is contained in:
AndrewChubatiuk 2024-07-09 11:07:10 +03:00 committed by Andrii Chubatiuk
parent 4c5ca49297
commit c0ca03aee5
No known key found for this signature in database
GPG key ID: 96D776CC99880667
7 changed files with 34 additions and 41 deletions

View file

@ -72,7 +72,7 @@ func (da *dedupAggr) itemsCount() uint64 {
return n
}
func (da *dedupAggr) pushSamples(samples []pushSample, dedupIdx int) {
func (da *dedupAggr) pushSamples(samples []pushSample, _ int64, dedupIdx int) {
pss := getPerShardSamples()
shards := pss.shards
for _, sample := range samples {

View file

@ -21,7 +21,7 @@ func TestDedupAggrSerial(t *testing.T) {
sample.value = float64(i + j)
expectedSamplesMap[sample.key] = *sample
}
da.pushSamples(samples, 0)
da.pushSamples(samples, 0, 0)
}
if n := da.sizeBytes(); n > 5_000_000 {
@ -73,7 +73,7 @@ func TestDedupAggrConcurrent(_ *testing.T) {
sample.key = fmt.Sprintf("key_%d", j)
sample.value = float64(i + j)
}
da.pushSamples(samples, 0)
da.pushSamples(samples, 0, 0)
}
}()
}

View file

@ -27,7 +27,7 @@ func benchmarkDedupAggr(b *testing.B, samplesPerPush int) {
b.RunParallel(func(pb *testing.PB) {
for pb.Next() {
for i := 0; i < loops; i++ {
da.pushSamples(benchSamples, 0)
da.pushSamples(benchSamples, 0, 0)
}
}
})

View file

@ -118,7 +118,7 @@ func (d *Deduplicator) Push(tss []prompbmarshal.TimeSeries) {
}
for idx, ps := range pss {
d.da.pushSamples(ps, idx)
d.da.pushSamples(ps, 0, idx)
}
ctx.pss = pss

View file

@ -136,8 +136,8 @@ func (as *rateAggrState) flushState(ctx *flushCtx) {
}
// Delete outdated entries in state
var rate float64
var totalItems int
rate := 0.0
countSeries := 0
for k1, state := range sv.state {
if ctx.flushTimestamp > state.deleteDeadline {
delete(sv.state, k1)
@ -156,7 +156,7 @@ func (as *rateAggrState) flushState(ctx *flushCtx) {
rate += (v1.total) * 1000 / float64(rateInterval)
state.prevTimestamp = v1.timestamp
state.prevValue = v1.value
totalItems++
countSeries++
}
state.lastValues[ctx.idx] = rateLastValueState{}
sv.state[k1] = state
@ -164,9 +164,9 @@ func (as *rateAggrState) flushState(ctx *flushCtx) {
sv.mu.Unlock()
if totalItems > 0 {
if countSeries > 0 {
if as.isAvg {
rate /= float64(totalItems)
rate /= float64(countSeries)
}
key := k.(string)
ctx.appendSeries(key, suffix, rate)

View file

@ -801,7 +801,7 @@ func (a *aggregator) runFlusher(pushFunc PushFunc, alignFlushToInterval, skipInc
dedupTime := ct.Truncate(tickInterval)
if a.ignoreOldSamples {
dedupIdx, flushIdx = a.getFlushIndices(dedupTime, flushDeadline)
dedupIdx, flushIdx = a.getAggrIdxs(dedupTime, flushDeadline)
}
pf := pushFunc
@ -848,28 +848,24 @@ func (a *aggregator) runFlusher(pushFunc PushFunc, alignFlushToInterval, skipInc
if !skipIncompleteFlush && ignoreFirstIntervals <= 0 {
dedupTime := time.Now().Truncate(tickInterval).Add(tickInterval)
if a.ignoreOldSamples {
dedupIdx, flushIdx = a.getFlushIndices(dedupTime, flushDeadline)
dedupIdx, flushIdx = a.getAggrIdxs(dedupTime, flushDeadline)
}
a.dedupFlush(flushDeadline.UnixMilli(), dedupIdx, flushIdx)
a.flush(pushFunc, flushDeadline.UnixMilli(), flushIdx)
}
}
func (a *aggregator) getFlushIndices(dedupTime, flushTime time.Time) (int, int) {
flushTimestamp := flushTime.UnixMilli()
flushIntervals := int(flushTimestamp / int64(a.interval/time.Millisecond))
var dedupIndex, flushIndex int
func (a *aggregator) getAggrIdxs(dedupTime, flushTime time.Time) (int, int) {
flushIdx := getStateIdx(a.interval.Milliseconds(), flushTime.Add(-a.interval).UnixMilli())
dedupIdx := flushIdx
if a.dedupInterval > 0 {
dedupTimestamp := dedupTime.UnixMilli()
dedupIntervals := int(dedupTimestamp / int64(a.dedupInterval/time.Millisecond))
intervalsRatio := int(a.interval / a.dedupInterval)
dedupIndex = dedupIntervals % aggrStateSize
flushIndex = flushIntervals % (aggrStateSize / intervalsRatio)
} else {
flushIndex = flushIntervals % aggrStateSize
dedupIndex = flushIndex
dedupIdx = getStateIdx(a.dedupInterval.Milliseconds(), dedupTime.Add(-a.dedupInterval).UnixMilli())
}
return dedupIndex, flushIndex
return dedupIdx, flushIdx
}
func getStateIdx(interval int64, ts int64) int {
return int(ts/interval) % aggrStateSize
}
func (a *aggregator) dedupFlush(deleteDeadline int64, dedupIdx, flushIdx int) {
@ -1000,7 +996,6 @@ func (a *aggregator) Push(tss []prompbmarshal.TimeSeries, matchIdxs []byte) {
continue
}
if ignoreOldSamples && s.Timestamp < minTimestamp {
a.ignoredOldSamples.Inc()
// Skip old samples outside the current aggregation interval
a.ignoredOldSamples.Inc()
continue
@ -1010,7 +1005,7 @@ func (a *aggregator) Push(tss []prompbmarshal.TimeSeries, matchIdxs []byte) {
maxLagMsec = lagMsec
}
if ignoreOldSamples {
flushIdx = int((s.Timestamp)/a.tickInterval+1) % aggrStateSize
flushIdx = getStateIdx(a.tickInterval, s.Timestamp)
}
samples[flushIdx] = append(samples[flushIdx], pushSample{
key: key,
@ -1019,20 +1014,20 @@ func (a *aggregator) Push(tss []prompbmarshal.TimeSeries, matchIdxs []byte) {
})
}
}
if len(samples) > 0 {
a.matchedSamples.Add(len(samples))
a.samplesLag.Update(float64(maxLagMsec) / 1_000)
}
ctx.samples = samples
ctx.buf = buf
pushSamples := a.pushSamples
if a.da != nil {
for idx, s := range samples {
a.da.pushSamples(s, idx)
}
} else {
for idx, s := range samples {
a.pushSamples(s, deleteDeadlineMsec, idx)
pushSamples = a.da.pushSamples
}
for idx, s := range samples {
if len(s) > 0 {
a.samplesLag.Update(float64(maxLagMsec) / 1_000)
a.matchedSamples.Add(len(s))
pushSamples(s, deleteDeadlineMsec, idx)
}
}
}

View file

@ -79,10 +79,8 @@ func (as *uniqueSamplesAggrState) flushState(ctx *flushCtx) {
state := len(sv.state[ctx.idx])
sv.state[ctx.idx] = make(map[float64]struct{})
sv.mu.Unlock()
if state > 0 {
key := k.(string)
ctx.appendSeries(key, "unique_samples", float64(state))
}
key := k.(string)
ctx.appendSeries(key, "unique_samples", float64(state))
return true
})
}