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

lib/streamaggr/avg.go

@@ -1,9 +1,9 @@
 package streamaggr
 
 import (
-	"strings"
 	"sync"
 
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
 )
 
@@ -36,7 +36,8 @@ func (as *avgAggrState) pushSamples(samples []pushSample) {
 				sum:   s.value,
 				count: 1,
 			}
-			vNew, loaded := as.m.LoadOrStore(strings.Clone(outputKey), v)
+			outputKey = bytesutil.InternString(outputKey)
+			vNew, loaded := as.m.LoadOrStore(outputKey, v)
 			if !loaded {
 				// The entry has been successfully stored
 				continue

lib/streamaggr/count_samples.go

@@ -1,9 +1,9 @@
 package streamaggr
 
 import (
-	"strings"
 	"sync"
 
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
 )
 
@@ -34,7 +34,8 @@ func (as *countSamplesAggrState) pushSamples(samples []pushSample) {
 			v = &countSamplesStateValue{
 				n: 1,
 			}
-			vNew, loaded := as.m.LoadOrStore(strings.Clone(outputKey), v)
+			outputKey = bytesutil.InternString(outputKey)
+			vNew, loaded := as.m.LoadOrStore(outputKey, v)
 			if !loaded {
 				// The new entry has been successfully created.
 				continue

lib/streamaggr/count_series.go

@@ -1,7 +1,6 @@
 package streamaggr
 
 import (
-	"strings"
 	"sync"
 
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
@@ -42,7 +41,8 @@ func (as *countSeriesAggrState) pushSamples(samples []pushSample) {
 					h: {},
 				},
 			}
-			vNew, loaded := as.m.LoadOrStore(strings.Clone(outputKey), v)
+			outputKey = bytesutil.InternString(outputKey)
+			vNew, loaded := as.m.LoadOrStore(outputKey, v)
 			if !loaded {
 				// The entry has been added to the map.
 				continue

lib/streamaggr/dedup.go

@@ -1,7 +1,6 @@
 package streamaggr
 
 import (
-	"strings"
 	"sync"
 	"unsafe"
 
@@ -25,7 +24,7 @@ type dedupAggrShard struct {
 
 type dedupAggrShardNopad struct {
 	mu sync.Mutex
-	m  map[string]*dedupAggrSample
+	m  map[string]dedupAggrSample
 }
 
 type dedupAggrSample struct {
@@ -170,13 +169,14 @@ func (das *dedupAggrShard) pushSamples(samples []pushSample) {
 
 	m := das.m
 	if m == nil {
-		m = make(map[string]*dedupAggrSample, len(samples))
+		m = make(map[string]dedupAggrSample, len(samples))
 		das.m = m
 	}
 	for _, sample := range samples {
 		s, ok := m[sample.key]
 		if !ok {
-			m[strings.Clone(sample.key)] = &dedupAggrSample{
+			key := bytesutil.InternString(sample.key)
+			m[key] = dedupAggrSample{
 				value:     sample.value,
 				timestamp: sample.timestamp,
 			}
@@ -184,8 +184,11 @@ func (das *dedupAggrShard) pushSamples(samples []pushSample) {
 		}
 		// Update the existing value according to logic described at https://docs.victoriametrics.com/#deduplication
 		if sample.timestamp > s.timestamp || (sample.timestamp == s.timestamp && sample.value > s.value) {
-			s.value = sample.value
+			key := bytesutil.InternString(sample.key)
-			s.timestamp = sample.timestamp
+			m[key] = dedupAggrSample{
+				value:     sample.value,
+				timestamp: sample.timestamp,
+			}
 		}
 	}
 }
@@ -195,7 +198,7 @@ func (das *dedupAggrShard) flush(ctx *dedupFlushCtx, f func(samples []pushSample
 
 	m := das.m
 	if resetState && len(m) > 0 {
-		das.m = make(map[string]*dedupAggrSample, len(m))
+		das.m = make(map[string]dedupAggrSample, len(m))
 	}
 
 	das.mu.Unlock()

lib/streamaggr/deduplicator.go

@@ -94,8 +94,9 @@ func (d *Deduplicator) Push(tss []prompbmarshal.TimeSeries) {
 		}
 		labels.Sort()
 
-		buf = lc.Compress(buf[:0], labels.Labels)
+		bufLen := len(buf)
-		key := bytesutil.InternBytes(buf)
+		buf = lc.Compress(buf, labels.Labels)
+		key := bytesutil.ToUnsafeString(buf[bufLen:])
 		for _, s := range ts.Samples {
 			pss = append(pss, pushSample{
 				key:       key,

lib/streamaggr/histogram_bucket.go

@@ -2,10 +2,10 @@ package streamaggr
 
 import (
 	"math"
-	"strings"
 	"sync"
 	"time"
 
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
 	"github.com/VictoriaMetrics/metrics"
 )
@@ -43,7 +43,8 @@ func (as *histogramBucketAggrState) pushSamples(samples []pushSample) {
 		if !ok {
 			// The entry is missing in the map. Try creating it.
 			v = &histogramBucketStateValue{}
-			vNew, loaded := as.m.LoadOrStore(strings.Clone(outputKey), v)
+			outputKey = bytesutil.InternString(outputKey)
+			vNew, loaded := as.m.LoadOrStore(outputKey, v)
 			if loaded {
 				// Use the entry created by a concurrent goroutine.
 				v = vNew

lib/streamaggr/last.go

@@ -1,9 +1,9 @@
 package streamaggr
 
 import (
-	"strings"
 	"sync"
 
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
 )
 
@@ -36,7 +36,8 @@ func (as *lastAggrState) pushSamples(samples []pushSample) {
 				last:      s.value,
 				timestamp: s.timestamp,
 			}
-			vNew, loaded := as.m.LoadOrStore(strings.Clone(outputKey), v)
+			outputKey = bytesutil.InternString(outputKey)
+			vNew, loaded := as.m.LoadOrStore(outputKey, v)
 			if !loaded {
 				// The new entry has been successfully created.
 				continue

lib/streamaggr/max.go

@@ -1,9 +1,9 @@
 package streamaggr
 
 import (
-	"strings"
 	"sync"
 
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
 )
 
@@ -34,7 +34,8 @@ func (as *maxAggrState) pushSamples(samples []pushSample) {
 			v = &maxStateValue{
 				max: s.value,
 			}
-			vNew, loaded := as.m.LoadOrStore(strings.Clone(outputKey), v)
+			outputKey = bytesutil.InternString(outputKey)
+			vNew, loaded := as.m.LoadOrStore(outputKey, v)
 			if !loaded {
 				// The new entry has been successfully created.
 				continue

lib/streamaggr/min.go

@@ -1,9 +1,9 @@
 package streamaggr
 
 import (
-	"strings"
 	"sync"
 
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
 )
 
@@ -34,7 +34,8 @@ func (as *minAggrState) pushSamples(samples []pushSample) {
 			v = &minStateValue{
 				min: s.value,
 			}
-			vNew, loaded := as.m.LoadOrStore(strings.Clone(outputKey), v)
+			outputKey = bytesutil.InternString(outputKey)
+			vNew, loaded := as.m.LoadOrStore(outputKey, v)
 			if !loaded {
 				// The new entry has been successfully created.
 				continue

lib/streamaggr/quantiles.go

@@ -2,7 +2,6 @@ package streamaggr
 
 import (
 	"strconv"
-	"strings"
 	"sync"
 
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
@@ -42,7 +41,8 @@ func (as *quantilesAggrState) pushSamples(samples []pushSample) {
 			v = &quantilesStateValue{
 				h: h,
 			}
-			vNew, loaded := as.m.LoadOrStore(strings.Clone(outputKey), v)
+			outputKey = bytesutil.InternString(outputKey)
+			vNew, loaded := as.m.LoadOrStore(outputKey, v)
 			if loaded {
 				// Use the entry created by a concurrent goroutine.
 				histogram.PutFast(h)

lib/streamaggr/rate.go

@@ -4,6 +4,7 @@ import (
 	"sync"
 	"time"
 
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
 )
 
@@ -59,6 +60,7 @@ func (as *rateAggrState) pushSamples(samples []pushSample) {
 			v = &rateStateValue{
 				lastValues: make(map[string]rateLastValueState),
 			}
+			outputKey = bytesutil.InternString(outputKey)
 			vNew, loaded := as.m.LoadOrStore(outputKey, v)
 			if loaded {
 				// Use the entry created by a concurrent goroutine.
@@ -89,6 +91,8 @@ func (as *rateAggrState) pushSamples(samples []pushSample) {
 			lv.value = s.value
 			lv.timestamp = s.timestamp
 			lv.deleteDeadline = deleteDeadline
+
+			inputKey = bytesutil.InternString(inputKey)
 			sv.lastValues[inputKey] = lv
 			sv.deleteDeadline = deleteDeadline
 		}

lib/streamaggr/stddev.go

@@ -2,9 +2,9 @@ package streamaggr
 
 import (
 	"math"
-	"strings"
 	"sync"
 
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
 )
 
@@ -35,7 +35,8 @@ func (as *stddevAggrState) pushSamples(samples []pushSample) {
 		if !ok {
 			// The entry is missing in the map. Try creating it.
 			v = &stddevStateValue{}
-			vNew, loaded := as.m.LoadOrStore(strings.Clone(outputKey), v)
+			outputKey = bytesutil.InternString(outputKey)
+			vNew, loaded := as.m.LoadOrStore(outputKey, v)
 			if loaded {
 				// Use the entry created by a concurrent goroutine.
 				v = vNew

lib/streamaggr/stdvar.go

@@ -1,9 +1,9 @@
 package streamaggr
 
 import (
-	"strings"
 	"sync"
 
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
 )
 
@@ -34,7 +34,8 @@ func (as *stdvarAggrState) pushSamples(samples []pushSample) {
 		if !ok {
 			// The entry is missing in the map. Try creating it.
 			v = &stdvarStateValue{}
-			vNew, loaded := as.m.LoadOrStore(strings.Clone(outputKey), v)
+			outputKey = bytesutil.InternString(outputKey)
+			vNew, loaded := as.m.LoadOrStore(outputKey, v)
 			if loaded {
 				// Use the entry created by a concurrent goroutine.
 				v = vNew

lib/streamaggr/streamaggr.go

@@ -398,7 +398,11 @@ type aggregator struct {
 }
 
 type aggrState interface {
+	// pushSamples must push samples to the aggrState.
+	//
+	// samples[].key must be cloned by aggrState, since it may change after returning from pushSamples.
 	pushSamples(samples []pushSample)
+
 	flushState(ctx *flushCtx, resetState bool)
 }
 

lib/streamaggr/sum_samples.go

@@ -1,9 +1,9 @@
 package streamaggr
 
 import (
-	"strings"
 	"sync"
 
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
 )
 
@@ -34,7 +34,8 @@ func (as *sumSamplesAggrState) pushSamples(samples []pushSample) {
 			v = &sumSamplesStateValue{
 				sum: s.value,
 			}
-			vNew, loaded := as.m.LoadOrStore(strings.Clone(outputKey), v)
+			outputKey = bytesutil.InternString(outputKey)
+			vNew, loaded := as.m.LoadOrStore(outputKey, v)
 			if !loaded {
 				// The new entry has been successfully created.
 				continue

lib/streamaggr/total.go

@@ -2,10 +2,10 @@ package streamaggr
 
 import (
 	"math"
-	"strings"
 	"sync"
 	"time"
 
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
 )
 
@@ -81,7 +81,8 @@ func (as *totalAggrState) pushSamples(samples []pushSample) {
 			v = &totalStateValue{
 				lastValues: make(map[string]totalLastValueState),
 			}
-			vNew, loaded := as.m.LoadOrStore(strings.Clone(outputKey), v)
+			outputKey = bytesutil.InternString(outputKey)
+			vNew, loaded := as.m.LoadOrStore(outputKey, v)
 			if loaded {
 				// Use the entry created by a concurrent goroutine.
 				v = vNew
@@ -109,9 +110,8 @@ func (as *totalAggrState) pushSamples(samples []pushSample) {
 			lv.value = s.value
 			lv.timestamp = s.timestamp
 			lv.deleteDeadline = deleteDeadline
-			if !ok {
+
-				inputKey = strings.Clone(inputKey)
+			inputKey = bytesutil.InternString(inputKey)
-			}
 			sv.lastValues[inputKey] = lv
 			sv.deleteDeadline = deleteDeadline
 		}

lib/streamaggr/unique_samples.go

@@ -1,9 +1,9 @@
 package streamaggr
 
 import (
-	"strings"
 	"sync"
 
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
 )
 
@@ -36,7 +36,8 @@ func (as *uniqueSamplesAggrState) pushSamples(samples []pushSample) {
 					s.value: {},
 				},
 			}
-			vNew, loaded := as.m.LoadOrStore(strings.Clone(outputKey), v)
+			outputKey = bytesutil.InternString(outputKey)
+			vNew, loaded := as.m.LoadOrStore(outputKey, v)
 			if !loaded {
 				// The new entry has been successfully created.
 				continue