Speedup DistinctValue collector and exit early for ingesters

CHANGELOG.md

@@ -1,5 +1,6 @@
 ## main / unreleased
 
+* [ENHANCEMENT] Speedup DistinctValue collector and exit early for ingesters [#4104](https://github.com/grafana/tempo/pull/4104) (@electron0zero)
 * [ENHANCEMENT] Add disk caching in ingester SearchTagValuesV2 for completed blocks [#4069](https://github.com/grafana/tempo/pull/4069) (@electron0zero)
 * [BUGFIX] Replace hedged requests roundtrips total with a counter. [#4063](https://github.com/grafana/tempo/pull/4063) [#4078](https://github.com/grafana/tempo/pull/4078) (@galalen)
 * [CHANGE] TraceByID: don't allow concurrent_shards greater than query_shards. [#4074](https://github.com/grafana/tempo/pull/4074) (@electron0zero)

modules/frontend/combiner/search_tag_values.go

@@ -43,8 +43,8 @@ func NewTypedSearchTagValues(limitBytes int) GRPCCombiner[*tempopb.SearchTagValu
 }
 
 func NewSearchTagValuesV2(limitBytes int) Combiner {
-	// Distinct collector with no limit
-	d := collector.NewDistinctValue(limitBytes, func(tv tempopb.TagValue) int { return len(tv.Type) + len(tv.Value) })
+	// Distinct collector with no limit and diff enabled
+	d := collector.NewDistinctValueWithDiff(limitBytes, func(tv tempopb.TagValue) int { return len(tv.Type) + len(tv.Value) })
 
 	return &genericCombiner[*tempopb.SearchTagValuesV2Response]{
 		httpStatusCode: 200,

modules/ingester/instance_search.go

@@ -573,7 +573,7 @@ func (i *instance) SearchTagValuesV2(ctx context.Context, req *tempopb.SearchTag
 	}
 
 	if valueCollector.Exceeded() {
-		level.Warn(log.Logger).Log("msg", "size of tag values in instance exceeded limit, reduce cardinality or size of tags", "tag", req.TagName, "userID", userID, "limit", limit, "total", valueCollector.TotalDataSize())
+		_ = level.Warn(log.Logger).Log("msg", "size of tag values exceeded limit, reduce cardinality or size of tags", "tag", req.TagName, "userID", userID, "limit", limit, "size", valueCollector.Size())
 	}
 
 	resp := &tempopb.SearchTagValuesV2Response{}

modules/querier/querier.go

@@ -504,6 +504,9 @@ func (q *Querier) SearchTagsBlocks(ctx context.Context, req *tempopb.SearchTagsB
 
 		for _, t := range s.Tags {
 			distinctValues.Collect(t)
+			if distinctValues.Exceeded() {
+				break // stop early
+			}
 		}
 	}
 
@@ -539,9 +542,14 @@ func (q *Querier) SearchTags(ctx context.Context, req *tempopb.SearchTagsRequest
 	if err != nil {
 		return nil, fmt.Errorf("error querying ingesters in Querier.SearchTags: %w", err)
 	}
+
+outerLoop:
 	for _, resp := range lookupResults {
 		for _, res := range resp.response.(*tempopb.SearchTagsResponse).TagNames {
 			distinctValues.Collect(res)
+			if distinctValues.Exceeded() {
+				break outerLoop // break out of all loops
+			}
 		}
 	}
 
@@ -573,10 +581,14 @@ func (q *Querier) SearchTagsV2(ctx context.Context, req *tempopb.SearchTagsReque
 	limit := q.limits.MaxBytesPerTagValuesQuery(userID)
 	distinctValues := collector.NewScopedDistinctString(limit)
 
+outerLoop:
 	for _, resp := range lookupResults {
 		for _, res := range resp.response.(*tempopb.SearchTagsV2Response).Scopes {
 			for _, tag := range res.Tags {
 				distinctValues.Collect(res.Name, tag)
+				if distinctValues.Exceeded() {
+					break outerLoop // break out of all loops
+				}
 			}
 		}
 	}
@@ -610,6 +622,7 @@ func (q *Querier) SearchTagValues(ctx context.Context, req *tempopb.SearchTagVal
 
 	// Virtual tags values. Get these first.
 	for _, v := range search.GetVirtualTagValues(req.TagName) {
+		// virtual tags are small so no need to stop early here
 		distinctValues.Collect(v)
 	}
 
@@ -619,9 +632,14 @@ func (q *Querier) SearchTagValues(ctx context.Context, req *tempopb.SearchTagVal
 	if err != nil {
 		return nil, fmt.Errorf("error querying ingesters in Querier.SearchTagValues: %w", err)
 	}
+
+outerLoop:
 	for _, resp := range lookupResults {
 		for _, res := range resp.response.(*tempopb.SearchTagValuesResponse).TagValues {
 			distinctValues.Collect(res)
+			if distinctValues.Exceeded() {
+				break outerLoop // break out of all loops
+			}
 		}
 	}
 
@@ -648,6 +666,7 @@ func (q *Querier) SearchTagValuesV2(ctx context.Context, req *tempopb.SearchTagV
 	// Virtual tags values. Get these first.
 	virtualVals := search.GetVirtualTagValuesV2(req.TagName)
 	for _, v := range virtualVals {
+		// no need to stop early here, virtual tags are small
 		distinctValues.Collect(v)
 	}
 
@@ -664,14 +683,18 @@ func (q *Querier) SearchTagValuesV2(ctx context.Context, req *tempopb.SearchTagV
 	if err != nil {
 		return nil, fmt.Errorf("error querying ingesters in Querier.SearchTagValues: %w", err)
 	}
+
+outerLoop:
 	for _, resp := range lookupResults {
 		for _, res := range resp.response.(*tempopb.SearchTagValuesV2Response).TagValues {
-			distinctValues.Collect(*res)
+			if distinctValues.Collect(*res) {
+				break outerLoop // break out of all loops
+			}
 		}
 	}
 
 	if distinctValues.Exceeded() {
-		level.Warn(log.Logger).Log("msg", "size of tag values in instance exceeded limit, reduce cardinality or size of tags", "tag", req.TagName, "userID", userID, "limit", limit, "total", distinctValues.TotalDataSize())
+		_ = level.Warn(log.Logger).Log("msg", "size of tag values exceeded limit, reduce cardinality or size of tags", "tag", req.TagName, "userID", userID, "limit", limit, "size", distinctValues.Size())
 	}
 
 	return valuesToV2Response(distinctValues), nil

pkg/collector/distinct_string_collector.go

@@ -24,6 +24,8 @@ func NewDistinctString(maxDataSize int) *DistinctString {
 	}
 }
 
+// Collect adds a new value to the distinct string collector.
+// return indicates if the value was added or not.
 func (d *DistinctString) Collect(s string) bool {
 	if _, ok := d.values[s]; ok {
 		// Already present

pkg/collector/distinct_value_collector.go

@@ -5,95 +5,118 @@ import (
 )
 
 type DistinctValue[T comparable] struct {
-	values   map[T]struct{}
-	new      map[T]struct{}
-	len      func(T) int
-	maxLen   int
-	currLen  int
-	totalLen int
-	mtx      sync.RWMutex
+	values      map[T]struct{}
+	new         map[T]struct{}
+	len         func(T) int
+	maxLen      int
+	currLen     int
+	limExceeded bool
+	diffEnabled bool
+	mtx         sync.Mutex
 }
 
 // NewDistinctValue with the given maximum data size. This is calculated
 // as the total length of the recorded strings. For ease of use, maximum=0
 // is interpreted as unlimited.
+// Use NewDistinctValueWithDiff to enable diff support, but that one is slightly slower.
 func NewDistinctValue[T comparable](maxDataSize int, len func(T) int) *DistinctValue[T] {
 	return &DistinctValue[T]{
-		values: make(map[T]struct{}),
-		new:    make(map[T]struct{}),
-		maxLen: maxDataSize,
-		len:    len,
+		values:      make(map[T]struct{}),
+		new:         make(map[T]struct{}),
+		maxLen:      maxDataSize,
+		diffEnabled: false, // disable diff to make it faster
+		len:         len,
 	}
 }
 
+// NewDistinctValueWithDiff is like NewDistinctValue but with diff support enabled.
+func NewDistinctValueWithDiff[T comparable](maxDataSize int, len func(T) int) *DistinctValue[T] {
+	return &DistinctValue[T]{
+		values:      make(map[T]struct{}),
+		new:         make(map[T]struct{}),
+		maxLen:      maxDataSize,
+		diffEnabled: true,
+		len:         len,
+	}
+}
+
+// Collect adds a new value to the distinct value collector.
+// return true when it reaches the limits and can't fit more values.
+// callers of return of Collect or call Exceeded to stop early.
 func (d *DistinctValue[T]) Collect(v T) (exceeded bool) {
-	d.mtx.RLock()
-	if _, ok := d.values[v]; ok {
-		d.mtx.RUnlock()
-		return // Already present
+	d.mtx.Lock()
+	defer d.mtx.Unlock()
+
+	if d.limExceeded {
+		return true
 	}
-	d.mtx.RUnlock()
 
 	// Calculate length
 	valueLen := d.len(v)
 
-	d.mtx.Lock()
-	defer d.mtx.Unlock()
+	// Can it fit?
+	// note: we will stop adding values slightly before the limit is reached
+	if d.maxLen > 0 && d.currLen+valueLen >= d.maxLen {
+		// No, it can't fit
+		d.limExceeded = true
+		return true
+	}
 
 	if _, ok := d.values[v]; ok {
 		return // Already present
 	}
 
-	// Record total inspected length regardless
-	d.totalLen += valueLen
-
-	// Can it fit?
-	if d.maxLen > 0 && d.currLen+valueLen > d.maxLen {
-		// No
-		return true
+	if d.diffEnabled {
+		d.new[v] = struct{}{}
 	}
 
-	d.new[v] = struct{}{}
 	d.values[v] = struct{}{}
 	d.currLen += valueLen
+
 	return false
 }
 
 // Values returns the final list of distinct values collected and sorted.
 func (d *DistinctValue[T]) Values() []T {
-	ss := make([]T, 0, len(d.values))
-
-	d.mtx.RLock()
-	defer d.mtx.RUnlock()
+	d.mtx.Lock()
+	defer d.mtx.Unlock()
 
+	ss := make([]T, 0, len(d.values))
 	for k := range d.values {
 		ss = append(ss, k)
 	}
 
 	return ss
 }
 
-// Exceeded indicates if some values were lost because the maximum size limit was met.
+// Exceeded indicates that we have exceeded the limit
+// can be used to stop early and to avoid collecting further values
 func (d *DistinctValue[T]) Exceeded() bool {
-	d.mtx.RLock()
-	defer d.mtx.RUnlock()
-	return d.totalLen > d.currLen
+	d.mtx.Lock()
+	defer d.mtx.Unlock()
+
+	return d.limExceeded
 }
 
-// TotalDataSize is the total size of all distinct strings encountered.
-func (d *DistinctValue[T]) TotalDataSize() int {
-	d.mtx.RLock()
-	defer d.mtx.RUnlock()
-	return d.totalLen
+// Size is the total size of all distinct items collected
+func (d *DistinctValue[T]) Size() int {
+	d.mtx.Lock()
+	defer d.mtx.Unlock()
+
+	return d.currLen
 }
 
 // Diff returns all new strings collected since the last time diff was called
+// returns nil if diff is not enabled
 func (d *DistinctValue[T]) Diff() []T {
-	ss := make([]T, 0, len(d.new))
+	d.mtx.Lock()
+	defer d.mtx.Unlock()
 
-	d.mtx.RLock()
-	defer d.mtx.RUnlock()
+	if !d.diffEnabled {
+		return nil
+	}
 
+	ss := make([]T, 0, len(d.new))
 	for k := range d.new {
 		ss = append(ss, k)
 	}

pkg/collector/distinct_value_collector_test.go

@@ -1,14 +1,17 @@
 package collector
 
 import (
+	"fmt"
 	"sort"
+	"strconv"
 	"testing"
 
+	"github.com/grafana/tempo/pkg/tempopb"
 	"github.com/stretchr/testify/require"
 )
 
 func TestDistinctValueCollectorDiff(t *testing.T) {
-	d := NewDistinctValue[string](0, func(s string) int { return len(s) })
+	d := NewDistinctValueWithDiff[string](0, func(s string) int { return len(s) })
 
 	d.Collect("123")
 	d.Collect("4567")
@@ -28,3 +31,44 @@ func stringsSlicesEqual(t *testing.T, a, b []string) {
 	sort.Strings(b)
 	require.Equal(t, a, b)
 }
+
+func BenchmarkCollect(b *testing.B) {
+	// simulate 100 ingesters, each returning 10_000 tag values
+	numIngesters := 100
+	numTagValuesPerIngester := 10_000
+	ingesterTagValues := make([][]tempopb.TagValue, numIngesters)
+	for i := 0; i < numIngesters; i++ {
+		tagValues := make([]tempopb.TagValue, numTagValuesPerIngester)
+		for j := 0; j < numTagValuesPerIngester; j++ {
+			tagValues[j] = tempopb.TagValue{
+				Type:  "string",
+				Value: fmt.Sprintf("value_%d_%d", i, j),
+			}
+		}
+		ingesterTagValues[i] = tagValues
+	}
+
+	limits := []int{
+		0,          // no limit
+		100_000,    // 100KB
+		1_000_000,  // 1MB
+		10_000_000, // 10MB
+	}
+
+	b.ResetTimer() // to exclude the setup time for generating tag values
+	for _, lim := range limits {
+		b.Run("limit:"+strconv.Itoa(lim), func(b *testing.B) {
+			for n := 0; n < b.N; n++ {
+				// NewDistinctValue is collecting tag values without diff support
+				distinctValues := NewDistinctValue(lim, func(v tempopb.TagValue) int { return len(v.Type) + len(v.Value) })
+				for _, tagValues := range ingesterTagValues {
+					for _, v := range tagValues {
+						if distinctValues.Collect(v) {
+							break // stop early if limit is reached
+						}
+					}
+				}
+			}
+		})
+	}
+}