From 984dd2abafb3fca6ec700c29b23314a5f1692472 Mon Sep 17 00:00:00 2001
From: Aliaksandr Valialkin <valyala@victoriametrics.com>
Date: Thu, 25 Apr 2024 00:57:48 +0200
Subject: [PATCH] lib/logstorage: optimize typical AND case

---
 lib/logstorage/filters.go | 66 +++++++++++++++++++++++++++++++++++----
 1 file changed, 60 insertions(+), 6 deletions(-)

diff --git a/lib/logstorage/filters.go b/lib/logstorage/filters.go
index 2c75d65cf..aab59d591 100644
--- a/lib/logstorage/filters.go
+++ b/lib/logstorage/filters.go
@@ -214,6 +214,9 @@ func (of *orFilter) apply(bs *blockSearch, bm *filterBitmap) {
 // It is expressed as `f1 AND f2 ... AND fN` in LogsQL.
 type andFilter struct {
 	filters []filter
+
+	msgTokensOnce sync.Once
+	msgTokens     []string
 }
 
 func (af *andFilter) String() string {
@@ -237,6 +240,22 @@ func (af *andFilter) updateReferencedColumnNames(m map[string]struct{}) {
 }
 
 func (af *andFilter) apply(bs *blockSearch, bm *filterBitmap) {
+	if tokens := af.getMsgTokens(); len(tokens) > 0 {
+		// Verify whether af tokens for the _msg field match bloom filter.
+		ch := bs.csh.getColumnHeader("_msg")
+		if ch == nil {
+			// Fast path - there is no _msg field in the block.
+			bm.resetBits()
+			return
+		}
+		if !matchBloomFilterAllTokens(bs, ch, tokens) {
+			// Fast path - af tokens for the _msg field do not match bloom filter.
+			bm.resetBits()
+			return
+		}
+	}
+
+	// Slow path - verify every filter separately.
 	for _, f := range af.filters {
 		f.apply(bs, bm)
 		if bm.isZero() {
@@ -247,6 +266,40 @@ func (af *andFilter) apply(bs *blockSearch, bm *filterBitmap) {
 	}
 }
 
+func (af *andFilter) getMsgTokens() []string {
+	af.msgTokensOnce.Do(af.initMsgTokens)
+	return af.msgTokens
+}
+
+func (af *andFilter) initMsgTokens() {
+	var a []string
+	for _, f := range af.filters {
+		switch t := f.(type) {
+		case *phraseFilter:
+			if isMsgFieldName(t.fieldName) {
+				a = append(a, t.getTokens()...)
+			}
+		case *sequenceFilter:
+			if isMsgFieldName(t.fieldName) {
+				a = append(a, t.getTokens()...)
+			}
+		case *exactFilter:
+			if isMsgFieldName(t.fieldName) {
+				a = append(a, t.getTokens()...)
+			}
+		case *exactPrefixFilter:
+			if isMsgFieldName(t.fieldName) {
+				a = append(a, t.getTokens()...)
+			}
+		case *prefixFilter:
+			if isMsgFieldName(t.fieldName) {
+				a = append(a, t.getTokens()...)
+			}
+		}
+	}
+	af.msgTokens = a
+}
+
 // notFilter negates the filter.
 //
 // It is expressed as `NOT f` or `!f` in LogsQL.
@@ -2933,11 +2986,8 @@ type stringBucket struct {
 }
 
 func (sb *stringBucket) reset() {
-	a := sb.a
-	for i := range a {
-		a[i] = ""
-	}
-	sb.a = a[:0]
+	clear(sb.a)
+	sb.a = sb.a[:0]
 }
 
 func getStringBucket() *stringBucket {
@@ -2983,12 +3033,16 @@ func toUint64Clamp(f float64) uint64 {
 }
 
 func quoteFieldNameIfNeeded(s string) string {
-	if s == "_msg" || s == "" {
+	if isMsgFieldName(s) {
 		return ""
 	}
 	return quoteTokenIfNeeded(s) + ":"
 }
 
+func isMsgFieldName(fieldName string) bool {
+	return fieldName == "" || fieldName == "_msg"
+}
+
 func toUint8String(bs *blockSearch, bb *bytesutil.ByteBuffer, v string) string {
 	if len(v) != 1 {
 		logger.Panicf("FATAL: %s: unexpected length for binary representation of uint8 number: got %d; want 1", bs.partPath(), len(v))