MB-66395: Support for batch processing of vector search requests

Requires:
- blevesearch/scorch_segment_api#62
- blevesearch/zapx#332

Author: abhinavdangeti

go.mod

@@ -14,7 +14,7 @@ require (
 	github.com/blevesearch/go-porterstemmer v1.0.3
 	github.com/blevesearch/goleveldb v1.0.1
 	github.com/blevesearch/gtreap v0.1.1
-	github.com/blevesearch/scorch_segment_api/v2 v2.3.10
+	github.com/blevesearch/scorch_segment_api/v2 v2.3.11-0.20250523184632-44c633af14bf
 	github.com/blevesearch/segment v0.9.1
 	github.com/blevesearch/snowball v0.6.1
 	github.com/blevesearch/snowballstem v0.9.0
@@ -26,7 +26,7 @@ require (
 	github.com/blevesearch/zapx/v13 v13.4.2
 	github.com/blevesearch/zapx/v14 v14.4.2
 	github.com/blevesearch/zapx/v15 v15.4.2
-	github.com/blevesearch/zapx/v16 v16.2.3
+	github.com/blevesearch/zapx/v16 v16.2.4-0.20250523184909-e4999955331b
 	github.com/couchbase/moss v0.2.0
 	github.com/golang/protobuf v1.3.2
 	github.com/spf13/cobra v1.8.1

go.sum

@@ -20,8 +20,8 @@ github.com/blevesearch/gtreap v0.1.1/go.mod h1:QaQyDRAT51sotthUWAH4Sj08awFSSWzgY
 github.com/blevesearch/mmap-go v1.0.2/go.mod h1:ol2qBqYaOUsGdm7aRMRrYGgPvnwLe6Y+7LMvAB5IbSA=
 github.com/blevesearch/mmap-go v1.0.4 h1:OVhDhT5B/M1HNPpYPBKIEJaD0F3Si+CrEKULGCDPWmc=
 github.com/blevesearch/mmap-go v1.0.4/go.mod h1:EWmEAOmdAS9z/pi/+Toxu99DnsbhG1TIxUoRmJw/pSs=
-github.com/blevesearch/scorch_segment_api/v2 v2.3.10 h1:Yqk0XD1mE0fDZAJXTjawJ8If/85JxnLd8v5vG/jWE/s=
-github.com/blevesearch/scorch_segment_api/v2 v2.3.10/go.mod h1:Z3e6ChN3qyN35yaQpl00MfI5s8AxUJbpTR/DL8QOQ+8=
+github.com/blevesearch/scorch_segment_api/v2 v2.3.11-0.20250523184632-44c633af14bf h1:qkKQJ4gw00t82UEVPMVObAaGl+5iC8jMCdwNykEqKoM=
+github.com/blevesearch/scorch_segment_api/v2 v2.3.11-0.20250523184632-44c633af14bf/go.mod h1:Z3e6ChN3qyN35yaQpl00MfI5s8AxUJbpTR/DL8QOQ+8=
 github.com/blevesearch/segment v0.9.1 h1:+dThDy+Lvgj5JMxhmOVlgFfkUtZV2kw49xax4+jTfSU=
 github.com/blevesearch/segment v0.9.1/go.mod h1:zN21iLm7+GnBHWTao9I+Au/7MBiL8pPFtJBJTsk6kQw=
 github.com/blevesearch/snowball v0.6.1 h1:cDYjn/NCH+wwt2UdehaLpr2e4BwLIjN4V/TdLsL+B5A=
@@ -44,8 +44,8 @@ github.com/blevesearch/zapx/v14 v14.4.2 h1:2SGHakVKd+TrtEqpfeq8X+So5PShQ5nW6GNxT
 github.com/blevesearch/zapx/v14 v14.4.2/go.mod h1:rz0XNb/OZSMjNorufDGSpFpjoFKhXmppH9Hi7a877D8=
 github.com/blevesearch/zapx/v15 v15.4.2 h1:sWxpDE0QQOTjyxYbAVjt3+0ieu8NCE0fDRaFxEsp31k=
 github.com/blevesearch/zapx/v15 v15.4.2/go.mod h1:1pssev/59FsuWcgSnTa0OeEpOzmhtmr/0/11H0Z8+Nw=
-github.com/blevesearch/zapx/v16 v16.2.3 h1:7Y0r+a3diEvlazsncexq1qoFOcBd64xwMS7aDm4lo1s=
-github.com/blevesearch/zapx/v16 v16.2.3/go.mod h1:wVJ+GtURAaRG9KQAMNYyklq0egV+XJlGcXNCE0OFjjA=
+github.com/blevesearch/zapx/v16 v16.2.4-0.20250523184909-e4999955331b h1:U5hPYxsnmffW4rzNuI4651GOboySSPR03fS8feG5PZI=
+github.com/blevesearch/zapx/v16 v16.2.4-0.20250523184909-e4999955331b/go.mod h1:sQ1SoGL1DpaPmAlwSVTRHmoi7cGXPRJGP0U+ZDtSwtY=
 github.com/couchbase/ghistogram v0.1.0 h1:b95QcQTCzjTUocDXp/uMgSNQi8oj1tGwnJ4bODWZnps=
 github.com/couchbase/ghistogram v0.1.0/go.mod h1:s1Jhy76zqfEecpNWJfWUiKZookAFaiGOEoyzgHt9i7k=
 github.com/couchbase/moss v0.2.0 h1:VCYrMzFwEryyhRSeI+/b3tRBSeTpi/8gn5Kf6dxqn+o=

index/scorch/optimize_knn.go

@@ -38,8 +38,12 @@ type OptimizeVR struct {
 	requiresFiltering bool
 }
 
-// This setting _MUST_ only be changed during init and not after.
-var BleveMaxKNNConcurrency = 10
+// These settings _MUST_ only be changed during init and not after.
+var (
+	BleveMaxKNNConcurrency               = 10
+	BleveVectorSearchBatchExecution      = false
+	BleveVectorSearchBatchExecutionDelay = segment_api.DefaultBatchExecutionDelay
+)
 
 func (o *OptimizeVR) invokeSearcherEndCallback() {
 	if o.ctx != nil {
@@ -80,7 +84,11 @@ func (o *OptimizeVR) Finish() error {
 				}()
 				for field, vrs := range o.vrs {
 					vecIndex, err := segment.InterpretVectorIndex(field,
-						o.requiresFiltering, origSeg.deleted)
+						o.requiresFiltering, origSeg.deleted,
+						segment_api.InterpretVectorIndexOptions{
+							Batch:               BleveVectorSearchBatchExecution,
+							BatchExecutionDelay: BleveVectorSearchBatchExecutionDelay,
+						})
 					if err != nil {
 						errorsM.Lock()
 						errors = append(errors, err)

search_knn_test.go

@@ -1644,7 +1644,6 @@ func TestNestedVectors(t *testing.T) {
 }
 
 func TestNumVecsStat(t *testing.T) {
-
 	dataset, _, err := readDatasetAndQueries(testInputCompressedFile)
 	if err != nil {
 		t.Fatal(err)
@@ -1701,3 +1700,108 @@ func TestNumVecsStat(t *testing.T) {
 		}
 	}
 }
+
+func TestSerialAndBatchRequestsToVectorIndex(t *testing.T) {
+	tmpIndexPath := createTmpIndexPath(t)
+	defer cleanupTmpIndexPath(t, tmpIndexPath)
+
+	docs := []map[string]interface{}{
+		{
+			"id":  "1",
+			"vec": []float32{1.0, 2.0, 3.0},
+		},
+		{
+			"id":  "2",
+			"vec": []float32{4.0, 5.0, 6.0},
+		},
+		{
+			"id":  "3",
+			"vec": []float32{7.0, 8.0, 9.0},
+		},
+		{
+			"id":  "4",
+			"vec": []float32{10.0, 11.0, 12.0},
+		},
+		{
+			"id":  "5",
+			"vec": []float32{13.0, 14.0, 15.0},
+		},
+		{
+			"id":  "6",
+			"vec": []float32{16.0, 17.0, 18.0},
+		},
+		{
+			"id":  "7",
+			"vec": []float32{19.0, 20.0, 21.0},
+		},
+		{
+			"id":  "8",
+			"vec": []float32{22.0, 23.0, 24.0},
+		},
+		{
+			"id":  "9",
+			"vec": []float32{25.0, 26.0, 27.0},
+		},
+		{
+			"id":  "10",
+			"vec": []float32{28.0, 29.0, 30.0},
+		},
+	}
+	// Index mapping
+	indexMapping := NewIndexMapping()
+	vm := mapping.NewVectorFieldMapping()
+	vm.Dims = 3
+	vm.Similarity = "l2_norm"
+	indexMapping.DefaultMapping.AddFieldMappingsAt("vec", vm)
+
+	// Create index and upload documents
+	idx, err := New(tmpIndexPath, indexMapping)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer func() {
+		err := idx.Close()
+		if err != nil {
+			t.Fatal(err)
+		}
+	}()
+
+	batch := idx.NewBatch()
+	for _, doc := range docs {
+		err = batch.Index(doc["id"].(string), doc)
+		if err != nil {
+			t.Fatal(err)
+		}
+	}
+
+	err = idx.Batch(batch)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	tests := []struct {
+		queryVec []float32
+	}{
+		{
+			queryVec: []float32{2, 5, 8},
+		},
+		{
+			queryVec: []float32{11, 14, 17},
+		},
+		{
+			queryVec: []float32{2, 5, 8, 11, 14, 17},
+		},
+	}
+
+	for _, test := range tests {
+		searchReq := NewSearchRequest(query.NewMatchNoneQuery())
+		searchReq.AddKNN("vec", test.queryVec, 3, 1)
+
+		res, err := idx.Search(searchReq)
+		if err != nil {
+			t.Fatal(err)
+		}
+
+		fmt.Println(test.queryVec, res.Hits)
+	}
+}