OAS-11787: Add Vector index feature and it's related testcases

Makefile

@@ -40,6 +40,10 @@ ifndef VST_ENABLED
 	VST_ENABLED := "false"
 endif
 
+ifndef ENABLE_VECTOR_INDEX
+	ENABLE_VECTOR_INDEX := "true"
+endif
+
 TESTV2PARALLEL ?= 4
 
 ORGPATH := github.com/arangodb
@@ -491,7 +495,7 @@ endif
 	@-docker rm -f -v $(TESTCONTAINER) &> /dev/null
 	@TESTCONTAINER=$(TESTCONTAINER) ARANGODB=$(ARANGODB) ALPINE_IMAGE=$(ALPINE_IMAGE) ENABLE_BACKUP=$(ENABLE_BACKUP) \
 	  ARANGO_LICENSE_KEY=$(ARANGO_LICENSE_KEY) STARTER=$(STARTER) STARTERMODE=$(TEST_MODE) TMPDIR="${TMPDIR}" \
-	  ENABLE_DATABASE_EXTRA_FEATURES=$(ENABLE_DATABASE_EXTRA_FEATURES) DEBUG_PORT=$(DEBUG_PORT) $(CLUSTERENV) DOCKER_NETWORK=${TEST_NET} "${ROOTDIR}/test/cluster.sh" start
+	  ENABLE_DATABASE_EXTRA_FEATURES=$(ENABLE_DATABASE_EXTRA_FEATURES) ENABLE_VECTOR_INDEX=$(ENABLE_VECTOR_INDEX) DEBUG_PORT=$(DEBUG_PORT) $(CLUSTERENV) DOCKER_NETWORK=${TEST_NET} "${ROOTDIR}/test/cluster.sh" start
 endif
 
 __test_cleanup:

test/cluster.sh

@@ -52,6 +52,9 @@ if [ "$CMD" == "start" ]; then
     if [ -n "$ENABLE_DATABASE_EXTRA_FEATURES" ]; then
         STARTERARGS="$STARTERARGS --all.database.extended-names-databases=true --args.all.http.compress-response-threshold=1 --args.all.http.handle-content-encoding-for-unauthenticated-requests=true"
     fi
+    if [ -n "$ENABLE_VECTOR_INDEX" ]; then
+        STARTERARGS="$STARTERARGS --args.all.experimental-vector-index=true"
+    fi
     # Use DOCKER_PLATFORM if set (e.g., from CircleCI for ARM), otherwise use macOS default
     if [ -n "$DOCKER_PLATFORM" ]; then
         DOCKERPLATFORMARG="$DOCKER_PLATFORM"

v2/CHANGELOG.md

@@ -5,6 +5,7 @@
 - Add ARM Support for V2 testcases 
 - Set TESTV2PARALLEL from 1 to 4
 - Disabled V8 related testcases in V1 and V2
+- Add Vector index feature
 
 ## [2.1.6](https://github.com/arangodb/go-driver/tree/v2.1.6) (2025-11-06)
 - Add missing endpoints from replication

v2/arangodb/collection_indexes.go

@@ -80,6 +80,12 @@ type CollectionIndexes interface {
 
 	// DeleteIndexByID deletes an index from the collection.
 	DeleteIndexByID(ctx context.Context, id string) error
+
+	// EnsureVectorIndex creates a vector index in the collection, if it does not already exist.
+	// The index is returned, together with a boolean indicating if the index was newly created (true) or pre-existing (false).
+	// Available in ArangoDB 3.12.4 and later.
+	// VectorParams is an obligatory parameter and must contain at least Dimension,Metric and NLists fields.
+	EnsureVectorIndex(ctx context.Context, fields []string, params *VectorParams, options *CreateVectorIndexOptions) (IndexResponse, bool, error)
 }
 
 // IndexType represents an index type as string
@@ -118,6 +124,9 @@ const (
 
 	// InvertedIndexType can be used to speed up a broad range of AQL queries, from simple to complex, including full-text search
 	InvertedIndexType = IndexType("inverted")
+
+	// VectorIndexType is used for efficient similarity searches on high-dimensional embeddings, enabling fast and scalable AI use cases.
+	VectorIndexType = IndexType("vector")
 )
 
 // IndexResponse is the response from the Index list method
@@ -135,6 +144,9 @@ type IndexResponse struct {
 
 	// InvertedIndex is the inverted index object. It is not empty only for InvertedIndex type.
 	InvertedIndex *InvertedIndexOptions `json:"invertedIndexes"`
+
+	// VectorIndex is the vector index params. It is not empty only for VectorIndex type.
+	VectorIndex *VectorParams `json:"params,omitempty"`
 }
 
 // IndexSharedOptions contains options that are shared between all index types
@@ -310,3 +322,45 @@ type CreateMDIPrefixedIndexOptions struct {
 	// Array expansions are not allowed.
 	PrefixFields []string `json:"prefixFields,required"`
 }
+
+type CreateVectorIndexOptions struct {
+	// Allow writes during creation.
+	InBackground *bool `json:"inBackground,omitempty"`
+	// Optional index name.
+	Name *string `json:"name,omitempty"`
+	// Number of threads to use for index creation.
+	Parallelism *int `json:"parallelism,omitempty"`
+	// Exclude docs missing the field.
+	Sparse *bool `json:"sparse,omitempty"`
+	// Introduced in v3.12.7
+	// Up to 32 additional attributes can be stored in the index.
+	StoredValues []string `json:"storedValues,omitempty"`
+}
+
+type VectorParams struct {
+	// Neighbors considered in search.
+	DefaultNProbe *int `json:"defaultNProbe,omitempty"`
+	// Vector length.
+	Dimension *int `json:"dimension,omitempty"`
+	// Faiss factory string.
+	Factory *string `json:"factory,omitempty"`
+	// Similarity measure.
+	Metric *VectorMetric `json:"metric,omitempty"`
+	// Number of centroids.
+	NLists *int `json:"nLists,omitempty"`
+	// Faiss training iterations
+	TrainingIterations *int `json:"trainingIterations,omitempty"`
+}
+
+// VectorMetric defines the type of similarity metric for vector comparison.
+type VectorMetric string
+
+const (
+	// Cosine similarity between vectors.
+	VectorMetricCosine VectorMetric = "cosine"
+	// Introduced in v3.12.6
+	// Inner product similarity.
+	VectorMetricInnerProduct VectorMetric = "innerProduct"
+	// Euclidean (L2) distance between vectors.
+	VectorMetricL2 VectorMetric = "l2"
+)

v2/arangodb/collection_indexes_impl.go

@@ -23,6 +23,7 @@ package arangodb
 import (
 	"context"
 	"encoding/json"
+	"fmt"
 	"net/http"
 	"net/url"
 
@@ -306,23 +307,111 @@ func (i *IndexResponse) UnmarshalJSON(data []byte) error {
 	i.Name = respSimple.Name
 	i.Type = respSimple.Type
 
-	if respSimple.Type == InvertedIndexType {
+	switch respSimple.Type {
+	case InvertedIndexType:
 		result := responseInvertedIndex{}
 		if err := json.Unmarshal(data, &result); err != nil {
 			return err
 		}
-
 		i.IndexSharedOptions = result.IndexSharedOptions
 		i.InvertedIndex = &result.InvertedIndexOptions
-	} else {
+	case VectorIndexType:
+		result := responseVectorIndex{}
+		if err := json.Unmarshal(data, &result); err != nil {
+			return err
+		}
+		i.IndexSharedOptions = result.IndexSharedOptions
+		i.VectorIndex = result.Params
+	default:
 		result := responseIndex{}
 		if err := json.Unmarshal(data, &result); err != nil {
 			return err
 		}
-
 		i.IndexSharedOptions = result.IndexSharedOptions
 		i.RegularIndex = &result.IndexOptions
 	}
+	return nil
+}
+
+func (p *VectorParams) validate() error {
+	if p == nil {
+		return errors.New("params must be provided for vector index")
+	}
+
+	if p.Dimension == nil || *p.Dimension <= 0 {
+		return errors.New("params.Dimension must be provided and greater than zero for vector index")
+	}
+
+	if p.Metric == nil {
+		return errors.New("params.Metric must be provided for vector index")
+	}
+
+	switch *p.Metric {
+	case VectorMetricCosine, VectorMetricL2, VectorMetricInnerProduct:
+		// valid
+	default:
+		return errors.New("params.Metric must be one of 'cosine', 'l2', or 'innerProduct' for vector index")
+	}
+
+	if p.NLists != nil && *p.NLists <= 0 {
+		return errors.New("params.NLists must be greater than zero for vector index")
+	}
 
 	return nil
 }
+
+func (c *collectionIndexes) EnsureVectorIndex(
+	ctx context.Context,
+	fields []string,
+	params *VectorParams,
+	options *CreateVectorIndexOptions,
+) (IndexResponse, bool, error) {
+
+	if len(fields) != 1 {
+		return IndexResponse{}, false, errors.New("vector index requires exactly one field")
+	}
+	if err := params.validate(); err != nil {
+		return IndexResponse{}, false, err
+	}
+
+	reqData := struct {
+		Type   IndexType     `json:"type"`
+		Fields []string      `json:"fields"`
+		Params *VectorParams `json:"params"`
+		*CreateVectorIndexOptions
+	}{
+		Type:                     VectorIndexType,
+		Fields:                   fields,
+		Params:                   params,
+		CreateVectorIndexOptions: options,
+	}
+
+	result := responseVectorIndex{}
+	fmt.Printf("reqData: %+v", reqData)
+	created, err := c.ensureIndex(ctx, &reqData, &result)
+	if err != nil {
+		return IndexResponse{}, false, err
+	}
+
+	return newVectorIndexResponse(&result), created, nil
+}
+
+type responseVectorIndex struct {
+	Name               string    `json:"name,omitempty"`
+	Type               IndexType `json:"type"`
+	IndexSharedOptions `json:",inline"`
+	Params             *VectorParams `json:"params,omitempty"`
+}
+
+func newVectorIndexResponse(res *responseVectorIndex) IndexResponse {
+	if res == nil {
+		return IndexResponse{}
+	}
+
+	return IndexResponse{
+		Name:               res.Name,
+		Type:               res.Type,
+		IndexSharedOptions: res.IndexSharedOptions,
+		VectorIndex:        res.Params,
+	}
+}