milvus-logo
LFAI
Home
  • User Guide
  • Benchmarks

Build an Index on Vectors

This guide describes how to build an index on vectors in Milvus.

Vector indexes are an organizational unit of metadata used to accelerate vector similarity search. You need to create an index before you can perform ANN searches against your Milvus.

See Vector Index for more information about the mechanism and varieties of vector indexes.

By default, Milvus does not index a segment with less than 1,024 rows. To change this parameter, configure rootCoord.minSegmentSizeToEnableIndex in milvus.yaml.

The following example builds a 1024-cluster IVF_FLAT index with Euclidean distance (L2) as the similarity metric. You can choose the index and metrics that suit your scenario. See Similarity Metrics for more information.

When interacting with Milvus using Python code, you have the flexibility to choose between PyMilvus and MilvusClient (new). For more information, refer to Python SDK.

Prepare index parameter

Prepare the index parameters as follows:

Python Java GO Node.js 
index_params = {
  "metric_type":"L2",
  "index_type":"IVF_FLAT",
  "params":{"nlist":1024}
}

const index_params = {
  metric_type: "L2",
  index_type: "IVF_FLAT",
  params: JSON.stringify({ nlist: 1024 }),
};

idx, err := entity.NewIndexIvfFlat(   // NewIndex func
    entity.L2,                        // metricType
    1024,                             // ConstructParams
)
if err != nil {
  log.Fatal("fail to create ivf flat index parameter:", err.Error())
}

final IndexType INDEX_TYPE = IndexType.IVF_FLAT;   // IndexType
final String INDEX_PARAM = "{\"nlist\":1024}";     // ExtraParam

create index

Collection name (book): book

The name of the field to create an index for (book_intro): book_intro

Index type (FLAT, IVF_FLAT, IVF_SQ8, IVF_PQ, RNSG, HNSW): IVF_FLAT

Index metric type (L2, IP, HAMMING): L2

Index params nlist: 1024

Timeout []:

curl -X 'POST' \
  'http://localhost:9091/api/v1/index' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json' \
  -d '{
    "collection_name": "book",
    "field_name": "book_intro",
    "extra_params":[
      {"key": "metric_type", "value": "L2"},
      {"key": "index_type", "value": "IVF_FLAT"},
      {"key": "params", "value": "{\"nlist\":1024}"}
    ]
  }'
Parameter Description Options
metric_type Type of metrics used to measure the similarity of vectors. For floating point vectors:
  • L2 (Euclidean distance)
  • IP (Inner product)
  • COSINE (Cosine similarity)
For binary vectors:
  • JACCARD (Jaccard distance)
  • HAMMING (Hamming distance)
index_type Type of index used to accelerate the vector search. For floating point vectors:
  • FLAT (FLAT)
  • IVF_FLAT (IVF_FLAT)
  • IVF_SQ8 (IVF_SQ8)
  • IVF_PQ (IVF_PQ)
  • GPU_IVF_FLAT* (GPU_IVF_FLAT)
  • GPU_IVF_PQ*> (GPU_IVF_PQ)
  • HNSW (HNSW)
  • DISKANN* (DISK_ANN)
For binary vectors:
  • BIN_FLAT (BIN_FLAT)
  • BIN_IVF_FLAT (BIN_IVF_FLAT)
params Building parameter(s) specific to the index. See In-memory Index and On-disk Index for more information.
  • * DISKANN has certain prerequisites to meet. For details, see On-disk Index.
  • * GPU_IVF_FLAT and GPU_IVF_PQ are available only when you install Milvus with the GPU feature enabled. For details, see
Parameter Description Option
metric_type Type of metrics used to measure the similarity of vectors. For floating point vectors:
  • L2 (Euclidean distance)
  • IP (Inner product)
  • COSINE (Cosine similarity)
For binary vectors:
  • JACCARD (Jaccard distance)
  • HAMMING (Hamming distance)
index_type Type of index used to accelerate the vector search. For floating point vectors:
  • FLAT (FLAT)
  • IVF_FLAT (IVF_FLAT)
  • IVF_SQ8 (IVF_SQ8)
  • IVF_PQ (IVF_PQ)
  • GPU_IVF_FLAT* (GPU_IVF_FLAT)
  • GPU_IVF_PQ*> (GPU_IVF_PQ)
  • HNSW (HNSW)
  • DISKANN* (DISK_ANN)
For binary vectors:
  • BIN_FLAT (BIN_FLAT)
  • BIN_IVF_FLAT (BIN_IVF_FLAT)
params Building parameter(s) specific to the index. See In-memory Index and On-disk Index for more information.
  • * DISKANN has certain prerequisites to meet. For details, see On-disk Index.
  • * GPU_IVF_FLAT and GPU_IVF_PQ are available only when you install Milvus with the GPU feature enabled. For details, see
Parameter Description Options
NewIndex func Function to create entity. Index according to different index types. For floating point vectors:
  • NewIndexFlat (FLAT)
  • NewIndexIvfFlat (IVF_FLAT)
  • NewIndexIvfSQ8 (IVF_SQ8)
  • NewIndexIvfPQ (IVF_PQ)
  • NewIndexGPUIvfFlat (GPU_IVF_FLAT)
  • NewIndexGPUIvfPQ (GPU_IVF_PQ)
  • NewIndexHNSW (HNSW)
  • NewIndexDISKANN* (DISK_ANN)
For binary vectors:
  • NewIndexBinFlat (BIN_FLAT)
  • NewIndexBinIvfFlat (BIN_IVF_FLAT)
metricType Type of metrics used to measure the similarity of vectors. For floating point vectors:
  • L2 (Euclidean distance)
  • IP (Inner product)
  • COSINE (Cosine similarity)
For binary vectors:
  • JACCARD (Jaccard distance)
  • HAMMING (Hamming distance)
ConstructParams Building parameter(s) specific to the index. See In-memory Index and On-disk Index for more information.
  • * DISKANN has certain prerequisites to meet. For details, see On-disk Index.
  • * GPU_IVF_FLAT and GPU_IVF_PQ are available only when you install Milvus with the GPU feature enabled. For details, see
Parameter Description Options
IndexType Type of index used to accelerate the vector search. For floating point vectors:
  • FLAT (FLAT)
  • IVF_FLAT (IVF_FLAT)
  • IVF_SQ8 (IVF_SQ8)
  • IVF_PQ (IVF_PQ)
  • GPU_IVF_FLAT* (GPU_IVF_FLAT)
  • GPU_IVF_PQ*> (GPU_IVF_PQ)
  • HNSW (HNSW)
  • DISKANN* (DISK_ANN)
For binary vectors:
  • BIN_FLAT (BIN_FLAT)
  • BIN_IVF_FLAT (BIN_IVF_FLAT)
ExtraParam Building parameter(s) specific to the index. See In-memory Index and On-disk Index for more information.
  • * DISKANN has certain prerequisites to meet. For details, see On-disk Index.
  • * GPU_IVF_FLAT and GPU_IVF_PQ are available only when you install Milvus with the GPU feature enabled. For details, see

Build index

Build the index by specifying the vector field name and index parameters.

Python Java GO Node.js 
from pymilvus import Collection, utility
# Get an existing collection.
collection = Collection("book")      
collection.create_index(
  field_name="book_intro", 
  index_params=index_params
)

utility.index_building_progress("book")
# Output: {'total_rows': 0, 'indexed_rows': 0}

await milvusClient.createIndex({
  collection_name: "book",
  field_name: "book_intro",
  extra_params: index_params,
});

err := milvusClient.CreateIndex(
  context.Background(),        // ctx
  "book",                      // CollectionName
  "book_intro",                // fieldName
  idx,                         // entity.Index
  false,                       // async
)
if err != nil {
  log.Fatal("fail to create index:", err.Error())
}

milvusClient.createIndex(
  CreateIndexParam.newBuilder()
    .withCollectionName("book")
    .withFieldName("book_intro")
    .withIndexType(INDEX_TYPE)
    .withMetricType(MetricType.L2)
    .withExtraParam(INDEX_PARAM)
    .withSyncMode(Boolean.FALSE)
    .build()
);

# Follow the previous step.

# Follow the previous step.
Parameter Description
field_name Name of the vector field to build index on.
index_params Parameters of the index to build.
Parameter Description
collection_name Name of the collection to build index in.
field_name Name of the vector field to build index on.
extra_params Parameters of the index to build.
Parameter Description
ctx Context to control API invocation process.
CollectionName Name of the collection to build index on.
fieldName Name of the vector field to build index on.
entity.Index Parameters of the index to build.
async Switch to control sync/async behavior. The deadline of context is not applied in sync building process.

What’s next

Table of contents

Try Managed Milvus for Free

Zilliz Cloud is hassle-free, powered by Milvus and 10x faster.

Get Started
Feedback

Was this page helpful?