All Products
Search
Document Center

OpenSearch:Common configurations of vector indexes

Last Updated:Jul 11, 2025

When you create a table, you can configure advanced configurations for vector indexes in the Index Schema step. This topic describes the parameters for the advanced configurations of vector indexes.

When you create a table, configure the index schema in the Index Schema step.cn向量索引结构配置.png

The following figure shows the parameters for advanced configurations.

image.png

The following table describes the parameters.

Parameter

Valid value

Description

Vector Dimension

N/A

The number of features or attributes of a vector. The Vector Dimension parameter specifies the complexity of the information and features that the vector can represent. You must configure the Vector Dimension parameter based on the vector generated by your vector model.

Distance Type

  • SquareEuclidean

  • InnerProduct

  • Cosine

The smaller the SquareEuclidean score, the more relevence.

The larger the InnerProduct score, the more relevence.

The Cosine score ranges from [-1, 1]. A value of -1 indicates that the two vectors are in opposite directions, with the lowest similarity. A value of 1 indicates that the two vectors are in the same direction, with the highest similarity.

Vector Index Algorithm

  • Qc

  • HNSW

  • Linear

  • QGraph

  • CAGRA

  • DiskANN

The vector indexing algorithm. For more information, see Introduction to vectors.

Note

The DiskANN algorithm is supported only when the data node specification family is SSD.

Real-time Indexing

  • true

  • false

Specifies whether to enable the real-time indexing feature. If you set this parameter to true, the real-time indexing feature is enabled. The OpenSearch Vector Search Edition instance builds indexes for the real-time data that you push by calling API operations. Then, you can query the data in real time.

Real-time Indexing Parameters

{"proxima.oswg.streamer.segment_size":2048}

The parameters for real-time indexing. We recommend that you use the default value.

Index Retrieval Parameters

N/A

The parameters for real-time retrieval. You must configure this parameter based on the vector indexing algorithm. For more information, see the following topics:

Vector Separator

Customizable

The delimiter that is used to separate dimensions during vector retrieval. For example, a comma (,) is used as the delimiter in vector:'1.05066,0.15610,0.156145...'.

Threshold for Linear Building

Default value: 5000

The threshold value for operations that do not create indexes in order. A value of 5000 specifies that indexes are created in order if the number of documents is less than 5,000.

Ignore Invalid Vector Data

  • true

  • false

Specifies whether to ignore invalid vector data. If you set this parameter to true, the system creates indexes for full or batch incremental data as expected when the vector dimension is invalid and the vector data is empty.