All Products
Search

This Product

    OpenSearch:Engine versions

    Document Center

    OpenSearch:Engine versions

    Last Updated:Jan 02, 2025

    OpenSearch Vector Search Edition is upgraded to a new engine version to improve user experience and simplify the use of services. You can store vector data to tables in databases instead of clusters that involve complex operations. This way, you can build the vector search engine based on the databases that store vector data. This topic describes the information about service upgrades.

    To ensure that the existing users can use the existing resources, both the old and new engine versions of OpenSearch Vector Search Edition are available. You can view the engine version of an instance on the instance details page.

    image.png

    The value of the Current Engine Version parameter indicates the engine version:

    • vector_service_1.0.1: the new engine version of OpenSearch Vector Search Edition.

    • ha3_3.10.0: the old engine version of OpenSearch Vector Search Edition.

    Important

    ha3_3.10.0 is an old engine version, and OpenSearch no longer provides feature upgrades for this engine version.

    Comparison of core features between the old and new engine versions

    Feature

    Old engine version (ha3_3.10.0)

    New engine version (vector_service_1.x.x)

    Supported data source

    • MaxCompute + API

    • OSS + API

    • API

    • MaxCompute + API

    • OSS + API

    • DLF

    • API

    Supported query test on the Query Test page in the OpenSearch Vector Search Edition console

    • Vector-based query

    • SQL development

    • Primary key-based query

    • Vector-based query

    • Hybrid query by vector and text

    SQL query statement

    Supported

    Not supported

    Vectorization model

    • Text vectorization

    • Image vectorization

    • Dense vectorization

    • Sparse vectorization

    • Image vectorization

    • Multi-subject identification

    • Optical character recognition (OCR)

    Vector-based query

    • Query by using a single vector

    • Query by using multiple vectors

    • Namespace-based query

    • Prediction-based query

    • Parameters that are used to return top N results or specify the threshold value based on which the results are filtered

    • Query by using a single vector

    • Namespace-based query

    • Query by using multiple namespaces

    • Query by using multiple vectors

    • Multi-vector-based query for a single document

    • Prediction-based query

    • Multi-query

    • Hybrid query

    • Parameters that are used to return top N results or specify the threshold value based on which the results are filtered

    Query syntax

    SDK

    Features supported by the vector_service engine version

    V1.4.0

    New features:

    • Text indexes are supported.

    • The search API is supported to implement hybrid queries of vector and text indexes.

    • The aggregate API is supported to implement statistical queries.

    • The Deflate algorithm can be used to compress query bodies.

    • The fetch API can be used to query fields of the TEXT type.

    V1.3.0

    New features:

    • Data Lake Formation (DLF) data sources are supported.

    • Models on the search development platform are supported.

    • A time to live (TTL) period can be configured for documents. After the TTL period ends, documents are automatically deleted.

    • The performance of the CONTAINS function is optimized.

    • The conditional operator ?: is supported in the SORT and FILTER functions. b : c)

    V1.2.2

    New features:

    • Aliases can be created for tables.

    • The logic for converting filter condition-based queries into inverted index-based queries is optimized.

    • Timeout periods can be configured for queries.

    Fixed issues:

    • The following issue is fixed: Index switching may be stuck when continuous query traffic exists.

    • The following issue is fixed: The engine core may crash during entity recognition.

    V1.2.0

    New features:

    • The number of Searcher workers is decoupled from the number of table shards. One Searcher worker can load multiple table shards.

    • Table creation is accelerated.

    • More filter functions are supported.

    V1.1.3

    New features:

    • Sparse vector scores can be calculated during filtering optimization.

    • Vector score thresholds can be configured for filtering optimization.

    V1.1.2

    New features:

    • Multiple vector indexes can be created for one field. This feature can be used in scenarios where filter condition-based queries need to be converted into inverted index-based queries.

    • Sparse vectors can be updated in real time by using HNSWStreamer.

    • Tags related to table names are added in QRS metrics.

    V1.1.1

    New features:

    • Subject identification is supported.

    • You can specify filter conditions to fetch data.

    • The issue that a navi collect resource timeout occurs when a model is loaded for an extended period of time is fixed.

    V1.1.0

    New features:

    • Query by using sparse vectors is supported.

    • Query by using multiple vectors is supported.

    • Custom sort expressions are supported.

    • Inverted indexes can be used to filter data.

    V1.0.3

    New features:

    • The core issue that occurs when a multi-value field of the STRING type in a table is deserialized if the size of the field value exceeds 4 KB is fixed.

    • The issue that memory leaks occur because data is segmented but not deleted when ann uses qc streamer is fixed.

    V1.0.2

    New features:

    • The issue that the new full data cannot be loaded and the old full data cannot be unloaded when you load data is fixed.

    V1.0.1

    New features:

    • Turing functions can be used in the filter conditions.

    • The issue that the multi-value results of the STRING type are serialized is fixed.

    • The issue that only the last table on which model prediction must be performed is returned is fixed.

    V1.0.0

    New features:

    • Namespace-based query is supported.

    • Query by using multiple namespaces is supported.

    • Index information can be retrieved.

    • The built-in model for prediction is supported.

    Usage notes

    • You cannot upgrade the engine version of an instance from ha3_3.10.0 to vector_service. You must create a new instance.

    • If you want to use features that are supported by the vector_service engine version, contact us to upgrade your engine version.