Index management

Index management in OpenSearch covers how index schemas define field types for search and filtering, how composite indexes span multiple fields, how analyzers process text and vector data, and how attribute fields support filter, sort, aggregate, and distinct operations.

An index schema defines how OpenSearch stores and retrieves your data. It consists of two types of fields:

• Index fields — analyzed during indexing and searched at query time. Use these for full-text and keyword searches.

• Attribute fields — used in filter, sort, aggregation, and distinct clauses.

Field type compatibility

The following table shows which field types support indexing, attribute configuration, or both.

Composite index

A composite index spans multiple TEXT or SHORT_TEXT fields. Unlike querying multiple single-field indexes with OR, a composite index treats the combined content of all its fields as a single searchable unit, so a query term that spans across fields can be matched.

Consider an application with these indexes:

• title_index — index on the title field

• body_index — index on the body field

• union_index — composite index on both title and body

Given this document:

id:123456,title:开放,body:搜索

# Querying two separate indexes with OR does not match the document.
query=title_index:'开放搜索' OR body_index:'开放搜索'

# Querying the composite index matches the document.
query=union_index:'开放搜索'

The term 开放搜索 cannot be found in either field alone, but the composite index finds it by searching across both fields together.

Text analyzers

Text analyzers: Text analyzers segment text for retrieval across the following domains: Chinese, English, general industry, IT, e-commerce, and education. They also support:

• Exact search by numeric or fixed-length value

• Fuzzy search

• Range search by location, time range, or numerical range

Vector analyzers

Vector analyzers retrieve documents based on multi-dimensional vectors. Two configurations are available:

• General industry: retrieves documents based on multi-dimensional vectors

• Education: retrieves general documents based on 256-dimensional vectors

Exact match without tokenization

To make a field match exactly without tokenization, submit an offline change to set the field type to LITERAL and set its analyzer to Keyword Analyzer. With Keyword Analyzer, the entire field value is treated as a single token, so queries against the field require an exact match instead of a tokenized match.

Because changing a field type is not available in the console, this configuration must go through the offline change process. The field cannot take effect immediately after saving.

You can also avoid tokenization in other scenarios:

• To avoid tokenization for only part of a field's content, use a custom analyzer together with intervention entries and disable secondary tokenization. See Custom analyzers.

• To avoid tokenization for an entire query term, wrap the term in double quotation marks, for example, query=default:"开放搜索". See Index recall - query clauses.

{
  "searchFields": {
    "vector": {
      "analyzer": "qwen3_vl_embedding",
      "fields": [
        "title",
        "img_url"
      ]
    }
  }
}

{
  "cluster": {
    "vectorIndexConfigs": [
      {
        "indexName": "vector_index1",
        "namespace": "",
        "indexType": "Qc",
        "distanceType": "SquaredEuclidean",
        "sourceFields": [
          { "fieldName": "title", "modality": "text" },
          { "fieldName": "img_url", "modality": "image" }
        ]
      }
    ]
  }
}

Attribute fields support the following clauses: