Skip to content

Azure AI Search 2025-03-01-Preview #32748

Open
@alzimmermsft

Description

@alzimmermsft

This issue outlines the work to be done for the Search 2025-03-01-Preview release.

These are the links to generate from:

https://raw.githubusercontent.com/Azure/azure-rest-api-specs/8c53aa7023e66a9ec24ede6a9fad0ed730d62515/specification/search/data-plane/Azure.Search/preview/2025-03-01-preview/searchindex.json

https://raw.githubusercontent.com/Azure/azure-rest-api-specs/8c53aa7023e66a9ec24ede6a9fad0ed730d62515/specification/search/data-plane/Azure.Search/preview/2025-03-01-preview/searchservice.json

Feature: Expose Indexes Statistics Summary

API changes

New API for listing statistics for all indexes in an Azure Search service.

Test plan

  • List index statistics when the Azure Search service has no indexes.
  • List index statistics when the Azure Search service has one index.
  • List index statistics when the Azure Search servive has many indexes.

The only validation to be done in all these tests is checking the number of index statistics returned and that expected index names are returned.

Notes

This is a paged API but we don't provision Azure Search services large enough to reach the page limitation. This will be a gap in testing.

Generative AI Chat Completion Skill

API changes

Adds a new skill GenAISkill and related configuration models GenAISkillCommonModelParameters, GenAISkillExtraParametersBehavior, and GenAISkillResponseFormat.

Test plan

Simple validation that the most basic GenAISkill can be created as part of a skillset pipeline and retrieved properly.

Notes

Only creation and retrieval need to be tested. We do not test skillset pipeline behaviors, so no additional testing or work is needed.

BQ Rescoring by Full Query

API changes

NONE

Runtime changes

In previous service versions configuring enableRescoring: true when rescoreStorageMethod: discardOriginals was configured in VectorSearchCompressions.rescoringOptions
was invalid. This has been changed to be allowed.

Test plan

Create a Search Index where the VectorSearchCompressions.rescoringOptions has enableRescoring: true and rescoreStorageMethod: discardOriginals. Ensure that works.

Add a few documents with vector field data and run a search and ensure that works without error (do not need to test any scoring information).

Text Chunk Reference

API changes

NONE

Notes

This is a behavior change in the outputs of the Text Split skill when running in the skillset pipeline to include additional data. We do not test skillset
pipeline behaviors, no work needed for this change.

Image Reference

API changes

NONE

Notes

This is a behavior change in the outputs of the Search Indexer when processing images to include additional data. We do not test indexer behaviors,
no work needed for this change.

Allow Identity field on AIServicesByIdentity to be null

API changes

NONE

Notes

This is a change to allow AIServicesAccountIdentity.identity to be nullable. This is used by the skillset pipeline, we do not test skillset pipeline behaviors
so no work is needed for this change.

Just ensure that the property is properly nullable.

Sum Aggregations

API changes

NONE

Runtime changes

The facets string now allows for sum aggregations to be performed on facet results. For example, a price field could have the values summed and returned as a metric.

Examples facet strings:

Top-level field

"facets": ["price, metric: sum"]

In the resulting facet results you'd have

"@search.facets": {
  "price": [
     {
       "sum": 50.00
     }
   ]
}

Nested field

"facets": ["parent > price, metric: sum"]

In the resulting facet results you'd have

"@search.facets": {
  "parent": [
    {
       // Other facet data
       "@search.facets": [
         {
           "sum": 50.00
         }
      ]
    }
}

Test plan

Testing here is optional as we don't provide conveniences for facets. If you wish to test just do two simple aggregations with a top-level and nested field.

Semantic Model Flighting

API changes

New boolean configuration on SemanticConfiguration flightingOptIn that configures which semantic and query rewrite models during flighting / upgrades.

Test plan

Ensure this value can be set properly.

Notes

Given model upgrades are opaque, we don't have a good way to test actual behaviors and should treat this similar to how we treat skillsets where we only check the CRUD operation.

Metadata

Metadata

Assignees

Labels

Type

No type

Projects

  • Status

    Untriaged

Milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions