fix(opensearch): Use the same method for getting title that the title embedding logic uses; small cleanup for content embedding

[acaprau]

Description

_index_vespa_chunk uses get_title_for_document_index() to populate the title field in a document index entry, which also happens to be same data used to create a title embedding. The OpenSearch flow needs to use this too.

This PR also encapsulates the enrichment logic for creating a content embedding in a similar way that content text enrichment logic is encapsulated in backend/onyx/document_index/chunk_content_enrichment.py.

How Has This Been Tested?

@evan-onyx first wrote this title fix in https://github.com/onyx-dot-app/onyx/pull/7560/changes#diff-d09eebe79e128bf021e791c21a4ff9cf98d85d79e4da36b237885de09bbd5618 and it works.

Additional Options

• [Optional] Override Linear Check

---

Summary by cubic

Use get_title_for_document_index in OpenSearch so the stored title matches the title used for embeddings and corrects fallback behavior. Also add a helper for content-embedding enrichment and rename the text-enrichment helper for clarity.

• Bug Fixes

  • OpenSearch sets title via get_title_for_document_index(), matching embedder logic and falling back to semantic_identifier when title is None.

• Refactors

  • Added generate_enriched_content_for_chunk_embedding() and used it in the embedder.
  • Renamed generate_enriched_content_for_chunk → generate_enriched_content_for_chunk_text and updated OpenSearch/Vespa call sites.

^{Written for commit f786e38. Summary will update on new commits.}

[greptile-apps]

Greptile Summary

Fixed OpenSearch to use get_title_for_document_index() for the title field, ensuring consistency with Vespa and the title embedding logic. This method properly falls back to semantic_identifier when title is None (but not empty string), preventing title/embedding mismatches.

• Refactored content enrichment logic by extracting generate_enriched_content_for_chunk_embedding() function that encapsulates the embedding-specific enrichment (uses metadata_suffix_semantic)
• Renamed existing function to generate_enriched_content_for_chunk_text() to clarify it's for text/BM25 indexing (uses metadata_suffix_keyword)
• Updated all call sites in OpenSearch, Vespa, and embedder to use the appropriate enrichment functions

Confidence Score: 5/5

• This PR is safe to merge with no issues found
• The changes are straightforward refactoring and bug fixes that improve code consistency. The title fix addresses a real inconsistency between OpenSearch and Vespa/embedder. The content enrichment refactoring follows existing patterns and properly separates concerns between embedding and text indexing. All changes are well-contained with clear intent.
• No files require special attention

Important Files Changed

Filename	Overview
backend/onyx/document_index/chunk_content_enrichment.py	Extracted content embedding enrichment logic into a dedicated function, matching the existing pattern for text enrichment
backend/onyx/document_index/opensearch/opensearch_document_index.py	Fixed title field to use get_title_for_document_index() matching Vespa and embedder logic; renamed function to use new text enrichment function

Sequence Diagram

sequenceDiagram
    participant Chunk as DocMetadataAwareIndexChunk
    participant Embedder as DefaultIndexingEmbedder
    participant EnrichFunc as generate_enriched_content_for_chunk_embedding
    participant OpenSearch as opensearch_document_index
    participant TextEnrichFunc as generate_enriched_content_for_chunk_text
    participant Doc as Document

    Note over Embedder,EnrichFunc: Content Embedding Flow
    Embedder->>EnrichFunc: Call with DocAwareChunk
    EnrichFunc-->>Embedder: Return enriched content (with metadata_suffix_semantic)
    Embedder->>Doc: get_title_for_document_index()
    Doc-->>Embedder: Return title (or semantic_identifier if title is None)
    
    Note over OpenSearch,TextEnrichFunc: OpenSearch Indexing Flow
    OpenSearch->>Doc: get_title_for_document_index()
    Note right of OpenSearch: Fixed: Now uses same<br/>method as embedder
    Doc-->>OpenSearch: Return consistent title
    OpenSearch->>TextEnrichFunc: Call with DocMetadataAwareIndexChunk
    TextEnrichFunc-->>OpenSearch: Return enriched content (with metadata_suffix_keyword)

Loading

[cubic-dev-ai]

No issues found across 4 files