fix: #10063 prevent oversized tracing payloads for ByteStream and Ima…

[jovicdev97]

Related Issues

• fixes [Tracing] ByteStream objects cause oversized payloads in tracing backends #10063

Proposed Changes:

This PR optimizes how ByteStream and ImageContent objects are serialized during tracing to prevent oversized payloads, which fixes #10063

Previously, tracing these objects would attempt to serialize their full binary or base64 content, leading to massive trace sizes. The updated logic in _serializable_value:

ByteStream: Now excludes the raw data payload. Instead, it captures metadata (mime type, size in bytes) and generates a short text preview (first 100 bytes) if the content is text-based.

ImageContent: Now excludes the full base64_image. It captures metadata and provides a truncated preview of the base64 string.

General: Added a helper _get_text_preview to safely decode and truncate text previews.

How did you test it?

I added comprehensive unit tests in test/tracing/test_utils.py covering the following scenarios:

ByteStream Binary: Verifying that large binary data is excluded from serialization while retaining metadata (size_bytes, mime_type).

ByteStream Text: Verifying that text-based streams generate a correct, truncated preview string.

ImageContent: Verifying that large base64 strings are truncated to prevent payload bloat.

Nested Structures: Verifying that these objects are correctly handled when nested inside lists or dictionaries.

Type Coercion: Verifying coerce_tag_value handles these objects correctly via the updated serialization logic.

Notes for the reviewer

Checklist

• I have read the contributors guidelines and the code of conduct.
• I have updated the related issue with new insights and changes.
• I have added unit tests and updated the docstrings.
• I've used one of the conventional commit types for my PR title: fix:, feat:, build:, chore:, ci:, docs:, style:, refactor:, perf:, test: and added ! in case the PR includes breaking changes.
• I have documented my code.
• I have added a release note file, following the contributors guidelines.
• I have run pre-commit hooks and fixed any issue.

[vercel]

@jovicdev97 is attempting to deploy a commit to the deepset Team on Vercel.

A member of the Team first needs to authorize it.

[LastRemote]

Thank you for the quick response!

Have you tested the scenario of a list of ChatMessage with oversized ImageContent? I think that is probably the most important and the most problematic one in terms of tracing. ChatMessage.to_dict() would call ImageContent.to_dict() directly, and that might escape the special handling logic.

[sjrl]

Hey @jovicdev97 thanks for the contribution! However, since this change affects a core part of our tracing functionality, we’d prefer to handle it internally.

If you’d like to explore other ways to contribute, feel free to check out our Contributions Wanted board: Haystack - Contributions wanted