●Change8

v4.10.3

📦 datadog-sdkView on GitHub →
🐛 4 fixes🔧 4 symbols

Summary

This release primarily focuses on stability and observability improvements, including increasing the default CI Visibility timeout, fixing APM span naming for pydantic integrations, and ensuring LLM Observability spans are not dropped due to deep JSON nesting.

Migration Steps

  1. To override the new default HTTP timeout for CI Visibility backend requests, set the `DD_CIVISIBILITY_BACKEND_API_TIMEOUT_MILLIS` environment variable (value in milliseconds).

🐛 Bug Fixes

  • CI Visibility: Increased the default HTTP timeout for backend requests from 15 seconds to 30 seconds, and ensured this timeout applies uniformly to all backend requests, including skippable test fetches.
  • pydantic_ai: Corrected APM span naming convention to use the generic category (e.g., `pydantic_ai.tool`) as the operation name and the specific name as the resource name, aligning with Datadog APM standards.
  • sqlalchemy: Resolved an issue causing duplicate SQLAlchemy event listeners when tracing the same engine multiple times.
  • LLM Observability: Implemented handling for deeply nested JSON fields in spans by stringifying fields that exceed Datadog's allowed depth, preventing spans from being dropped.

Affected Symbols

DD_CIVISIBILITY_BACKEND_API_TIMEOUT_MILLISpydantic_ai.toolpydantic_ai.agenttrace_engine()