●Change8

v4.8.0rc1

📦 datadog-sdkView on GitHub →
✨ 6 features🐛 7 fixes⚡ 1 deprecations🔧 2 symbols

Summary

This release introduces new tracing support for Azure CosmosDB, an AI Guard integration for LiteLLM, and OTLP export capabilities. It also removes the deprecated RAGAS integration for LLM Observability.

Migration Steps

  1. If using RAGAS integration for LLM Observability evaluations, switch to manually submitting evaluation results via the external evaluation mechanism.

✨ New Features

  • ASM adds a LiteLLM proxy guardrail integration for Datadog AI Guard using `ddtrace.appsec.ai_guard.integrations.litellm.DatadogAIGuardGuardrail`, requiring `litellm>=1.46.1`.
  • Tracing support added for Azure CosmosDB, tracing CRUD operations on databases, containers, and items.
  • CI Visibility adds automatic log correlation and submission for test logs to appear alongside test runs. Enable via `DD_AGENTLESS_LOG_SUBMISSION_ENABLED=true` or `DD_LOGS_INJECTION=true`.
  • Tracing now supports exporting traces in OTLP HTTP/JSON format via libdatadog by setting `OTEL_TRACES_EXPORTER=otlp`.
  • LLM Observability introduces a `decorator` tag to spans traced by a function decorator.
  • LLM Observability experiments now accept a `pydantic_evals` `ReportEvaluator` as a summary evaluator if its `evaluate` return annotation is exactly `ScalarResult`, recording the scalar `value`.

🐛 Bug Fixes

  • Fixes lock profiling samples not appearing in the Thread Timeline view for events collected on macOS.
  • Fixes a potential internal thread leak in fork-heavy applications.
  • Resolves an issue where a `ModuleNotFoundError` could be raised at startup in Python environments lacking the `_ctypes` extension module.
  • Fixes a crash that could occur post-fork in fork-heavy applications.
  • Fixes incorrect span hierarchy in LLMObs traces when using ddtrace SDK alongside OTel instrumentation (e.g., Strands Agents), ensuring OTel gen_ai spans are nested under their SDK parent spans.
  • Fixes incorrect `model_name` and `model_provider` reporting on AWS Bedrock LLM spans; they now correctly match backend pricing data.
  • Fixes an issue where deferred tools in Anthropic and OpenAI integrations caused LLMObs span payloads to include full tool descriptions/schemas; deferred tool definitions now strip description and schema from span metadata, preserving only the tool name.

Affected Symbols

ddtrace.appsec.ai_guard.integrations.litellm.DatadogAIGuardGuardrailddtrace.llmobs.LLMObs

⚡ Deprecations

  • Support for the RAGAS integration in LLM Observability is removed. Users should manually submit RAGAS evaluation results using the external evaluation documentation as an alternative.