v4.11.1
📦 datadog-sdkView on GitHub →
🐛 5 fixes🔧 7 symbols
Summary
This release addresses several bugs across bootstrap, Code Security (IAST), and LLM Observability. Key fixes include improved PyYAML compatibility, resolution of IAST taint tracking issues, and enhanced LLM observability for metric keys and OpenAI token capture.
🐛 Bug Fixes
- bootstrap: Resolved an issue where `yaml`/`_yaml` were not kept loaded during module cleanup, which affected PyYAML consumers like Airflow.
- Code Security (IAST): Fixed a taint tracking bug that could cause the Python process to abort when old-style `%` string formatting encountered tainted text with literal IAST evidence marker delimiters.
- Code Security (IAST): Corrected a weak-hash false positive that occurred in ddtrace's own code when both IAST and the live debugger (Symbol Database) were enabled.
- LLM Observability: Fixed an issue where spans with metric keys containing dots (e.g., `anomaly.query_count`) were not ingested. Dots in metric keys are now replaced with underscores.
- LLM Observability: Resolved a problem where cache write input tokens were not captured by the OpenAI integration for newer models that include cache write token charges.