Datadog

Datadog is tightening tracing instrumentation control and hardening stability across Python and Node runtimes.

APM semantic convention lookup changed in Agent 7.78.0 — Agent now applies per-concept precedence ordering when resolving OpenTelemetry attributes with multiple equivalent keys (e.g., http.status_code vs http.response.status_code). Previously, any key found in span attributes would win regardless of resource attributes; now the registry defines the lookup order per concept.

dd-trace-py 4.7.0 shipped profiling performance gains via Cython — Lock profiler hot path compiled to C reduces overhead ~49% at default 1% capture rate, ~15-19% at 100% capture. No configuration needed. Also bumped openfeature-sdk minimum to 0.8.0 for hook metrics tracking, and flag evaluations now return Reason.ERROR with FLAG_NOT_FOUND instead of Reason.DEFAULT when a flag doesn't exist.

RAGAS integration removed from LLM Observability in v4.8.0 RC track — Users should manually submit evaluation results via external evaluation APIs. The pin parameter in database-related classes (TracedConnection, TracedCursor, TracedAsyncConnection) is deprecated ahead of v5.0.0; use environment variables or integration config instead.

Fork-safety fixes across dd-trace-py patch releases — Multiple versions (4.5.7+, 4.6.5+, 4.7.1) addressed race conditions in periodic background threads during fork, preventing crashes and hangs in Gunicorn, uWSGI, and on Python shutdown. CI Visibility now handles 429 rate-limit responses correctly, respecting X-RateLimit-Reset headers.

dd-trace-js added process tags and test optimization caching — v5.93.0 enables process tags by default; v5.95.0 adds filesystem caching for test optimization API requests and bumps LLM Observability payload limit from 1MB to 5MB to match Python and docs.

Browser SDK v6.32.0 expands Vue/Next.js support — Added LCP subparts collection by default, scaffolded new @datadog/browser-rum-vue package, and introduced addVueError and addNextjsError helpers. Error handling stack is now updatable via error.handling_stack.

Single Step Instrumentation rules now control which services are traced — New APM feature lets teams selectively reduce unnecessary trace data collection across hosts.