●Change8

v4.7.0rc2

📦 datadog-sdkView on GitHub →
✨ 27 features🔧 31 symbols

Summary

This release introduces significant performance improvements to the profiler via Cython compilation and adds extensive new features across LLM Observability, tracing integrations (MLFlow, Azure Durable Functions, Google Cloud Pub/Sub), and general telemetry propagation via process tags.

Migration Steps

  1. If you wish to disable the new process tag propagation feature across profiler, runtime metrics, remote configuration, dynamic instrumentation, crash tracking, data streams monitoring, database monitoring, stats computation, and tracer payloads, set the environment variable `DD_EXPERIMENTAL_PROPAGATE_PROCESS_TAGS_ENABLED=false`.
  2. To preserve raw `mongodb.query` values, ensure `DD_APM_OBFUSCATION_MONGODB_ENABLED=false` is set on the Datadog Agent, in addition to using the new configuration option.
  3. If using AI Guard evaluation blocking, note that the parameter `block` now defaults to `True`.

✨ New Features

  • Profiling lock profiler's hot path is now compiled to C via Cython, resulting in significant performance improvements for lock operations.
  • Added a request header provider (auth plugin) for MLFlow that includes DD-API-KEY and DD-APPLICATION-KEY headers if environment variables are set.
  • AI Guard calls now block by default if blocking is enabled in the UI; this can be disabled with `block=False`.
  • AI Guard SDK response now includes Sensitive Data Scanner (SDS) results.
  • Introduced AI Guard support for Strands Agents (Plugin API requires `strands-agents>=1.29.0`).
  • Added tracing support for Azure Durable Functions, tracing durable activity and entity functions.
  • Process tags are now added to profiler payloads (can be deactivated with `DD_EXPERIMENTAL_PROPAGATE_PROCESS_TAGS_ENABLED=false`).
  • Process tags are now added to runtime metrics tags (can be deactivated with `DD_EXPERIMENTAL_PROPAGATE_PROCESS_TAGS_ENABLED=false`).
  • Process tags are now added to remote configuration payloads (can be deactivated with `DD_EXPERIMENTAL_PROPAGATE_PROCESS_TAGS_ENABLED=false`).
  • Process tags are now added to debugger payloads (dynamic instrumentation) (can be deactivated with `DD_EXPERIMENTAL_PROPAGATE_PROCESS_TAGS_ENABLED=false`).
  • Process tags are now added to crash tracking payloads (can be deactivated with `DD_EXPERIMENTAL_PROPAGATE_PROCESS_TAGS_ENABLED=false`).
  • Process tags are now added to Data Streams Monitoring payloads (can be deactivated with `DD_EXPERIMENTAL_PROPAGATE_PROCESS_TAGS_ENABLED=false`).
  • Process tags are now added to Database Monitoring SQL service hash propagation (can be deactivated with `DD_EXPERIMENTAL_PROPAGATE_PROCESS_TAGS_ENABLED=false`).
  • Process tags are now added to stats computation payloads (can be deactivated with `DD_EXPERIMENTAL_PROPAGATE_PROCESS_TAGS_ENABLED=false`).
  • LLM Observability experiment tasks can optionally receive dataset record metadata as a third `metadata` parameter.
  • Introduced `RemoteEvaluator` for referencing LLM-as-Judge evaluations configured in the Datadog UI by name during local experiments.
  • Added cache creation breakdown metrics (`ephemeral_5m_input_tokens` and `ephemeral_1h_input_tokens`) for the Anthropic integration when prompt caching is used.
  • Added support for reasoning and extended thinking content in Anthropic, LiteLLM, and OpenAI-compatible integrations.
  • LLMJudge now forwards extra `client_options` to the underlying provider client constructor.
  • Added `dataset.add_tags`, `dataset.remove_tags`, and `dataset.replace_tags` methods to operate on dataset record tags.
  • Changed experiment execution to run evaluators immediately after each record's task completes, posting spans and metrics incrementally.
  • Added support for Pydantic AI evaluations in LLM Observability Experiments.
  • Tracer now supports API endpoint discovery for Tornado applications.
  • Process tags are now added to trace payloads (can be deactivated with `DD_EXPERIMENTAL_PROPAGATE_PROCESS_TAGS_ENABLED=false`).
  • Added instrumentation support for `mlflow>=2.11.0`.
  • Added a new configuration option `DD_TRACE_MONGODB_OBFUSCATION` to control obfuscation of `mongodb.query` resource names.
  • Added tracing support for the `google-cloud-pubsub` library, instrumenting publishing and subscribing.

Affected Symbols

profiling (lock profiler)MLFlow (auth plugin)AI Guard (evaluate calls)AI Guard API clientStrands Agents Plugin APIazure_durable_functionsprofiling (payloads)runtime metrics (tags)remote configuration (payloads)dynamic instrumentation (payloads)crashtracking (payloads)data streams monitoring (payloads)database monitoring (SQL service hash propagation)stats computation (payloads)LLMObs.experiment task signatureRemoteEvaluatorAnthropic integration (caching metrics)Anthropic integration (thinking blocks)LiteLLM (reasoning content)OpenAI-compatible integrations (reasoning content)LLMJudgedataset.add_tagsdataset.remove_tagsdataset.replace_tagsLLMObs.experiment execution flowTornado applications (endpoint discovery)tracer (payloads)mlflow>=2.11.0pymongo (resource names)google_cloud_pubsub (PublisherClient.publish)google_cloud_pubsub (SubscriberClient.subscribe)