APM and OpenTelemetry

1. Instrumenting #

Three options.

Choosing #

pip install ddtrace
DD_SERVICE=checkout DD_ENV=prod DD_VERSION=2.4.1 DD_LOGS_INJECTION=true DD_AGENT_HOST=localhost \
  ddtrace-run gunicorn app:app

OpenTelemetry to Datadog #

# datadog.yaml : l'agent reçoit l'OTLP
otlp_config:
  receiver:
    protocols:
      grpc: { endpoint: 0.0.0.0:4317 }
      http: { endpoint: 0.0.0.0:4318 }

Applications instrumented with OpenTelemetry send to the agent (ports 4317/4318); the deployment.environment, service.name and service.version attributes are converted to env, service and version.

2. Using APM #

Services, traces, profiles.

Views #

Good to know: Trace retention relies on ingestion sampling and retention filters: not all traces are kept for 15 days, only those retained by the filters (errors, slow traces, sample).

3. Linking logs and traces #

One click from one signal to the other.

Correlation #

With DD_LOGS_INJECTION=true, the library adds dd.trace_id and dd.span_id to logs; Datadog shows the logs of a trace and the trace of a log.

2026-10-19 10:00:00 ERROR [dd.service=checkout dd.env=prod dd.version=2.4.1 dd.trace_id=5735239282836592 dd.span_id=1234567] Paiement refusé

Good practice: If logs are in JSON, the dd.* attributes are recognized directly; otherwise, add them to the log format and extract them in the pipeline.