OpenTelemetry

We instrument our applications with OpenTelemetry (OTel), the vendor-neutral CNCF standard for traces, metrics and logs. It provides the APIs, SDKs and the OTLP wire protocol, and practically every observability backend accepts it.

Why OpenTelemetry?

• Instrument once: Code is instrumented against OTel, not against a monitoring vendor. Where the data goes is decided in configuration, so switching backends does not mean touching application code.

• Backend choice: OTLP is accepted by Grafana and its stack as well as by managed services such as Scaleway Cockpit, which keeps the observability stack as portable as the rest of our infrastructure.

• Traces across services: Distributed tracing is what lets you follow a request through several services, queues and model calls. For Event-driven Architecture and AI pipelines in particular, it has to be in place from the start.

• Broad support: Auto-instrumentation is available for the languages and frameworks we use, and a growing number of tools (including LLM gateways and agent frameworks) emit OTel natively.