The Future of Observability with OpenTelemetry

Chapter 7. Suggested Setups and Telemetry Pipelines

Now that we understand the individual building blocks that make up OpenTelemetry, how should we combine them into a robust production pipeline?

The answer depends upon where you are starting from. OpenTelemetry is modular and designed to work at a variety of different scales. You need to use only the pieces that feel relevant. That said, we’ve created a suggested roadmap for you to follow.

Installing the OpenTelemetry Client

It is possible to use the OpenTelemetry client on its own without deploying a Collector. This basic setup is usually a sufficient starting point for greenfield deployments, both for testing and for initial production. The OpenTelemetry SDK can be configured to transmit telemetry directly to most observability services.

Picking an Exporter

By default, OpenTelemetry exports data using OTLP. The SDK ships with exporters for several common formats: Zipkin, Prometheus, StatsD, etc. If the observability backend you are using does not support OTLP natively, one of these other formats will most likely be supported. Install the correct exporter and send data directly to your backend system.

Installing Library Instrumentation

In addition to the SDK, OpenTelemetry instrumentation must be installed for all HTTP clients, web frameworks, databases, and message queues contained within the application. If one of these libraries is missing instrumentation, context propagation will break, leading to incomplete ...