OpenFeature as a Control Plane for OpenTelemetry

Talk to us

Questions we didn’t get to, want to try this in your own setup, or have opinions on the open questions? Reach out:

• Open an issue on the poc repo
• Find us on the CNCF Slack in #openfeature and #opentelemetry
• Through our websites: Simon and Juliano

About the talk

Observability is usually configured up front: too much telemetry while everything is healthy, too little once an incident starts. This talk uses OpenFeature flags to control OpenTelemetry instrumentation at runtime, so a service moves from lightweight tracing to detailed diagnostics without a redeploy.

The demo turns tracing off with one flag, drops INTERNAL spans while keeping the I/O edges, and targets by OpenTelemetry semantic conventions used as evaluation context, down to sampling that follows the error rate. The question we leave you with: should semantic conventions become the vendor-neutral control surface for this?

Speakers

The code

The project behind the talk is called OTelFeature:

• poc — the demo from the talk
• otelfeature-instrument — zero-code OpenTelemetry launcher with flagd-controlled span suppression
• otelfeature-java-extension — the Java agent counterpart

Found it useful? A star on the repos tells us to keep going.

Take it further

Learn more

• OpenFeature
• flagd
• OpenTelemetry
• Flagsmith
• Dash0