ReadMeFeed / Community knowledge
Real questions. Useful conversations. Find the people who know your stack.
A two-tier OpenTelemetry Collector design: local agents export OTLP to a gateway that runs tail_sampling, preserving errors and slow traces while bounding cost and blast radius.
Head sampling drops traces before knowing if they matter. Tail sampling in the OpenTelemetry Collector waits for the whole trace -- here's how to configure it and what it costs.
Use the OpenTelemetry Collector's Tail Sampling Processor to keep errors and slow traces while probabilistically sampling the rest. This guide covers a working configuration, policy ordering, memory limits, and common placement mistakes.
Goal Reduce trace ingestion costs for a low-traffic service instrumented with OpenTelemetry. Constraints and uncertainty The SDK's TraceIdRatioBased head sampler can drop spans at creation with minimal overhead, while the Collector's tail_sampling processor buffers complete traces to retain only those matching latency, error, or attribute policies. Tail samp
opentelemetry