GoKawiilLast month, the VictoriaMetrics team gave a talk on retroactive sampling at KubeCon Europe 2026.
By writing this blog post, as a transcript of the session, we want to explain how retroactive sampling reduces outbound traffic, CPU, and memory usage in the data collection pipeline significantly compared to tail sampling in OpenTelemetry.
Benchmark Comparison (Lower is better. Details in section 3)
1. Background
#
We’ll start with concepts for the beginners, but feel free to skip ahead to section 2 if you’re already familiar with these concepts - you won’t miss anything important.
1.1 Distributed Tracing
#
A Trace
The diagram above shows how a trace is visualized. The concept of distributed tracing is straightforward: you have a distributed system with dozens of microservices that call each other as they process a single request. Each microservice records its work as a span, and spans are organized into a trace by the parent-child relationships.
... continue reading