Tracing is an essential tool for developers and engineers to gain insights into the performance of distributed applications, and Grafana Labs utilizes frameworks like Jaeger and OpenTelemetry for this purpose. While traditional head-based sampling methods often miss important traces due to their probabilistic nature, tail-based sampling evaluates complete trace data to make more informed decisions. However, the challenge arises when the OpenTelemetry Collector is scaled horizontally, leading to partial trace sampling and incomplete data. Grafana Labs addresses this issue by introducing the aggregate processor, which ensures all spans of a trace are sent to the same collector instance, allowing for accurate tail-based sampling. This processor uses a combination of fnv hash and jumphash to effectively distribute spans and minimize data loss during scaling. The implementation has proven successful in Grafana Labs' internal infrastructure, maintaining consistent trace sampling even during traffic spikes, and the processor is available for public use through their GitHub repository and Docker image.