A platform team scales a collector tier running tail sampling and starts seeing traces with missing spans because related spans land on different collector replicas. Which design best fixes this?

Question

Accepted Answer

B. Place a load-balancing exporter layer in front of the stateful processors and route by trace ID. The scaling guide explains that stateful components like tail sampling cannot be naively scaled because related spans for one trace may hit different replicas. The recommended fix is a load-balancing exporter layer that consistently hashes by trace ID to keep related spans together. Option d is wrong because reusing the same receiver name does not solve cross-replica routing for a distributed collector tier.

A platform team scales a collector tier running tail sampling and starts seeing traces with missing spans because related spans land on different collector replicas. Which design best fixes this?

Related Questions

Discussion