Tail Sampling
A trace-sampling strategy that decides whether to retain a trace after observing some or all of its completed spans, enabling error- or latency-aware selection.
Observability Context
Tail sampling delays the keep/drop decision until some or all spans of a trace have been observed. This allows policies that preferentially retain rare errors or high-latency requests, which can be difficult for head sampling to predict.
Resource Boundary
The decision requires buffering and state in the collector, increasing memory use and decision latency. Tail sampling is also unrelated to the statistical concept of tail latency except that latency can be one of the retention signals.