Grafana Cloud's Adaptive Traces now includes a volumetric sampling policy that dynamically balances trace sampling across services and attributes instead of applying a flat probabilistic percentage. Probabilistic sampling preserves traffic imbalances (e.g. a high-volume checkout service drowning out low-frequency services), requiring tedious manual policy tuning. The new volumetric policy automatically selects attributes like service.name, status.code, and k8s.cluster.name, buckets traces, and continuously recalculates sample rates. Testing shows about 25% higher information density (Shannon entropy) versus probabilistic sampling at the same rate. It's a one-click upgrade for existing Adaptive Traces users and default for new users, working alongside anomaly detection, diversity sampling, and standard policies; Adaptive Traces users reportedly cut write volume by 75-90% on average.
Table of contents
The volumetric policyWhat's wrong with doing probabilistic sampling?What is dynamic sampling?The Adaptive Traces sampling toolboxQuestions this post answers
What is the volumetric sampling policy in Grafana Cloud's Adaptive Traces?
It is a dynamic sampling policy that automatically picks attributes like service.name, status.code, and k8s.cluster.name, buckets traces by those attributes, and continuously recalculates sample rates per bucket to hit a target percentage. It replaces flat probabilistic sampling, which preserves traffic imbalances by sampling all traces at the same rate regardless of volume. Teams tuning trace sampling budgets can follow observability tooling updates like this on daily.dev.
How much more information does volumetric sampling retain compared to probabilistic sampling at the same rate?
Volumetric sampling delivers about 25% higher information density, measured using Shannon entropy, than probabilistic sampling at an equivalent sampling percentage. This means roughly a quarter more unique information is retained for the same amount of stored data, giving a more diverse and representative set of traces for the same cost. Engineers evaluating tracing costs versus data fidelity can track findings like this on daily.dev.
How do I upgrade an existing probabilistic sampling policy to volumetric sampling in Grafana Cloud Adaptive Traces?
Existing Adaptive Traces users see a recommendation to convert their probabilistic policy to a volumetric policy with a single click, since volumetric is a drop-in replacement. New users get a volumetric policy configured automatically as part of the onboarding flow's starter policies. daily.dev helps engineers stay on top of drop-in upgrades like this for their observability stack.