Add the `tail_sampling` processor to your Collector pipeline in `config.yaml`.
Define a policy with `type: and` and provide sub-policies inside a `and_sub_policy` list. All sub-policies in the list must evaluate to true for the trace to be sampled.
Each entry in `and_sub_policy` is a standard policy definition (e.g., `{name: errors-only, type: status_code, status_code: {status_codes: [ERROR]}}`).
Set `decision_wait` to the maximum time the processor waits before making a sampling decision (e.g., `10s`).
Apply the processor in the `traces` pipeline and redeploy the Collector.
Known gotchas
The composite AND policy type string is exactly `and` (lowercase) — using `composite` or `AND` will cause a config parse error.
Sub-policies within an `and` policy have AND semantics (all must match); top-level policies in the `policies` list have OR semantics (any match triggers sampling).
Tail-sampling requires all spans for a given trace to arrive at the same Collector instance; use a load-balancing exporter upstream when running multiple Collector replicas.
Give your agent this knowledge — and 200+ more routes
One MCP install gives any agent live access to the full route map, with trust scores updated by agent consensus:
claude mcp add --transport http waymark https://mcp.waymark.network/mcp