Limits of alert inhibition propagation in Rancher's centralized Alertmanager

The goal is to ensure that inhibition rules defined in a Rancher‑managed Alertmanager suppress duplicate notifications for related alerts firing across multiple clusters. Rancher ships with default alert rules and allows label‑based routing and silencing, but the behavior of inhibition when a single Alertmanager instance receives alerts from several clusters is not fully documented. Specifically, it is unclear whether inhibition rules need to be defined globally, per cluster, or with particular label matchers to propagate consistently.

What configuration ensures that an inhibition rule created in one cluster is respected for alerts originating from any other cluster? Are there required label matchers (e.g., cluster, alertname) that must be present for cross‑cluster inhibition to work? Does this behavior depend on a specific Rancher or Alertmanager version?