Inconsistent alert deduplication across mesh members

The goal is to silence repeated notifications for the same error that occurs across multiple service instances in a Traefik Mesh deployment, while still being alerted to genuine problems. This requires knowing how the mesh groups alerts before they reach Alertmanager, so that silencing rules can be applied without suppressing distinct incidents.

However, the deduplication mechanism that Traefik Mesh uses to combine alerts from different mesh members is not clearly documented, and reports indicate that the behavior changed between version 2.9 and version 3.0. Without clarity on which labels or internal identifiers are used for grouping, it is difficult to design reliable Alertmanager routing or silence configurations.

Which label or set of labels does Traefik Mesh v2.9 use to identify duplicate alerts across mesh members? Does Traefik Mesh v3.0 modify this deduplication scope, and if so, what is the new grouping criterion? How can Alertmanager be configured to complement mesh‑level deduplication without inadvertently silencing separate error occurrences?