Network & Infrastructure / DNS, DNSSEC & Service Discovery
Anycast resolver changes and cache-dependent user behavior
After an anycast routing shift, some users see different cached answers or latency. The new resolver instance may be healthy but hold a different cache and upstream history.
Environment and inputs: Authoritative and recursive DNS infrastructure; BIND examples are conceptual and require release-matched configuration. Diagnostic/design example; not tested on production equipment.
Collect the evidence
Correlate routing changes, serving-instance identity, cache hit ratio, upstream latency and returned TTLs. Compare authoritative truth before treating every answer difference as corrupted DNS.
Explain the result
Anycast stabilizes a service address, not necessarily the state of every instance. Resolver caches and temporary failures can differ even when instances share the same policy.
Plan the fix
Standardize configuration and health withdrawal while treating cache state as locally transient unless the design explicitly synchronizes it. Ensure monitoring can identify the serving instance.
Check the result and rollback
Test controlled route movement and both warm and cold cache behavior. Revert an anycast change if the destination instance cannot meet service objectives or serves a different policy view.