Why a DNS change does not apply everywhere at once
Resolvers remember answers for as long as the record's TTL allows. If you change an A record with a TTL of 86400, resolvers that saw it yesterday serve the old value for up to 24 hours – each resolver on its own, depending on when it last asked. That is why a visitor in Berlin already sees the new site while one in Munich still sees the old one. "Propagation" is not distribution, it is caches expiring.
How to read the result
- At the top is the state of the authoritative name servers – the target. If it still returns the old value, the zone has not been reloaded or the change was not saved at the hoster.
- "current": the resolver returns exactly the authoritative state. "still stale": it holds a different value in its cache, which expires with the old TTL.
- "no answer": the resolver was unreachable or exceeded the time budget – no verdict.
- A single anycast resolver such as 8.8.8.8 answers from many data centres; our request hits one of them. That another location of the same provider is still stale cannot be ruled out this way.
Before a migration: lower the TTL
Set the TTL of the affected records to 300 seconds one or two days before the change. Then caches expire within five minutes after the switch instead of producing mixed states for a day. Afterwards you can raise the TTL again. DomainWarn monitors DNS records continuously and reports every change – including the ones you did not make yourself.