DNS Infrastructure
I Created the Record. Why Does DNS Still Say It Doesn't Exist?
DNS caches missing names as well as existing records. Here's how negative caching delays a newly created record, how to read the SOA in the response, and what to verify before waiting.
November 19, 2026
You create launch.example.com with a five-minute TTL. The DNS dashboard shows the right address, but a public resolver still says the hostname doesn't exist.
The five-minute TTL may not be the timer you're waiting on. Before you created the record, that resolver may have cached the fact that the name was missing.
That's negative caching: remembering an unsuccessful lookup so the same question doesn't have to go back to the authoritative server every time.
There are two kinds of “missing”
An NXDOMAIN answer reports that a name doesn't exist. A NODATA answer indicates that the name exists but the requested record type is absent; its response code is normally NOERROR.
For example, a host could already have an A record while an AAAA query returns no IPv6 record. Adding AAAA later doesn't immediately clear a resolver's earlier negative answer for that type.
If the status is SERVFAIL or the query times out, begin with the DNS error guide. Those symptoms require a different investigation.
The negative answer has its own lifetime
The standard negative-cache lifetime comes from the smaller of the SOA record's TTL and its MINIMUM field. It isn't the TTL of the record you create afterward. RFC 2308 defines this behavior.
The SOA appears in the authority section of a negative response. In a cached response, its displayed TTL can show the remaining lifetime of that negative answer.
For the other fields, see the SOA explainer. Managed DNS providers may control these settings rather than exposing every field for you to edit.
A timeline makes the delay easier to see
Imagine this sequence with a 900-second negative-cache lifetime:
- 10:00: A resolver asks for
launch.example.comand learns that it doesn't exist. - 10:02: You create its A record with a TTL of 300 seconds.
- 10:03: A client asks that same resolver. It still has the earlier negative answer.
- 10:15: The original negative answer expires. A new lookup can retrieve the A record.
The new record's five-minute TTL applies when its positive answer is fetched and cached. It doesn't replace the earlier negative timer.
A different resolver that never asked before creation may return the record immediately. That difference is useful evidence, rather than a reason to keep deleting and recreating the record.
This timeline is an illustration of one cache. Real clients can use multiple caching layers, and resolver policies can affect the observed wait. Many resolvers also cap how long they keep negative answers, regardless of the SOA.
If your zone is signed with DNSSEC, one more wrinkle applies. Validating resolvers such as 1.1.1.1 can use the signed "nothing exists between these two names" proofs they've already cached to answer NXDOMAIN for names they've never asked about (RFC 8198). So a resolver can return a cached negative answer for your new hostname even if nobody ever queried that exact name.
Compare authoritative and recursive answers
Use your domain's actual authoritative nameservers in place of these examples:
dig @ns1.example.net launch.example.com A +norecurse
dig @ns2.example.net launch.example.com A +norecurse
dig @1.1.1.1 launch.example.com A
dig @8.8.8.8 launch.example.com A
If every authoritative server returns the intended record but a recursive resolver still returns NXDOMAIN, a cached negative answer is a plausible explanation.
If authoritative servers disagree, investigate zone synchronization first. If they all say the name is missing, verify the exact hostname, the active provider, and whether the record was saved in the right zone.
To inspect a resolver's response without losing the header and authority section:
dig @1.1.1.1 launch.example.com A +noall +comments +answer +authority
An illustrative negative response might include:
;; ->>HEADER<<- opcode: QUERY, status: NXDOMAIN, id: 12345
example.com. 720 IN SOA ns1.example.net. hostmaster.example.com. 2026111901 3600 600 1209600 900
Read the two numbers separately. The final 900 is the SOA MINIMUM field, the zone's configured ceiling for negative caching. The 720 near the start is the time this resolver has left: it will keep returning NXDOMAIN for about 12 more minutes. Query again and that number keeps counting down. It doesn't reset.
What helps, and what doesn't
Creating and verifying the record before advertising the hostname avoids many early negative lookups. That is especially useful before sending launch emails, configuring integrations, or asking users to try a new endpoint.
Be careful how you check, too. Asking a public resolver "does launch.example.com exist yet?" before the record is published is exactly what creates the negative cache entry you'll later be waiting on. Check against your authoritative nameservers instead, with dig +norecurse as shown above or the DNS Lookup tool, which queries authoritative servers by default. Those queries don't touch any public resolver's cache.
Flushing your own computer's DNS cache may clear a local copy. It doesn't flush a public resolver, a colleague's router, or an application's separate cache.
Some resolver operators provide a cache-purge facility, such as Cloudflare's 1.1.1.1 purge page and Google's Public DNS flush tool. That can help with that operator's cache, but it isn't an internet-wide reset. Lowering the new A record's TTL also won't invalidate an existing negative answer.
Avoid repeated delete-and-recreate cycles during troubleshooting. They add changes to the evidence without necessarily shortening the delay.
SERVFAIL can be cached too
Resolution failures can also be cached, but that isn't the same mechanism as caching an authoritative “name missing” answer. RFC 9520 addresses caching of resolution failures, including DNSSEC validation failures.
If you're seeing SERVFAIL, investigate the failure before assuming an SOA negative-cache timer explains it. Fixing a missing name and repairing a broken chain of trust are different tasks.
How DNS monitoring fits in
During a launch, the first question is always "is the record actually published?" OneDollarDNS answers that from your authoritative nameservers. Add the new hostname as a monitored host when you create it, and its record history shows what your nameservers are serving, independent of any resolver's cache. If the history shows the record published and a resolver still says NXDOMAIN, you're waiting on a cache, not looking at a mistake. From then on, the same monitoring will alert you if that record is changed or removed.
From there, the resolver comparison view shows which public resolvers have picked up the new record and which are still holding the old negative answer. That separates "the record is published" from "this particular resolver has fetched it," and saves you from editing a record that was right all along.
Monitor your DNS for $1/domain/month
OneDollarDNS watches your DNS records and alerts you when anything changes.
Get started free