DNS Delegation & Nameserver Health
Query the parent referral and every delegated server directly and compare NS sets, glue, serials, and DS.
DNS SecurityA domain resolves only when every step from the parent zone to your name servers agrees. When one step does not, the failure is partial and random-looking — some resolvers succeed, some time out, some get stale data — and an ordinary recursive lookup hides exactly the disagreement you need to see.
The parent zone — for a registered domain, the TLD registry — publishes NS records for your domain and, where needed, glue addresses. Those parent-side records are not authoritative data of the parent; they are a referral. Your own zone publishes the authoritative NS set at its apex. RFC 1034 expects both sets to be exactly the same, yet they are maintained through different channels: the parent through your registrar, the child in your DNS provider's zone. After migrations and provider changes, they drift apart.
Resolvers start with the parent's set and often switch to the child's once they reach your servers, so the effect of drift depends on the resolver implementation and its cache state. A name listed only at the parent receives queries it may not be able to answer. A name listed only at the child is never used by a resolver that starts cold at the parent. Neither shows up in an ordinary lookup, which returns whatever one recursive resolver currently has cached. The DNS Delegation & Nameserver Health check therefore asks the parent's own servers for the referral and compares it with the apex NS set that each of your servers returns.
If example.com is served by ns1.example.com, a resolver would need the server's address before it could ask that server for the address. Glue breaks the loop: the parent includes the A and AAAA records of in-domain name servers in its referral. RFC 9471 makes this explicit — a referral must include all available glue for in-domain name servers, or set the TC flag so the resolver retries over TCP. For sibling-domain servers, which sit elsewhere in the same parent zone, glue should be included where it fits; servers in unrelated zones need none, because resolvers look up their addresses separately.
; referral from the com servers for example.com
example.com. 86400 IN NS ns1.example.com.
example.com. 86400 IN NS ns.dns-provider.example.
ns1.example.com. 86400 IN A 192.0.2.53 ; in-domain: glue required
Glue is registered at the registrar, usually as a host object, and it does not follow changes in your zone. Renumber a name server, update its A record in your zone, forget the registrar, and resolvers starting at the parent keep going to the old address until it stops answering. The check compares glue with the addresses seen through a resolver and flags mismatches, missing required glue, name servers that are aliases (an NS target must not be a CNAME, RFC 2181 Section 10.3), and names that resolve to non-public addresses.
The decisive test is direct: send each delegated name server address a non-recursive query for the zone's SOA and NS and read the reply. A healthy server answers NOERROR with the AA (authoritative answer) bit set. REFUSED, or an answer without AA, means the server is listed in the delegation but not configured for the zone — traditionally called a lame delegation. SERVFAIL means it is configured but cannot load or serve the zone. A timeout means no reply at all, over UDP or TCP. Resolvers paper over a minority of broken servers by retrying others, which turns a hard failure into extra latency and occasional SERVFAILs that are almost impossible to reproduce on demand.
Count and placement matter as well. A delegation needs at least two name servers, and RFC 2182 recommends three for most organization-level zones, with at least one well removed from the others. Two names at the same provider still share much of one failure domain. Servers must also answer over TCP, not only UDP: a response that does not fit into UDP is truncated, and a server that truncates but cannot complete the answer over TCP fails resolvers on exactly the large answers — DNSSEC keys, long TXT sets — that need it (RFC 7766).
Secondary servers transfer the zone from the primary when they see a higher SOA serial. A brief serial difference right after a change is normal. A lasting one means a secondary no longer receives updates and serves stale data, and once the SOA EXPIRE interval passes without a successful refresh, that server is no longer authoritative for the zone at all (RFC 1035). Compare the serials of all authoritative answers after every change; the check lists them side by side. An SOA MNAME that does not appear in the NS set is common with hidden primaries and only informational.
For a signed zone, the parent also publishes DS records — digests of the zone's key-signing key. If the DS RRset is authenticated and contains a record with a supported DNSKEY algorithm and digest type, a supported DS record must authenticate a matching child DNSKEY and the DNSKEY RRset; otherwise validation fails as bogus and the resolver returns SERVFAIL. If no DS record uses a supported algorithm and digest type, the delegation is insecure, not automatically bogus. The check verifies this DS-to-DNSKEY link from the parent's and the child's own answers. It does not validate RRSIG signatures or the chain above the parent; the DNSSEC Validator and a validating resolver cover those, and Diagnosing DNSSEC Outages walks through the failure modes.
Most delegation incidents are migrations performed as a flip. Run them as an overlap instead. Build the zone at the new provider and verify it by querying the new servers directly before touching the delegation, including the apex NS set the new zone will serve. Decide DNSSEC before the move: either keep a valid chain throughout with a key procedure coordinated between both operators, or make the zone insecure first by removing the DS at the parent and waiting until it has expired from caches. Then change the NS set and any glue at the registrar, keep the old provider serving the identical zone until both the parent's NS TTL and your own have passed, and only then decommission it.
After each step, check from the parent down — referral, glue, each server's authoritative answer, serials, DS — rather than from a recursive resolver up; the DNS Lookup remains the right tool for the resolver's view once the delegation is sound. Keep the vantage point in mind: the direct queries run from this site's hosting environment, within a fixed query and time budget, and test reachability from there. If that host cannot send direct DNS queries at all, the check says so instead of reporting your servers as down; a server filtered only for some networks needs a second vantage point.