Networks

Why DNS Changes Take Time to Propagate

Light at the ends of optical fibres
AI-generated illustration.
Inside this article
  1. What DNS propagation means
  2. How DNS TTLs set expectations
  3. Why users see different answers
  4. How DNS caches missing names
  5. Nameserver moves add another cache
  6. How to find the stale answer
  7. When old data survives expiry

After you change records in the Domain Name System (DNS), some visitors still reach the old server. DNS links domain names to records such as server addresses. DNS resolvers look up those records for users. Resolvers and devices save answers for set periods and refresh at different times. Users can therefore get different answers even after the authoritative DNS servers, which hold the domain’s official records, have the new record.

“Propagation” describes how DNS answers change over time and differ among users. To troubleshoot, confirm the authoritative servers have the new record, then identify which resolver or device is returning an outdated answer.

What DNS propagation means

The DNS provider publishes records on authoritative nameservers, which hold the official data for a domain. No central point sends a DNS change to every device. Recursive resolvers, such as those run by internet providers or public DNS services, look up that data for users and cache their answers. Devices can also keep local DNS answers.

Each resolver and device saves and expires DNS answers on its own schedule. One resolver might have fetched a record shortly before the change, while another fetched it afterward, so the same domain can return an old address on one network and a new address on another.

Start by checking the authoritative nameservers. If they still return the old record, waiting for user caches to expire will not fix the underlying problem. Confirm that the change was made at the DNS provider currently responsible for the domain.

How DNS TTLs set expectations

A record’s time to live, or TTL, tells caches how long they may reuse that answer before checking again. A TTL of one hour, for example, allows a resolver to use its cached answer for up to an hour. It does not schedule a global update at the end of that hour, and resolvers do not all start their timers together.

Changing a TTL after updating a record does not shorten the lifetime of answers that resolvers already cached under the previous TTL. For a planned change, lower the TTL in advance. Let the old, longer TTL pass before making the change. After the change has settled, restore a longer TTL if that suits the domain.

A TTL estimates how long a cache can use an answer. It does not set an exact deadline for every resolver. The DNS standards let resolvers limit how long they keep a received answer. How they do this depends on their implementation. A zero TTL tells caches not to keep the answer beyond the transaction, but it does not erase answers already cached elsewhere.

Ethernet cables in a network rack
AI-generated illustration.

Why users see different answers

A recursive resolver can answer from its cache instead of contacting an authoritative nameserver. A device can also check its own DNS cache before contacting that resolver. Those caches can hold different answers and have different amounts of time left before they expire.

This explains why switching networks or DNS providers can change what a device sees: the new connection may use a different resolver with a different cached answer. To narrow down the cause, compare the answer from the affected device’s configured resolver with answers from other resolvers and the authoritative nameservers.

How DNS caches missing names

DNS caches store some negative answers as well as records that exist. An NXDOMAIN answer means the queried name does not exist. A NODATA answer means the name exists but has no record of the requested type. For example, a resolver could cache the absence of an IPv6 address even though an IPv4 address exists.

If someone looked up a hostname before it was created, a resolver may keep returning NXDOMAIN until its negative cache entry expires. The negative-cache lifetime comes from timing information in the zone’s Start of Authority (SOA) record. When checking a negative answer, inspect the SOA details in the response as well as the answer itself.

Resolutions can also fail for reasons other than a name or record being absent. DNS standards cover caching resolution failures, including DNSSEC validation failures. If a lookup returns SERVFAIL, check for a resolution problem rather than assuming the record simply needs more time.

Nameserver moves add another cache

Changing a record inside a DNS zone and changing the domain’s nameservers are different updates. A nameserver change alters the delegation, the information that directs resolvers to the servers responsible for the zone. Resolvers can cache that direction separately from the records inside the zone.

After you move DNS hosting, some resolvers might keep sending queries to the old provider until their cached delegation information expires. Check which nameservers the domain currently delegates to, and verify that the new zone contains the intended records. If a hostname uses a CNAME, which points one name to another, check each name in the chain too.

Access-card forms and a security key
AI-generated illustration.

How to find the stale answer

Compare answers from the authoritative nameservers with answers from the recursive resolvers used by affected users. Use tools such as dig and nslookup to query DNS. Check the returned address and record type. Then check the TTL. A remaining TTL on a positive answer indicates how long that cached answer may still be used under ordinary caching. For NXDOMAIN or NODATA, examine the SOA information in the response.

Use the results to find where the old answer is coming from:

  1. Authoritative servers return the old answer: Confirm that you changed the active DNS zone and that its authoritative servers serve the update.
  2. Authoritative servers return the new answer, but a recursive resolver returns the old one: Check the resolver’s cached answer and delegation. Also check any CNAME targets.
  3. The resolver returns a negative answer: Check whether the name or requested record type existed when the query was made. Then inspect the negative-cache information.
  4. The resolver returns the new answer, but one device does not: Check the device’s local DNS cache and confirm which resolver it uses.

On Windows, ipconfig /displaydns shows the DNS client cache, and ipconfig /flushdns clears it, including cached negative answers. That command clears only the Windows device’s local cache. It does not clear an internet provider’s or public resolver’s cache, or the caches on other devices.

When old data survives expiry

Some recursive resolvers keep serving expired data if they cannot get an updated answer from authoritative servers. This behavior, called serve-stale, helps keep domains reachable during an outage at the authoritative servers. It also means an old answer can appear after its ordinary TTL has expired.

Serve-stale is a resilience option, not a universal resolver setting. If a resolver keeps returning an old answer after the TTL reaches zero, compare it with authoritative answers and check whether the resolver can reach those servers. The expired TTL alone does not prove that the resolver is serving stale data.

About this article

Created with AI assistance using published sources. Illustrations are AI-generated. Redbag does not claim hands-on testing or exclusive reporting. Check the linked sources for changing product details.

Our approach ↗

Keep connecting the dots.