Domains, CDN & Web Delivery

DNS Propagation

Definition

The name is misleading and it causes real confusion. Your edit reaches the authoritative nameservers the instant you save it, normally within a second or two, and from then on every fresh query gets the new value. What takes time is everyone who already asked the question and was told how long to remember the answer. That length was the TTL on the record as it stood before the edit, not the one you set during it. So the period people call propagation is not new data travelling outwards — it is old data dying off on a timetable you chose in advance, possibly years in advance.

Why It Matters

The practical consequence is that the useful moment to act is before the change, not after. Drop a record's TTL to 300 seconds a day ahead of a migration and the switch is effectively done in five minutes. Edit a record that has been sitting at 86400 and then lower the TTL, and you have missed the window entirely: resolvers holding the old answer keep it for up to another 24 hours no matter what the zone now says. The visible symptom is a site that works for you, works for your colleague, and does not work for the client who looked at it an hour before you made the change.

How It Works

Each resolver counts down its own copy independently, and they all started at different moments, so they expire at different moments too. That is why the changeover is ragged rather than instant, and why a page can load for one person and fail for another in the same office. A query against the authoritative nameservers shows the current truth immediately, since it skips every cache in between; a query through a public resolver shows whatever that particular resolver is still holding. The world-map propagation checkers do nothing more exotic than the second of those against a list of resolvers, which is why the map fills in patchily. A resolver that never had the old answer sees the new one on its very first try.

Real-World Example

A studio moves files.hart.co.uk from an old server to a 99helpers-hosted press kit. Two days earlier they set the record's TTL to 300. On the day, they change the record at 09:00, and by 09:06 every check they run returns the new target, with the 80 MB kit downloading from the edge. Had the record still been at a day-long TTL, journalists who had opened the old link that morning would have gone on hitting the retired server until the following morning.

Common Mistakes

  • Lowering the TTL at the same time as making the change — the old value is already cached, so the shorter lifetime only helps the next move
  • Assuming 24 to 48 hours is the normal wait — at a 300-second TTL it is over in minutes, and the long figure really belongs to nameserver changes at the registry
  • Flushing your own cache and calling it fixed — your machine is one resolver among thousands, and the ones holding up your visitors are not yours to flush

Related Terms

Put a file online in seconds

Drop in a document, an image, a page or a whole static website and share the link — free, with no build step and no server to set up.

Host a file free →