[browningstreet]

Ehh do you remember the defaults back in the day? And how long local vs intermediary vs backbone TTLs could be cached for, even above and beyond the set TTL?

The propagation part refers to how long it would take for all those cached requests to expire and when you could tell some random client they should be able to see the new value. Especially when you forgot to lower the TTL ahead of time.

It’s a term of art and it’s fine.

Oh that reminds me. I made a bunch of DNS changes a while ago and left all the TTLs set to 5 minutes. I should up them.

[teddyh]

> local vs intermediary vs backbone TTLs

This is an example of the myth in action. There are no such things. There is a single resolver, which you use, and a set of authoritative server, which that resolver will query when the TTL in the resolver’s cache times out. There is no chain of resolvers.

[inigyou]

What do you call it when an application queries systemd-resolved and systemd-resolved queries PiHole and PiHole queries your home router and your home router queries your ISP? What is that if not a chain?

[JdeBP]

Forwarding proxy DNS servers leading to a (yes) single resolving proxy DNS server at the end.

* https://jdebp.uk/FGA/dns-server-roles.html

[teddyh]

See this old thread: <https://news.ycombinator.com/item?id=19654515>

[toast0]

> There is no chain of resolvers.

There absolutely can be a chain of resolvers.

It's very common to run a caching recursive resolver on client nodes that sends its requests to an external recursive resolver (usually the network owner's suggested server). Nicer local caching resolvers will recurse themselves if/when the configured forwarding destination is unavailable. Some decades ago, it wasn't unusual for ISPs to run distributed recursive caching resolvers that would go through a centralized cache ... some of those systems may still exist?

If all of those servers respect and propagate (or pass on, whatever) the authoritative TTL as I think RFCs suggest, then you should see all the caches behind a centralized resolver expire at the same time and then get refreshed on their next request. Of course, some caches don't respect TTLs and some might return the TTL they received (or modified to) in responses from cache and who knows what other garbage they do.

That's why people want to know approximately how much of the population can see my changed records. Sometimes where in the world can people see my changed records ... not because the change is organized geographically, but because showing where people see one value vs the other is more visually stimulating than a % line... and also because that one dumb ISP server that has no cache expiration is likely to cause problems for users in one geography and not worldwide, so it's handy to know that users in that country will have problems when you turn off old servers that you took out of dns weeks ago.

Maybe propagate isn't the best verb for DNS changes to make their way from your source of truth to everyone else's resolutions, but you're welcome to suggest another.

Merriam webster includes these definitions of propagate that I think fit [1]:

> (transitive verb) 3a to cause to spread out and affect a greater number or greater area

> (intransitive verb) to travel through space or a material

Maybe you don't think the first one fits because of 'cause' in there. When you start returning new results, you don't do anything specific to cause the caches to fetch a new result, but... the new results do affect a greater number of caches over time.

We could agree to use percolate though, it was suggested elsewhere in the thread and it's a fun word and easier to spell. It will take time for this change to percolate through our distributed conciousness though. :p

[1] https://www.merriam-webster.com/dictionary/propagate

[teddyh]

> We could agree to use percolate though, it was suggested elsewhere in the thread and it's a fun word and easier to spell.

A more accurate term would be to talk about waiting until old DNS entries “age out”, or “expire”.

[teddyh]

> There absolutely can be a chain of resolvers.

A chain of proxies does not change how we should view the situation. Proxies, by definition, should not affect anything. See this old thread: <https://news.ycombinator.com/item?id=19654515>