I don't know who needs to hear this, but it's never propagation.

[swelljoe@mas.to]

I don't know who needs to hear this, but it's never propagation. Not anymore.

Sure, maybe if you're updating glue records it'll take a few minutes, maybe an hour. But, DNS infrastructure is so much less aggressively cache-y than it used to be. It's not hanging on to old records for 24-48 hours, as lots of people say you should wait for propagation. If you're getting bogus records after an hour or even a few minutes, you did something wrong. Usually it's just the TTL, probably <5 minutes.

[matt@toot.cafe]

@swelljoe Is it a bad idea to just set the TTL to zero? I'm using Amazon Route 53 for DNS, so I guess I'm paying a little extra for additional queries that way, but it seems to work fine.

[swelljoe@mas.to]

@matt it will make things a bit slower for your users. I usually do 60 seconds for stuff that I expect to change regularly or when I'm planning a migration to a new server/IP, 5 minutes for stuff I don't have any plans to change.

Zero TTL would be, IMHO, for services that move around a lot via some dynamic mechanism. I can't think of any time I'd set TTL that low, or any benefit to doing so.

But, because Route 53 is very distributed with Anycast, it probably doesn't have a huge impact.