bristolianthrw

How would I use a load balancer to fail traffic between, say, London and Amsterdam with no fiber in place between them? Where would the load balancer physically exist in that scenario and how would it fail to the other when power is lost in one location? Would I make a third PoP to isolate it? What would then be my redundancy story for that PoP? How would I relocate traffic to my backup load balancer PoP number four?

Within a single network, sure, load balance all you want. That’s not the scenario low TTLs go after.

No, there isn’t. The specification as implemented requires no invalidation mechanism, which means no such mechanism across all caches exists, nor will it ever. The long tail kills you in such a failure scenario, and remember, people who make kitchen appliances write DNS resolvers.

For literally my entire career in SRE, well over a decade now, I’ve only interacted with systems that use DNS for this purpose, from small shops to parts of every HN reader’s life. That sentence in your quote, as well as the assertive nature of the post on such a weak foundation, are sufficient to disqualify a hiring candidate on account of lack of experience despite the authored software presented. It simply does not align with reality when presented with two logically separate networks and a required mechanism to transition between them.

The only other alternative for that scenario is using anycast addressing, and that has a colorful bag of limitations that are quite different from those of low-TTL DNS (including being out of reach for most).