Hacker News Reader

Comment by jtmarmon | Hacker News Reader

I'm updating a list of confirmed outages as I see them here https://news.ycombinator.com/item?id=12759520

So far twitter, etsy, soundcloud, spotify, github, pagerduty...crazy that this can even happen

I'm surprised; I would have thought such large sites would use more than one DNS provider? I mean:

    $ host -t NS twitter.com
    twitter.com name server ns4.p34.dynect.net.
    twitter.com name server ns3.p34.dynect.net.
    twitter.com name server ns2.p34.dynect.net.
    twitter.com name server ns1.p34.dynect.net.

I would have expected at least one of those to be somewhere else. What is the reason they would not have a backup provider?

I'd guess the reasoning is that DNS providers these days are all anycast-style DNS. A DDoS would usually just be a blip on a few servers around the world depending on where the attacks originate.

I'm not saying it's a good reason but it's a reason.

The other reason of course to use two different providers is to mitigate the automatic error propagation issue: https://twitter.com/devops_borat/status/41587168870797312

I know a lot about some things, but almost nothing about networking, so excuse me if this is a really dumb question but - would your physical location determine what hosts you returned from that query? Like if you were in Asia would you get different ones back?

Yes because DNS typically uses ANYCAST networking. The DNS request routes to the nearest location.

The network definitely can (and often does) do load balancing by offering different DNS results for different regions.

All of these work for me from Germany, and querying their authorative nameservers works just fine (so definitely no caching effect). Anycast for the win!

They don't work from Denmark, at 16:17 UTC.

https://www.whatsmydns.net/#A/twitter.com

Yep! Intercom is down as well

Isn't a major feature of DNS that it can be cached? Why aren't records being returned by ISP/Google DNS/OpenDNS servers? Is their TTL set that low?

They all work fine for me.

Caching is a magical thing :)

If anycast routing is in play, which is not unlikely with a DNS service like that, then it may also be that specific servers are being attacked so the outages don't affect users in all locations as some will be routed to infrastructure that is not affected.

Me too. I'm in Spain - definitely seems to be geographical.

Reply on news.ycombinator.com