Comment by colmmacc | Hacker News Reader

Comment by colmmacc | Hacker News Reader

Parent Full thread

Edge cases and niggling "works ok for me" problems really do matter for something that's intended to be used with testing.

Otherwise it's easy to rat-hole for a long time trying to determine why your test isn't working, when it turns out it was a problem between your DNS resolver and an upstream domain.

Problems between DNS resolvers and DNS authoritative servers are classically intermittent; they usually depend on the ordering of a chain of steps to occur. For example, I might get a resolution failure for a xip.io record if one of the following sequences occurs;

  Client asks resolver asks for AAAA - gets NXDOMAIN from ns-1.xip.io, caches it
  Client asks resolver for A - responds with NXDOMAIN

but if the queries happen in the reverse order, things are fine.

or, another example;

  Client asks an AD DNS server to perform resolution, server chokes on lack of the AA bit in authoritative answer.

And so on.

But then in either case if a tester fires up nslookup or dig, everything works on the command line, and so they may spend quite a while trying to figure out why my library routine for connecting to my service isn't working.

What I took away from this, though, is that they just want to be able to load a web application they're developing on their iPad/iPhone or otherwise "restrictive" device that doesn't allow you to easily make local DNS modifications (such as /etc/hosts files).

I honestly don't understand what you've written above (although I re-read it a couple of times, I guess I'm just not knowledgeable enough about DNS for it to make sense) but can you see those issues impacting the ability of someone to load an application on their iPad in order to test it out?

I guess the problem might arise that people start to use this "not as originally intended" and get into all sorts of strife but for the particular scenario they were originally intending it for it seems perfectly adequate, no?

The issues reported so far could absolutely cause errors across any device, including an iPad. The problems that the DNS setup will cause affect resolvers - which can be a combination of software in your browser, your c library, local caching daemon, on your cable/dsl modem, in your ISP, and a public provider.

Many crufty resolvers - on things like wifi routers in particular - don't deal well with the lack of an AA bit, or a REFUSED answer. So a tester could easily end up with "works for me" and "not for me" reports that are really just down to the particulars of their network and resolver software, whether they have IPv6 enabled, and so on.

Edited to add: Again, I don't mean to rain on the developers parade. It's a great idea.

Writing DNS implementations is hard, and requires a certain kind of technical archeology to get to grips with the detail. DNS is a tricky protocol, chaotically and ambiguously documented. I've helped write 3 different ones - and I still get things wrong. And that said; anyone interested in writing hardcore DNS implementations that have to operate on the scale of microseconds per query should drop me a line.

The state that xip.io is in right now could, theoretically, result in DNS failures 50% of the time on any device.

Reply on news.ycombinator.com