zlacker

Any DNS resolver libraries have a 4.5 second timeout? Maybe their infrastructure is just rotting.

>>lamont+(OP)
It'd be weird that it's limited to specific outbound domains in that case.

>>minima+D
domains load balanced across different servers in some overly complicated distributed topology with only one of them busted?

although seems unlikely it just happens to be the NYT.

>>lamont+(OP)
glibc defaults to 5 sec, but the server wouldn't need to resolve the redirect domain -- that'd be the client's job. Unless it's doing so for some proprietary reason, of course.

Or did you mean failing to resolve some internal service's hostname?

>>lamont+R4
I don't think it's a problem with rotting infrastructure. Using curl those requests are quick, unless you pass user-agent from browser. 4.7s with firefox user-agent and 1.3s without it. Using twitter link to the same NYT article.

>>runlev+36
> Or did you mean failing to resolve some internal service's hostname?

Yeah, something more like that where the internal service is somehow 'sharded' due to some overly complicated distributed database nonsense, and there's a DNS lookup that is failing. Of course that'd mean the DNS lookup wasn't cached, so you're taking that normal latency on every single hit, which would be terrible architecture. The curl-vs-wget performance isn't explained by that though (although that's a bit weird in and of itself, and might suggest that they had to allow that for some internal tool that they didn't want to punish).

> glibc defaults to 5 sec,

The timeout being close to 5 seconds is what made me wonder about it. Its just off though.