> On 17 Oct 2022, at 10:58 am, Michael Richardson <mcr+i...@sandelman.ca> wrote: > > > Geoff Huston <g...@apnic.net> wrote: >>> On 17 Oct 2022, at 7:53 am, Michael Richardson <mcr+i...@sandelman.ca> >>> wrote: >>> I think that's because >>> recursive nameserves effectively have always done an equivalent to "happy >>> eyeballs", so the risk is low. > >> That certainly was not the case in 2015: >> https://www.potaroo.net/presentations/2015-10-04-dns-dual-stack.pdf > >> I have not seen a large scale measurement since then but I suspect that >> nothing has changed. i.e.: >> recursive resolvers do not do the equivalent of happy eyeballs >> behaviour. > > What I'm saying (based upon my understanding, and some long ago reading of > code) is that recursive nameservers remember which NS were too slow or > non-functional, and try to avoid them in the future. > (I agree that this isn't exactly happy eyeballs: we don't do requests in > parallel and pick the winner) > > So if my v6 infrastructure for my authoritative nameservers fails, but the v4 > continues to work, that I'm not too badly off. My experience is that this > resilience can often mask failed servers for months :-) >
that’s not happy eyeballs - that’s just resilience across multiple service access points. _______________________________________________ DNSOP mailing list DNSOP@ietf.org https://www.ietf.org/mailman/listinfo/dnsop
