> On 14 Nov 2018, at 22:18, Brandon Butterworth <[email protected]> wrote: > > On Wed Nov 14, 2018 at 09:26:15PM +0000, Chris Malton wrote: >> traceroute to bbc.co.uk (2a04:4e42:600::81), 30 hops max, 80 byte packets > > inet6num: 2a04:4e40::/29 > netname: US-FASTLY-20130718 > > So cdn, it's down to fastly/HE so neither directly under > our (BBC) control
Yes, around this time last year we placed the bbc.co.uk domain apex behind Fastly (who we use for other purposes) to provide a greater level of DDoS protection as well as better global redirect performance (as all bbc.co.uk does is 301 to www.bbc.co.uk, so the CDN serves this directly). > >> 19 2a04:4e42:600::81 (2a04:4e42:600::81) 326.976 ms 326.960 ms 326.940 ms > > That's a long way away, looks like it's not a HE routing issue as that > realy is in brisbane. The question is why for UK requests > Fastly are handing out an AU node, maybe it is supposed to be anycast > and something happened to our local nodes. Yes - in the case of requiring fixed IPs rather than CNAMEing to a CDN (required in the case of a zone apex) , Fastly offer up both v4 and v6 Anycast IPs. I strongly suspect an incorrect BGP announcement mucking up the Anycast routing is the root cause but can’t confirm that as it’s with the provider to check. > Looks like it's time to ask fastly unless the DNS is intended > to resolve locality (also outsourced, was easier when it was all > BBC stuff that might break) It’s been raised internally (OPS-209888) and has been passed on to Fastly support. Cheers Paul > >> Is that any help to you? > > Yes, any data is better than no data, thanks > > brandon >
