Open aidanharris opened 1 week ago
@rust-lang/infra ⬆
crates.io is served through AWS CloudFront, which is where the DNS records point. Just playing with dig
a bit myself, I can see that the name resolution rotates through different networks and changes based on geographical location.
We have very little control over the specific DNS records, since we have set up the A and AAAA records for crates.io
as aliases for the CloudFront distribution, so everything is handled by AWS internally.
Do you have the problem on other networks as well?
I can reach it just fine from other networks. It definitely seems to be some sort of routing issue on Amazon's part.
Right now I am routing 2600:9000:2046::/48 through a VPS over a gre6 tunnel because I got sufficiently annoyed by certain build systems not implementing Happy Eyeballs correctly and taking forever to fallback to IPv4.
It's working via France. I don't know how long this will last though. The DNS could change again:
$ traceroute6 -I crates.io
traceroute to crates.io (2600:9000:2171:fe00:c:7ed3:240:93a1), 30 hops max, 80 byte packets
1 * * *
2 2001:470:68c2::1 (2001:470:68c2::1) 0.328 ms * *
3 * * tunnel932964.tunnel.tserv5.lon1.ipv6.he.net (2001:470:1f08:fd::1) 10.049 ms
4 e0-19.core2.lon2.he.net (2001:470:0:67::1) 11.290 ms 11.525 ms 11.814 ms
5 * * *
6 * * *
7 amazon.par.franceix.net (2001:7f8:54::118) 17.809 ms * *
8 * * *
9 * * *
10 * 2a01:578:0:13::8 (2a01:578:0:13::8) 18.017 ms 18.034 ms
11 * * *
12 * * *
13 * * *
14 * * *
15 2600:9000:fff:ff00::401 (2600:9000:fff:ff00::401) 17.356 ms 17.351 ms *
16 * * *
17 * * *
18 2600:9000:2171:fe00:c:7ed3:240:93a1 (2600:9000:2171:fe00:c:7ed3:240:93a1) 17.307 ms 17.303 ms 17.327 ms
I'm not sure if this is the correct place to report this but I wanted to make you aware that I've been experiencing connectivity issues to crates.io over IPv6.
I'm not able to reach any of these addresses:
The above traceroutes are from two different machines in the UK using two different tunnel endpoints (also in the UK, using London endpoints)
I've spoken to the NOC of Hurricane Electric and this seems to be an artifact of some sort of load-balancing Amazon is doing. According to them they have withdrawn the entire :2046: prefix.
Is this something that can be looked into? From my end I can ping this :20e9: subnet and others but Amazon DNS still seems to be giving out records for the :2046: prefix I can't reach.