Is Fly.io slower today?

21

Oops - just noticed I haven’t been replying directly to your messages, so tagging y’all here! @kurt @eli.

To summarize I’m still having issues with:

  • Sub-optimal routing of my requests to fly.io
  • Fly.io VMs having weird variance in latency to my DB & cache hosted on DigitalOcean (mostly ~15ms but sometimes sub 1ms)
22

So in the traceroutes that you sent over it looks like your ISP is sending IPv6 traffic over a different transit provider, where it’s hitting our upstream ISP in New York:

curl ipinfo.io/2607:f740:70:101::6
{
  "ip": "2607:f740:70:101::6",
  "city": "New York City",
  "region": "New York",
  "country": "US",
  "loc": "40.7143,-74.0060",
  "org": "AS36236 NetActuate, Inc",
  "postal": "10004",
  "timezone": "America/New_York",
  "readme": "https://ipinfo.io/missingauth"
}

In that case I’d recommend reaching out to your client ISP to see if they are able to prefer a route that can reach us in Toronto.

For the DO issue, to clarify – only fdaa:0:756b:a7b:aa2:6019:cf16:2 has 1ms latency to DO, and the other app instances are consistently closer to ~15ms, but never much lower?

23

OK re: my ISP, that’s helpful thank you! Regarding the DigitalOcean issue, yes that’s correct. All my other app instances (~5 across 3 different apps) in the YYZ region are consistently closer to ~15ms.

1 Like
24

Thanks for confirming! Checking out these two instances for the DO issue, the traffic for the 1ms vm is initially coming from slightly different locations vs the 15ms ones.

If you wouldn’t mind, it could help to have a traceroute from one of the 15ms VMs to DO :slightly_smiling_face: In any case, we’ll check in with our providers to see if we can improve this for you.

As an aside, we’ve also reached out to Rogers for you, but you’ll likely have better luck as a customer of theirs.

1 Like
25

Amazing, thank you!

Here’s a traceroute for one of the 15ms VMs to DO:

traceroute to api.usepastel.com (174.138.112.93), 30 hops max, 46 byte packets
 1  172.19.8.17 (172.19.8.17)  0.145 ms  0.094 ms  0.069 ms
 2  unknown.ord.scnet.net (205.234.149.65)  0.684 ms  2.766 ms  0.694 ms
 3  67.223.96.121 (67.223.96.121)  0.726 ms  0.912 ms  0.804 ms
 4  *  *  *
 5  *  *  *
 6  *  *  *
 7  *  *  *

And here’s a traceroute for the 1ms VM to DO (for comparison):

traceroute to api.usepastel.com (174.138.112.93), 30 hops max, 46 byte packets
 1  172.19.2.153 (172.19.2.153)  0.277 ms  0.199 ms  0.156 ms
 2  153.220.248.162.ptr.anycast.net (162.248.220.153)  5.963 ms  11.670 ms  14.798 ms
 3  *  *  *
 4  *  *  *
 5  138.197.249.87 (138.197.249.87)  28.332 ms  0.516 ms  138.197.249.91 (138.197.249.91)  0.495 ms
 6  *  *  *
26

Our network folks chased down DigitalOcean and you should see a lot better latency to your DBs now.

2 Likes
27

Amazing, thank you! Latency from my Fly VMs to DigitalOcean DBs is down to 1ms or less across the board.

1 Like