UDP connectivity dead?

Questions / Help
1

Hello,

As of a few minutes ago, we’ve had UDP packets through Fly intermittently die - at least in the UK. This occurred previously below and it seems to be the same thing again:

https://community.fly.io/t/udp-connectivity-failing/

Is anyone else affected? We’re proxying DNS.

2

It now appears to have returned as of 2 minutes ago.

3

UDP, as a protocol, is susceptible to drops by design. So one cannot realistically expect it to always work at the internet scale. Prolonged UDP dropouts may point to problems with network though.

I think that those facts are widely known and understood, but maybe someone has high expectations regarding UDP. It’s always better to have a TCP fallback in place.

1 Like
4

Just because UDP doesn’t hold a connection, doesn’t mean that it’s ‘designed’ to have continuous drops that affect the program in use. If this was the case, online gaming would not be a thing. This was a prolonged outage at the UDP proxy layer Fly runs, it lasted around 40 minutes and completely killed UDP connectivity.

3 Likes
5

We’re seeing this outage again, is anyone else having this issue? TCP Fallback for DNS does not work for us for some reason, not sure why:

[[services]]
  internal_port = 53
  protocol = "udp"

  [[services.ports]]
    port = "53"

[[services]]
  internal_port = 53
  protocol = "tcp"

  [[services.ports]]
    port = "53"

EDIT: Came back now, lasted 6 minutes.

6

Are you deploying a new version of your app? I’ve noticed that I usually can’t connect to my UDP app for a few minutes after it boots up.

7

Unfortunately not, these are running dnsdist apps that have been going for about a year and just randomly start timing out. I have two clusters with different Fly regions, and they both go down at the same time which I believe is because of how the Fly UDP proxy works, and then it doesn’t failover to TCP for some reason.