Failed to fallback on IPv4 when IPv6 is not working

[tutaidalal]

I'm using an FQDN as the target in my server group. When I run nslookup, it returns both IPv4 and IPv6 addresses.

With curl, I can successfully fetch data from the Prometheus server group. However, Promxy always tries to connect using IPv6 first and doesn't switch to IPv4 if IPv6 is unreachable.

Is there any code change or configuration that can force my application to use only IPv4?

I am running the application in Kubernetes and the base os is alpine:3.18.10

[tutaidalal]

Seems like the Happy Eyeballs (RFC 6555) algorithm, which tries both IPv6 and IPv4 in parallel, is not implemented by default in Go’s net.Dial or http.Client. Is there any option to make it work to try all the IP resolved by Resolver Lookup

[jacksontj]

Promxy's connection to downstream prometheus API endpoints is all done through https://github.com/prometheus/client_golang -- which under the hood uses net.http which supposedly has this happy eyeballs RFC implemented.

If thats still not working; do you have an endpoint to test against?

[tutaidalal]

My endpoints are in private network and still having the issue. I can also see that when i set 'ignore_error' as false its clearly saying that its only trying to reach ipv6. For now I have modified the resolver so that any ipv6 which are not reachable will not be published and this work around worked for me. May be if i can simulate the same in some public network, will share the test endpoint.