kubelet now uses cgo for all DNS queries

[bcressey]

Image I'm using:
aws-k8s-1.28

What I expected to happen:
When bootstrapping a cluster, kubelet should try multiple entries for the same host in /etc/hosts:

127.0.0.1 localhost localhost.localdomain localhost4 localhost4.localdomain4
::1 localhost localhost.localdomain localhost6 localhost6.localdomain6
10.0.10.1 asdfasdfasfd.aws.dev
10.0.10.2 asdfasdfasfd.aws.dev
10.0.10.3 asdfasdfasfd.aws.dev

What actually happened:
kubelet tries to connect to the first IP repeatedly.

How to reproduce the problem:
Using GODEBUG=netdns=2 we can see that cgo rather than netgo is now used.

Sample program:

package main

import (
	"fmt"
	"net"
	"os"
)

func main() {
	addrs, err := net.LookupHost(os.Args[1])
	if err != nil {
		fmt.Println(err)
	} else {
		fmt.Println(addrs)
	}
}

List multiple IPs for the same host:

apiclient apply <<EOF
[settings.network]
hosts = [
 ["10.0.10.1", ["asdfasdfasfd.aws.dev"]],
 ["10.0.10.2", ["asdfasdfasfd.aws.dev"]],
 ["10.0.10.3", ["asdfasdfasfd.aws.dev"]],
]
EOF

Query the host:

# GODEBUG=netdns=2 dns-test asdfasdfasfd.aws.dev
go package net: confVal.netCgo = false  netGo = false
go package net: dynamic selection of DNS resolver
go package net: hostLookupOrder(asdfasdfasfd.aws.dev) = cgo
[10.0.10.1]

This happens because netgo sees the new "resolve" directive in nsswitch.conf, and knows it cannot use that NSS plugin so it falls back to cgo.