I have a K8S 1.14 cluster setup with CoreDNS 1.5.0 as default DNS server. My ubuntu 14.04 client failed to resolve service name when I got a response with TTL less than the minimun TTL value in SOA field(response with IPv6 result, no IPv6 address resolved).

I tried to use the same client code in Ubuntu 18.04 which can resolve the service name every time. I'm using python socket.getaddrinfo to resolve the service name, using socket.gethostbyname works around this issue since it only queries the A records.

This is the code i used to test

import socket
import time

while True:
    try:
        time.sleep(2)
        print socket.getaddrinfo("kubernetes", "443")
    except Exception as e:
        print e

The result of running this code is as followed:

[(2, 1, 6, '', ('10.233.0.1', 443)), (2, 2, 17, '', ('10.233.0.1', 443)), (2, 3, 0, '', ('10.233.0.1', 443))]

[Errno -5] No address associated with hostname

[Errno -5] No address associated with hostname

[(2, 1, 6, '', ('10.233.0.1', 443)), (2, 2, 17, '', ('10.233.0.1', 443)), (2, 3, 0, '', ('10.233.0.1', 443))]

[Errno -5] No address associated with hostname

[Errno -5] No address associated with hostname

When i update the default ttl to 30 seconds, I got more failures util I got the response with TTL 30