nginx-ingress-controller not starting if IPv6 DNS server configured

MarkusTeufelberger commented 7 years ago

Issue Report Template

Tectonic Version

1.5.2

Environment

What hardware/cloud provider/hypervisor is being used with Tectonic? Provisioner in an LXD container, 1 Worker/1 Controller in VirtualBox VMs

Expected Behavior

Tectonic console is readily available a few minutes after reaching the "connect nodes" stage in the installer.

Actual Behavior

It seems like I am hitting https://github.com/kubernetes/contrib/issues/2188 at the "connect nodes" stage of the installer.

After investigating a bit, the controller node seems to run just fine, but on the worker node 2 containers (in docker ps -a) constantly are restarted: gcr.io/google_containers/nginx-ingress-controller:0.8.3 and quay.io/coreos/tectonic-console:v0.9.1

Looking at the logs for the nginx container shows that it has the following error, before killing the container (xxxx instead of actual IPv6 address):

E0227 13:57:07.959164       1 command.go:87] failed to execute nginx -s reload: 2017/02/27 13:57:07 [emerg] 16#16: invalid port in resolver "xxxx:xxxx:xxxx:xxxx::1" in /etc/nginx/nginx.conf:76
nginx: [emerg] invalid port in resolver "xxxx:xxxx:xxxx:xxxx::1" in /etc/nginx/nginx.conf:76

The tectonics-console container just seems to sync its config for a while (which fails, since nginx won't serve any content there), then time out and die.

Reproduction Steps

define an IPv6 DNS resolver for your network
install Tectonics according to the bare metal guide
confirm that your machine has an IPv6 name server (e.g. "cat /etc/resolv.conf" shows one or more IPv6 name servers on your node)
check on the node where the tectonics console has been deployed for constantly restarting nginx + tectonic-console containers

Other Information

A workaround seems to be to just turn off IPv6 DNS or wait until Tectonics ships with an updated nginx ingress controller, since the issue seems to already have been fixed upstream.

MarkusTeufelberger commented 7 years ago

Just to clarify: I fixed this issue by turning off the IPv6 DNS server manually and rebooting the nodes, they are running fine now. Still something that others might run into and that should be kept in mind for the next version (which hopefully already contains this fix).