[CMR] CoreDNS goes to 'Unknown' state

[Peter Jose De Sousa]

Hi,

When deploying coredns charm revision 18, offering, consuming and relating CoreDNS the SAAS endpoint on the foreign model goes to "Unknown" and DNS is not accessible on Kubernetes cluster.

This is repeated on different clouds. Steps to reproduce are:

Prerequisites:

1. A charmed k8s cluster
2. a bootstrapped juju controller on k8s

Steps:

FOREIGN_MODEL=MYCONTROLLER:MYMODEL

1. Add CoreDNS model
2. juju deploy cs:~containers/coredns
3. juju offer coredns:dns-provider
4. juju consume -m $FOREIGN_MODEL coredns-k8s:admin/kube-system-coredns.coredns
5. juju relate -m $FOREIGN_MODEL kubernetes-master coredns

juju --version
2.9.17-ubuntu-amd64

crashdump attached, screendumps in private-fileshare

[WORKAROUND]

Add the k8s cloud to the underlying cloud controller. This should send the application into "Active" state and allow you run enable coredns successfully.

thank you

Peter

[Peter Jose De Sousa (pjds)]

screendump: https://private-fileshare.canonical.com/~pjds/lp-1950139/screendumps.txt

[Peter Jose De Sousa (pjds)]

checking debug-log theres a suspicious "error in remote application worker for coredns"

controller-0: 13:17:34 ERROR juju.worker.remoterelations error in remote application worker for coredns: opening facade to remote model: cannot resolve "controller-service.controller-coredns-k8s.svc.cluster.local": lookup controller-service.controller-coredns-k8s.svc.cluster.local on 127.0.0.53:53: server misbehaving

[Peter Jose De Sousa (pjds)]

Subscribing field medium, as this is blocking upgrade of coredns, but does not mean a degraging of functionality yet.

[Peter Jose De Sousa (pjds)]

After further digging it was discovered that adding the k8s-cloud to the existing VM controller resolved the situation. Updating the workaround.