driver=none sets --extra-config=kubelet.resolv-conf for incorrect path in Ubuntu 16.04

[anencore94]

Steps to reproduce the issue:

1. minikube --vm-driver=none start on Ubuntu 16.04
2. then coredns pod's status become CrashLoopBackOff as in none on Ubuntu should automatically set --extra-config=kubelet.resolv-conf #3511

Full output of failed command:

Since there isn't any file /run/systemd/resolve/resolv.conf in Ubuntu 16.04, The auto-setting --extra-config=kubelet.resolv-conf=/run/systemd/resolve/resolv.conf doesn't work.

• https://github.com/kubernetes/minikube/blob/master/pkg/minikube/driver/driver.go#L207

# ls /run/systemd/resolve/resolv.conf
ls: cannot access '/run/systemd/resolve/resolv.conf': No such file or directory

# ls -l /etc/resolv.conf
lrwxrwxrwx 1 root root 29  8월 12  2019 /etc/resolv.conf -> ../run/resolvconf/resolv.conf

So different from minikube start on Ubuntu 18.04, the stdout of minikube start on Ubuntu 16.04 doesn't have following line:
▪ kubelet.resolv-conf=/run/systemd/resolve/resolv.conf

How to fix it in Ubuntu 16.04:
By https://stackoverflow.com/questions/53075796/coredns-pods-have-crashloopbackoff-or-error-state/53414041#53414041

Following solution works also on Ubuntu 16.04:
Hacky solution: Disable the CoreDNS loop detection

Edit the CoreDNS configmap:

kubectl -n kube-system edit configmap coredns
Remove or comment out the line with loop, save and exit.

Then remove the CoreDNS pods, so new ones can be created with new config:

kubectl -n kube-system delete pod -l k8s-app=kube-dns

Other trial in Ubuntu 16.04:.
I've tried minikube start with --extra-config=kubelet.resolv-conf=/run/resolvconf/resolv.conf option, But it doesn't work as I expected. The coredns pod also become CrashLoopBackOff. I have no idea why this happens.

[tstromberg]

I didn't think this flag was required for Ubuntu 16.04. What happens if you run --extra-config=kubelet.resolv-conf=/etc/resolv.conf?

[anencore94]

@tstromberg The same issue happen with --extra-config=kubelet.resolv-conf=/etc/resolv.conf.

coredns pod became crashloopbackoff by loop plugin

[xieydd]

same error

[medyagh]

@xieydd is your error also on ubuntu 16.04 ?

@anencore94 does this happen only on ubuntu 16.04 and not ubuntu 18.04 or newer? (ubuntu 16.04 end of life is less than a year in april 2021)

[medyagh]

@anencore94 I am curious is there a reason you are not using our Docker driver instead of none ?
none is not our preferred driver
/triage needs-information
/triage support

[anencore94]

@medyagh Yeap It only happens on 16.04.
Since the path was right on 18.04

I've been using nvdia-docker to use gpu with minikube. With your official docs, only kvm2 and none driver supports gpu. Also, none driver setting was easier than kvm2 driver, so i choose to use none driver.

[xieydd]

@medyagh Yep, i use ubuntu 16.04, and i have same purpose as @anencore94 , use minikube and gpu-device-plugin.

[priyawadhwa]

Hey @anencore94 I just wanted to confirm that you ran --extra-config on a brand new cluster? That flag doesn't work on pre-existing clusters (that issue can be tracked here #8661)

To make sure it worked, could you run:

minikube delete
minikube start --extra-config=kubelet.resolv-conf=/etc/resolv.conf

and let us know if that fixes the issue?

[anencore94]

@priyawadhwa Thanks for checking. But I did this on a brand-new cluster.
I think it is not the same reason with #8661.

Since with log, I can see minikube start --extra-config=kubelet.resolv-conf=/etc/resolv.conf sets such extra-config successfully. But CrashLoopBackOff happens.

[medyagh]

@anencore94 do you still have this issue ?

do you mind sharing the logs for the crash that you see ?

btw KVM driver should not be too hard to install.

[anencore94]

@medyagh Sry.. I just changed my os to Ubuntu 20.04 now.
When I have some chance to work on Ubuntu 16.04, I'll test again and share log here.
Thanks.

[fejta-bot]

Issues go stale after 90d of inactivity.
Mark the issue as fresh with /remove-lifecycle stale.
Stale issues rot after an additional 30d of inactivity and eventually close.

If this issue is safe to close now please do so with /close.

Send feedback to sig-testing, kubernetes/test-infra and/or fejta.
/lifecycle stale