improve exit message when trying to pull Non-Existing Kubernetes Version

[medyagh]

I made a mistake and put in the wrong k8s version, we should have a more beautiful looking error, and ideally do not even try to create container when a non-existing k8s version is specified

mk start --kubernetes-version=20.10.7
😄  minikube v1.21.0-beta.0 on Darwin 11.4
✨  Automatically selected the docker driver. Other choices: hyperkit, parallels, virtualbox, ssh
👍  Starting control plane node minikube in cluster minikube
🚜  Pulling base image ...
🔥  Creating docker container (CPUs=2, Memory=4000MB) ...\ E0610 16:43:42.794488   82266 cache.go:211] Error caching images:  cached images list: version too new: 20.10.7

🐳  Preparing Kubernetes v20.10.7 on Docker 20.10.7 ...

❌  Exiting due to K8S_INSTALL_FAILED: kubeadm images: version too new: 20.10.7

╭────────────────────────────────────────────────────────────────────╮
│                                                                    │
│    😿  If the above advice does not help, please let us know:      │
│    👉  https://github.com/kubernetes/minikube/issues/new/choose    │
│                                                                    │
│    Please attach the following file to the GitHub issue:           │
│    - /Users/medya/.minikube/logs/lastStart.txt                     │
│                                                                    │
╰────────────────────────────────────────────────────────────────────╯

[JacekDuszenko]

I'd like to try and implement this

[JacekDuszenko]

/assign

[JacekDuszenko]

Hey @medyagh, I've just raised a PR with a fix #11631. Please, take a look.

[JacekDuszenko]

/assign @medyagh

[afbjorklund]

I think I just added some assertions to the backend, to avoid trying to run with strange numbers. So it needs to start with 1.
Whenever Kubernetes 2.0 comes out, this would have to be revisited. But it should allow a couple of future releases yet ?

        if v.Major > 1 {
                return nil, fmt.Errorf("version too new: %v", v)
        }
        if semver.MustParseRange("<1.12.0-alpha.0")(v) {
                return nil, fmt.Errorf("version too old: %v", v)
        }

Kubeadm will just make something up, and hope for the best:

$ kubeadm config images list --kubernetes-version=20.10.7
W0613 20:44:50.874013  628468 images.go:80] could not find officially supported version of etcd for Kubernetes v20.10.7, falling back to the nearest etcd version (3.2.24)
k8s.gcr.io/kube-apiserver:v20.10.7
k8s.gcr.io/kube-controller-manager:v20.10.7
k8s.gcr.io/kube-scheduler:v20.10.7
k8s.gcr.io/kube-proxy:v20.10.7
k8s.gcr.io/pause:3.4.1
k8s.gcr.io/etcd:3.2.24
k8s.gcr.io/coredns/coredns:v1.8.0

[afbjorklund]

The error message for old releases is slightly more friendly:

❗  Specified Kubernetes version 1.10.0 is less than the oldest supported version: v1.14.0
❗  You can force an unsupported Kubernetes version via the --force flag

❌  Exiting due to K8S_OLD_UNSUPPORTED: Kubernetes 1.10.0 is not supported by this release of minikube

The situation is not so good when trying a far future release:

✨  Automatically selected the docker driver. Other choices: kvm2, podman, virtualbox, ssh
👍  Starting control plane node minikube in cluster minikube
🚜  Pulling base image ...
🔥  Creating docker container (CPUs=2, Memory=7900MB) ...
❗  The image 'k8s.gcr.io/kube-scheduler:v1.25.0' was not found; unable to add it to cache.
❗  The image 'k8s.gcr.io/kube-apiserver:v1.25.0' was not found; unable to add it to cache.
❗  The image 'k8s.gcr.io/kube-controller-manager:v1.25.0' was not found; unable to add it to cache.
❗  The image 'k8s.gcr.io/kube-proxy:v1.25.0' was not found; unable to add it to cache.
🐳  Preparing Kubernetes v1.25.0 on Docker 20.10.7 ...
❌  Unable to load cached images: loading cached images: stat /home/anders/.minikube/cache/images/k8s.gcr.io/kube-scheduler_v1.25.0: no such file or directory

❌  Exiting due to K8S_INSTALL_FAILED: updating control plane: downloading binaries: downloading kubelet: download failed: https://storage.googleapis.com/kubernetes-release/release/v1.25.0/bin/linux/amd64/kubelet?checksum=file:https://storage.googleapis.com/kubernetes-release/release/v1.25.0/bin/linux/amd64/kubelet.sha256: getter: &{Ctx:context.Background Src:https://storage.googleapis.com/kubernetes-release/release/v1.25.0/bin/linux/amd64/kubelet?checksum=file:https://storage.googleapis.com/kubernetes-release/release/v1.25.0/bin/linux/amd64/kubelet.sha256 Dst:/home/anders/.minikube/cache/linux/v1.25.0/kubelet.download Pwd: Mode:2 Umask:---------- Detectors:[0x315d138 0x315d138 0x315d138 0x315d138 0x315d138 0x315d138 0x315d138] Decompressors:map[bz2:0x315d138 gz:0x315d138 tar.bz2:0x315d138 tar.gz:0x315d138 tar.xz:0x315d138 tar.zst:0x315d138 tbz2:0x315d138 tgz:0x315d138 txz:0x315d138 tzst:0x315d138 xz:0x315d138 zip:0x315d138 zst:0x315d138] Getters:map[file:0xc000663d10 http:0xc000c38f20 https:0xc000c38f40] Dir:false ProgressListener:0x3120890 Options:[0x190e6a0]}: invalid checksum: Error downloading checksum file: bad response code: 404

[afbjorklund]

Might have to add a new constant for this, the newest version that is supported with some similar version skew ?

    DefaultKubernetesVersion = "v1.20.7"
    NewestKubernetesVersion = "v1.22.0-alpha.2"
    OldestKubernetesVersion = "v1.14.0"

If assuming the same 6 versions (v1.20.x -> v1.14.0), then it would be a warning with anything above v1.26.0 ?

But I just hardcoded >= v2.0.0 as "too new", for now.

Parsing available versions etc is a different feature ?

[JacekDuszenko]

Thank you for your input @afbjorklund . As I've already implemented what @medyagh suggested (fetching and parsing available versions), let's see what he thinks and we'll go from there. Also, feel free to look at my updated PR if you want. Thanks!

[afbjorklund]

Hardcoding future versions was not great idea anyway. In theory, minikube should continue to work also with those ?

[afbjorklund]

PR improves the failure scenario:
❌ Exiting due to K8S_VERSION_NOT_FOUND: Specified Kubernetes version 1.25.0 was not found in Kubernetes releases

Just need to make it possible to run offline too, where needed.
❌ Exiting due to K8S_RELEASE_FETCH_FAILED: Failed fetching available Kubernetes releases from GitHub repository. Check your internet connection

[nishipy]

I created a PR (#13354) to add simple solution for this issue. Please take a look. Thank you!
/assign