Move to Kubernetes 1.11 #77
Conversation
dlipovetsky
changed the title
dlipovetsky
force-pushed
the
private/daniel/kubernetes-1.11
branch
from
54bfbb1 to
10bd01d
Compare
dlipovetsky
changed the title
dlipovetsky
force-pushed
the
private/daniel/kubernetes-1.11
branch
from
10bd01d to
f52daee
Compare
dlipovetsky
force-pushed
the
private/daniel/kubernetes-1.11
branch
from
f52daee to
6ff55eb
Compare
dlipovetsky
changed the title
dlipovetsky
force-pushed
the
private/daniel/kubernetes-1.11
branch
from
f4db4e7 to
6a3fadc
Compare
Do not silently ignore errors
dlipovetsky
force-pushed
the
private/daniel/kubernetes-1.11
branch
from
5d53b2d to
9bc7ffe
Compare
- Accept configuration file in nodeadm join - Use kubeadm config to override hostname - Do not write nodeadm kubelet systemd drop-in; no longer necessary, as values are fetched by kubeadm. See https://kubernetes.io/docs/setup/independent/kubelet-integration/#propagating-cluster-level-configuration-to-each-kubelet - Remove unused Kubelet and Networking fields from JoinConfiguration - Remove default IP and routerID for VIPConfiguration
CoreDNS is the default cluster DNS provider as of kubeadm 1.11
kubeadm 1.11 expects the arch-agnostic image. See kubernetes/kubernetes@0745b25
dlipovetsky
force-pushed
the
private/daniel/kubernetes-1.11
branch
from
e20c719 to
c0d8d88
Compare
Co-Authored-By: dlipovetsky <dlipovetsky@users.noreply.github.com>
| } | ||
| if netConfig.DNSDomain == "" { | ||
| netConfig.DNSDomain = constants.DefaultDNSDomain | ||
| if p.ExistsP("api.bindPort") { |
There was a problem hiding this comment.
Just adding a comment as per our discussion during code walk through. There is a tradeoff between knowing the type or using master configuration as just a blob of map[string]{}. Not knowing the type reemphasizes (and rightly so) the fact that nodeadm should just act as pass through of kubeadm's master configuration. However there might be cases like these where nodeadm would still have to be aware of partial if not entire schema of upstream type.
There was a problem hiding this comment.
This is the only case, and it's due to nodeadm configuring keepalived. I'd like to remove even this case in the future, if possible.
| func logAllWithPrefix(prefix string, r io.Reader) { | ||
| s := bufio.NewScanner(r) | ||
| for s.Scan() { | ||
| log.Infof("%v: %v", prefix, s.Text()) |
There was a problem hiding this comment.
Hi Daniel, just to understand, this means that even with the piped output the verbosity set at nodeadm will be honored?
There was a problem hiding this comment.
Yes 😄
If nodeadm is configured to omit Info-level messages, kubeadm output won't be seen.
| DefaultPodNetwork = "10.244.0.0/16" | ||
| DefaultDNSIP = "10.96.0.10" | ||
| DefaultServiceSubnet = "10.96.0.0/12" | ||
| DefaultDNSDomain = "cluster.local" | ||
| DefaultRouterID = 42 | ||
| KubeadmConfig = "/tmp/kubeadm.yaml" | ||
| KubeDNSVersion = "1.14.8" | ||
| CoreDNSVersion = "1.1.3" |
There was a problem hiding this comment.
Hi Daniel, sorry forgot to ask during code walk through, for upgrade scenarios do we need an explicit kubectl delete for kubedns
There was a problem hiding this comment.
Yes, but that will be done in cctl.
Fixes #76