feat: Fine-grained stack update

[mumoshu]

USAGE:

Update all the stacks including root, control-plane, worker:

kube-aws update --force

Update the control-plane stack only:

kube-aws update --targets control-plane --force

Note that, for the root stack, just the Resources.Controlplane.Properties.TemplateURL is updated in this case.

Update the worker node pool stack(s) only:

kube-aws update --targets worker --force

NOTES:

• The root stack is updated implicitly when you update all the stacks including worker, control-plane, etcd and network

TODO:

• The separation and extraction of the etcd stack and the network stack from the control-plane stack
• Add a new option in cluster.yaml rollbackOnFailure for control-plane and etcd stack only

Everything else should be there

Ref #1112

[mumoshu]

This is where the magic happens.

[codecov-io]

Codecov Report

Merging #1233 into master will decrease coverage by 0.39%.
The diff coverage is 29.88%.

@@            Coverage Diff            @@
##           master    #1233     +/-   ##
=========================================
- Coverage   36.63%   36.23%   -0.4%     
=========================================
  Files          63       64      +1     
  Lines        3882     3913     +31     
=========================================
- Hits         1422     1418      -4     
- Misses       2242     2279     +37     
+ Partials      218      216      -2

Impacted Files	Coverage Δ
core/controlplane/config/stack_config.go	0% <ø> (ø)	⬆️
model/subnet.go	1.39% <0%> (-0.06%)	⬇️
model/vpc.go	0% <0%> (ø)
model/etcd.go	0% <0%> (ø)	⬆️
model/derived/api_endpoint_lb.go	0% <0%> (ø)	⬆️
model/controller.go	0% <0%> (ø)	⬆️
cfnstack/assets.go	0% <0%> (ø)	⬆️
model/derived/etcd_node.go	0% <0%> (ø)	⬆️
model/subnets.go	23.33% <0%> (-76.67%)	⬇️
cfnstack/provisioner.go	0% <0%> (ø)	⬆️
... and 5 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update a55c665...1133d20. Read the comment docs.

[mumoshu]

This is triggering a considerable amount of refactorings... 😉

[mumoshu]

Whoa. Can't believe this is actually working 🤣

$ KUBECONFIG=kubeconfig kubectl get no
NAME                                            STATUS    ROLES     AGE       VERSION
ip-10-0-0-180.ap-northeast-1.compute.internal   Ready     node      7m        v1.9.1
ip-10-0-0-20.ap-northeast-1.compute.internal    Ready     master    15m       v1.9.1

[mumoshu]

You now have to enumerate node pools explicitly like kube-aws update --targets nodepool1,nodepool2 in order to update a part of node pools only. Previously, kube-aws update --targets worker had been updating all the node pools but it was removed in favor of this.

Also note that --targets is enhanced to accept all, so that you can target all the stacks with --targets all, rather than --targets worker,controller,etcd.

[mumoshu]

Go ahead!

[cknowles]

Is this intended to work with just --targets etcd? I did an update like that but it also updated controllers and started on workers too. The controllers appeared to be mostly blank container linux instances, they didn't even have a kubelet service! To fix it I cancelled the update and the cluster has a functioning control plane again.