Add timeout value to configure how long kubeadm init takes to timeout

[soggiest]

FEATURE REQUEST

Add a KubeADM configuration option or CLI flag that determines how long kubeadm init takes to timeout. Cloud-based installations have a chance to fail if the API Server's load balancer takes too long to recognize an instance is listening on the appropriate port. This can potentially be mitigated by modifying the load balancer's health checks. However, an option to extend the kubeadm init would be helpful.

Versions

kubeadm version (use kubeadm version): 1.11.3

Environment:

• Kubernetes version (use kubectl version): 1.11.3
• Cloud provider or hardware configuration: AWS
• OS (e.g. from /etc/os-release): Ubuntu 16.04
• Kernel (e.g. uname -a): kubeadm does not copy the cloud configuration file to nodes #77-Ubuntu SMP

What happened?

KubeADM init runs were consistently failing on AWS with the following error: "timed out waiting for condition"
After each KubeADM run the Kubernetes control plane was working as expected. During an investigation into what was causing this error I observed that the AWS ELB would become active after KubeADM had timed out. Modifying the ELB health checks to be quicker alleviated the issue.

From a UX stand point I should be able to set the init timeout from a config or cli flag rather than worry about a race against time with my ELB.

What you expected to happen?

KubeADM init would wait long enough for the control plane to come up.

How to reproduce it (as minimally and precisely as possible)?

Region:
ap-southeast-1 (Singapore)

Configure an AWS EC2 Instance with:
AMI: Ubuntu 16.04 LTS
Size: m5.xlarge
Storage: 50GB
Tag: kubernetes.io/cluster/kubernetes: owned

Configure an ELB with:
Standard Health Checks Intervals, Health check against TCP:6443
TCP Passthrough 443 -> TCP 6443

Run kubeadm init with a kubeadm config that points to the ELB as the controlPlaneEndpoint

[neolit123]

From a UX stand point I should be able to set the init timeout from a config or cli flag rather than worry about a race against time with my ELB.

i agree with this.

/kind feature
/cc @kubernetes/sig-cluster-lifecycle

[fabriziopandini]

@timothysc opinions?

[dixudx]

I'd love to add this. +1.

[timothysc]

This has been requested a number of times, but the use case was on the other end of the spectrum.

@fabriziopandini could we roll this into the v1beta1 component(apiserver) migration work?

[timothysc]

/assign @soggiest
/assign @fabriziopandini
/assign @timothysc

[fabriziopandini]

@timothysc ok
Just to set expectations: I'm going to propose a generic solution for defining timeouts in the kubeadm config API (not in flags) and use the new settings for making configurable only the timeout above + eventually the same timeout in the kubeadm --join-workflow (not all the timeouts in kubeadm).

[neolit123]

xref kubernetes/kubernetes#70480

[neolit123]

timeout for the api server was added in the v1beta1 config:
https://github.com/kubernetes/kubernetes/blob/master/cmd/kubeadm/app/apis/kubeadm/v1beta1/types.go#L135