Documenting Kubernetes getting started (personal background notes)

doc
1

This document is a bit of a personal brain dump. It won’t become official documentation.

Thoughts

Need to sync/update the semantics to match add-credentials.

The add-k8s command introduces new flows for add-cloud and add-credential. This introduces cognitive load.

For documentation purposes, we now need to consider four starting points:

Has a Kubernetes cluster? Familiar with Juju? Live controller? In Scope? Representative User(s)
:ballot_box_with_check: :ballot_box_with_check: :ballot_box_with_check: :ballot_box_with_check: Canonical IS staff. Heavily invested in Juju.
:ballot_box_with_check: :ballot_box_with_check: :ballot_box_with_check: Ex-Juju user.
:ballot_box_with_check: :ballot_box_with_check: :sparkle: Current k8s user. Exploring options for simplifying their deployment.
:ballot_box_with_check: :ballot_box_with_check: :ballot_box_with_check: Current Juju user. Someone currently on Discourse.
:ballot_box_with_check: :ballot_box_with_check: Someone who used Juju in the past, but has moved to other tools. JAAS users.
:negative_squared_cross_mark: Brand new to Juju and Kubernetes.

Examples

juju add-k8s myk8scloud
Behaviour
Loads the definition of myk8scloud into the current controller.
Assumptions
Assumes that a K8s cluster is available.

When the k8s cluster is a vanilla cluster on top of a cloud, assumes that an integrator charm has been deployed.

Questions
  • Q: where does the definition for myk8scloud come from? ~/.kube/config? A: yes.
  • how does it behave when k8s isn't available? A: `ERROR failed to read Kubernetes config file: /home/tsm/.kube/config not found`
  • how does it behave when a controller isn't available?

Scenarios

No k8s, no controller
$ juju add-k8s nope
ERROR failed to read Kubernetes config file: /home/ubuntu/.kube/config not found
Kubernetes-core deployed
Setup 
juju add-cloud devstack [openstack]
juju bootstrap devstack
juju add-model k8s-base
juju deploy kubernetes-core
juju wait [via juju-wait plugin]
juju scp kubernetes-master/0:config ~/.kube/config [only documented in charm readmes]
Running add-k8s 
$ juju add-k8s dev-k8s
.
ERROR 	Juju needs to query the k8s cluster to ensure that the recommended
	storage defaults are available and to detect the cluster's cloud/region.
	This was not possible in this case so run add-k8s again, using
	--storage= to specify the storage class to use and
	--cloud= --region=/ to specify the cloud/region.

$ juju add-k8s dev-k8s --cloud=devstack
.
ERROR 	No recommended storage configuration is defined on this cluster.
	Run add-k8s again with --storage= and Juju will use the
	specified storage class or create a storage-class using the recommended
	"Cinder Disk" provisioner.

juju deploy ~containers/openstack-integrator
juju trust openstac-integrator
juju wait
Questions:
  • Why is <cloud> required twice?
    • Microk8s, no controller
    $ juju add-k8s maybe
[todo]

    juju add-k8s myk8scloud --local
juju add-k8s myk8scloud --controller mycontroller
juju add-k8s --context-name mycontext myk8scloud
juju add-k8s myk8scloud --region <cloudNameOrCloudType>/<someregion>
juju add-k8s myk8scloud --cloud <cloudNameOrCloudType>
juju add-k8s myk8scloud --cloud <cloudNameOrCloudType> --region=<someregion>

KUBECONFIG=path-to-kubuconfig-file juju add-k8s myk8scloud --cluster-name=my_cluster_name
kubectl config view --raw | juju add-k8s myk8scloud --cluster-name=my_cluster_name

juju add-k8s --gke myk8scloud
juju add-k8s --gke --project=myproject myk8scloud
juju add-k8s --gke --credential=myaccount --project=myproject myk8scloud
juju add-k8s --gke --credential=myaccount --project=myproject --region=someregion myk8scloud

juju add-k8s --aks myk8scloud
juju add-k8s --aks --cluster-name mycluster myk8scloud
juju add-k8s --aks --cluster-name mycluster --resource-group myrg myk8scloud

    See also:
    remove-k8s