Skip to content

Latest commit

 

History

History
130 lines (93 loc) · 6.13 KB

kubernetes-basics.md

File metadata and controls

130 lines (93 loc) · 6.13 KB
Raw

Kubernetes basics

The following collection of sub-sections is intended to provide enough information about kubernetes to get you started. Obviously, to become proficient with kubernetes, you will need to gain experience through use and you will need to consult more complete sources of information, including the kubernetes documentation.

Installing kubectl on the ICP boot-master node

Copying kubectl from a container to the host /usr/local/bin It is very useful to be able to have kubectl available at the shell on the ICP boot-master or boot machine.

You can get kubectl from the kubernetes container already installed as part of ICP. (NOTE: In the command sample below you may need to modify the version tag on kubernetes. Use docker images | grep kubernetes to see your actual version tag for kubernetes.)

> docker run --net=host -v /usr/local/bin:/data ibmcom/kubernetes:v1.8.3 cp /kubectl /data

If the kubernetes image is not available in the local registry on the boot node, then another place that kubectl is available is the inception image.

NOTE: For ICP 2.1.0.3 the following command can be used to get a kubectl excutuable copied to the VM host:

docker run --net=host -e LICENSE=accept -v /usr/local/bin:/data ibmcom/icp-inception:2.1.0.3-ee cp /kubectl /data

Again, the exact name of the image may not be as shown in the command above. Check your docker images to determine the image name.

You may also have to open a shell on the image to figure out where the kubectl executable is in the image.

docker run -it -e LICENSE=accept ibmcom/icp-inception-amd64:3.1.0-ee /bin/bash

NOTE: For ICP 3.1.0 the following can be used to get a kubectl executable copied to the VM host:

docker run -it -e LICENSE=accept -v /usr/local/bin:/data ibmcom/icp-inception-amd64:3.1.0-ee cp /usr/local/bin/kubectl /data

Kubernetes command basics

The command line tool for working with Kubernetes is kubectl.

Kubernetes Getting Started

A quick source of kubectl help: Kubernetes kubectl "cheat sheet"

When working with kubectl you need to be authenticated in your shell. Authentication involves a number of steps.

To set up authentication in your shell:

  1. Log into the ICP console. The client setup command stream is available under the user name in the upper right margin of the ICP console.
  2. Click on the user name, then click on "Configure client" and in the pop-up window, click on the copy icon next to the "kubectl config" commands in the pop-up window.
  3. Do a paste in the shell window you are using. At that point, the shell is configured to properly use kubectl commands.

The authentication token is good for 12 hours. You have to log out of the ICP console and log back in to get a new token using the same procedure described in the steps above.

You can see your kubectl context using the kubectl config view command.

Get a list of namespaces

kubectl get namespaces

Get a list of pods

kubectl get pods --all-namespaces

NOTE: A newly deployed ICP cluster typically has pods defined only in the kube-system namespace.

You can limit the pod listing to a specific namespace with the --namespace=NAMESPACE_NAME, e.g.,

kubectl get pods --namesapce=kube-system

NOTE: If you want to see the terminated pods, i.e., those that have completed or errored out, then use the -a (--show-all) option with the get command.

Shorthand for --namespace is -n, e.g., -n kube-system

Get info about a pod

kubectl describe pod <pod_name>

The describe command is particularly useful on any of the different kinds of Kubernetes objects.

Get a list of deployments

kubectl get deployments

Get a log associated with a pod or container

kubectl logs pod/<pod_name> | tee -a mypod.log

or

kubectl logs pod/<pod_name> --container=<container_name> | tee -a mycontainer.log

If a pod has more than one container in it, then you need the --container (-c) option.

Check out the usage information for the logs command (kubectl logs --help) for details on all the options.

Combining kubectl commands

You can use all the usual Linux idioms for combining commands with kubectl.

Here is an example of deleting a bunch of pods with bluecompute-ce in the pod name:

kubectl delete pods $(kubectl get pods -a | grep bluecompute-ce | awk '{print $1}')

Change the current namespace preference

Typically the namespace associated with a given kubeconfig will be the default namespace. However you can change it to some other namespace.

The following command changes the currently preferred namespace to kube-system. By doing so the --namespace option for kubectl commands is not needed if the namespace if interest is kube-system.

kubectl config set-context $(kubectl config current-context) --namespace kube-system

For more detailed information on Kubernetes namespaces, see the Kubernetes Namespaces documentation.

kubectl without logging in

There may be times when you can get the "client config" command line content from the ICP console, but you still want to be able to use kubectl. In those cases get on the one of the master nodes (with kubectl installed) and use -s localhost:8888 as one of the command line options to kubectl followed by the command you want to run.

For example:

> kubectl -s localhost:8888 --namespace=kube-system get pods

NOTE: The use of the -s localhost:8888 option has been disabled. It can be enabled by manually configuring the service to listen on that port.

One approach for using kubectl when a console login is not available is the following command (run on a master node)

kubectl --kubeconfig=/var/lib/kubelet/kubectl-config get pods -n kube-system -o wide

At least as of ICP 3.1.1, the root user on the master nodes has the ~/.kube/config file set up with a permanent login context for kubectl. (To become root on the master once you log in be sure to include the dash in sudo su -.)