Kubernetes

kubectl

get vs. describe

The get command prints a table of the most important information about specified resources.

$ kubectl get --help
...
$ kubectl get componentsstatuses
$ kubectl get nodes
$ kubectl get -o wide nodes
$ kubectl get -o json nodes

The describe command shows details of specified resources with a name prefix.

$ kubectl describe --help
$ kubectl describe nodes <NAME_PREFIX>

namespace

Kubernetes uses namespaces to organize objects in the cluster.

$ kubectl get namespaces
NAME          STATUS    AGE
default       Active    8d
kube-public   Active    8d
kube-system   Active    8d

You need to specify the --namespace flag to set the difference namespace temporarily.

$ kubectl get deployments --namespace=kube-system
NAME                    DESIRED   CURRENT   UP-TO-DATE   AVAILABLE   AGE
event-exporter-v0.2.3   1         1         1            1           8d
fluentd-gcp-scaler      1         1         1            1           8d
heapster-v1.5.3         1         1         1            1           8d
kube-dns                2         2         2            2           8d
kube-dns-autoscaler     1         1         1            1           8d
l7-default-backend      1         1         1            1           8d
metrics-server-v0.2.1   1         1         1            1           8d

To save the namespace permanently, set a context with the ---namespace flag.

$ kubectl config set-context $(kube config current-context) --namespace=my-resources
$ kubectl config view | grep namespace

run vs. apply

The run command creates a deployment or job to manage the created containers.

# The run command is deprecated.
$ kubectl run <DEPLOYMENT_NAME> --image=<IMAGE>
kubectl run --generator=deployment/apps.v1beta1 is DEPRECATED and will be removed in a future version. Use kubectl create instead.

# Use the create command instead
$ kubectl create deployment <DEPLOYMENT_NAME> --image=<IMAGE>

If you want to just create a pod, use the apply command with the Pod configuration.

$ cat my-pod.yml
apiVersion: v1
kind: Pod
metadata:
  ...
spec:
  containers:
    - image: <IMAGE>
      name: <POD_NAME>
      ...

$ kubectl apply -f my-pod.yml

logs

$ kubectl logs <POD_NAME>

exec / cp

$ kubectl exec -it <POD_NAME> -- bash

$ kubectl cp <POD_NAME>:/path/to/pod/file /path/to/local/file

delete

$ kubectl delete service <SERVICE_NAME>
$ kubectl delete deployment <DEPLOYMENT_NAME>

port-forward

The port-forward command forwards local ports to the specified pod or a selected pod by the deployment.

$ kubectl port-forward --help
...
Usage:
  kubectl port-forward TYPE/NAME [LOCAL_PORT:]REMOTE_PORT [...[LOCAL_PORT_N:]REMOTE_PORT_N] [options]
...

# TYPE defaults to pod if omitted.
$ kubectl port-forward <POD_NAME> <LOCAL_PORT>:<REMOTE_PORT>
$ kubectl port-forward deployment/<DEPLOYMENT_NAME> <LOCAL_PORT>:<REMOTE_PORT>

proxy

The proxy command launches a proxy server between localhost and the Kubernetes API server.

$ kubectl proxy --port=8001
Starting to serve on 127.0.0.1:8001
...

$ curl http://localhost:8001/api/v1/nodes
...

Google Kubernetes Engine

When you authorize your gcloud account, the current project will be updated.

$ gcloud auth login
...
$ gcloud config get-value project
<PROJECT_ID>

$ gcloud config set compute/zone asia-northeast1-a

How to manage clusters:

$ gcloud container clusters create <CLUSTER_NAME> \
  --zone=asia-northeast1-a \
  --machinet-ype=n1-standard-1 \
  --num-nodes=1

$ gcloud container clusters resize <CLUSTER_NAME> --num-nodes=3
$ gcloud container clusters delete <CLUSTER_NAME>

How to update kubectl current-context:

$ gcloud container clusters get-credentials <CLUSTER_NAME>
...
$ kubectl config current-context
$ kubectl get nodes
...

How to reduce default add-on resources:

# Disable Stackdriver monitoring
$ gcloud container clusters update <CLUSTER_NAME> --monitoring-servce none
$ gcloud container clusters update <CLUSTER_NAME> --update-addons=HorizontalPodAutoscaling=DISABLED
$ kubectl --namespace=kube-system scale deployment metrics-server-v0.3.1 --replicas=0
# (The metrics-server version may vary depending on the GKE cluster version)

# Disable Stackdriver logging
$ gcloud container clusters update <CLUSTER_NAME> --logging-servce none

# Disalbe DNS autoscaling
$ kubectl scale --replicas=0 deployment/kube-dns-autoscaler --namespace=kube-system
$ kubectl scale --replicas=1 deployment/kube-dns --namespace=kube-system