Kubernetes Scripts

A collection of scripts and commands for various tasks in Kubernetes.
These were all written during my work with Kubernetes on various project. Enjoy and share. Contributions are more than welcome!

Usage

Each script has a usage function. See usage with

<script> --help

Scripts

• countPodsAndContainerPerNodeCSV.sh: Count number of pods and containers per node. Print in CSV format.
• findEmptyNamespaces.sh: Loop over all namespaces in a cluster and find empty ones.
• getPodsLoad.sh: Get formatted results of pods in a namespace underlying node's load average (using cat /proc/loadavg).
• getPodsTopCSV.sh: Get a pod's cpu and memory usage (optionally per container) written as CSV formatted file.
• getResourcesCSV.sh: Get all pods resources requests, limits and actual usage per container in a CSV format with values normalized.
• getRestartingPods.sh: Get all pods (all or single namespace) that have restarts detected in one or more containers. Formatted in CSV.
• podReady.sh: Simple script to check if pod is really ready. Check status is 'Running' and that all containers are ready. Returns 0 if ready. Returns 1 if not ready.
• getNodesLoadCSV.sh: Traverse over the kube-proxy pods to get the nodes load average and number of CPUs in a CSV format. Will also mark high load node with big YES in the output.
• runCommandOnPods.sh: Run a command on a list of pods.
• canIdo.sh: Check all or some permissions current user has in a namespace on all or some resources using the kubectl auth can-i command.

YAML

• memory.yaml: A pod using a given block of memory for a given time.
  
• podWithTools.yaml: A pod with some basic tools (vi and curl) for easy debugging.
  
• podWithHostFS.yaml: A pod with the host root file system mounted into it.
  
• podmanPod.yaml: A pod with podman in it.
  
• kind-config.yaml: An example kind configuration for a multi node K8s cluster WARNING: There is danger of corrupting your Kubernetes host. Use with extra care!

Commands

Kubectl

See all cluster nodes load (top)

kubectl top nodes

Get cluster events

# All cluster
kubectl get events

# Specific namespace events
kubectl get events --namespace=kube-system

Get all cluster nodes IPs and names

# Single call to K8s API
kubectl get nodes -o json | grep -A 12 addresses

# A loop for more flexibility
for n in $(kubectl get nodes -o name); do \
  echo -e "\nNode ${n}"; \
  kubectl get ${n} -o jsonpath='{.status.addresses}'; \
done

See all cluster nodes CPU and Memory requests and limits

# With node names
kubectl describe nodes | grep -A 3 "Name:\|Resource .*Requests .*Limits" | grep -v "Roles:"

# Just the resources
kubectl describe nodes | grep -A 3 "Resource .*Requests .*Limits"

Using kube-capacity

There is a great CLI for getting a cluster capacity and utilization - kube-capacity.
Install as described in the installation section.

# Get cluster current capacity
kube-capacity

# Get cluster current capacity with pods breakdown
kube-capacity --pods

# Get cluster current capacity and utilization
kube-capacity --util

# Displaying available resources
kube-capacity --available

# Roll over all clusters in your kubectl contexts
for a in $(kubectl ctx); do echo -e "\n---$a"; kubectl ctx $a; kube-capacity; done

# Roll over all clusters in your kubectl contexts and get just summary of each cluster
for a in $(kubectl ctx); do echo -e "\n---$a"; kubectl ctx $a; kube-capacity| grep -B 1 "\*"; done

Get all labels attached to all pods in a namespace

for a in $(kubectl get pods -n namespace1 -o name); do \
  echo -e "\nPod ${a}"; \
  kubectl -n namespace1 describe ${a} | awk '/Labels:/,/Annotations/' | sed '/Annotations/d'; \
done

Forward local port to a pod or service

# Forward localhost port 8080 to a specific pod exposing port 8080
kubectl port-forward -n namespace1 web 8080:8080

# Forward localhost port 8080 to a specific web service exposing port 80
kubectl port-forward -n namespace1 svc/web 8080:80

Port forwarding

• A great tool for port forwarding all services in a namespace + adding aliases to /etc/hosts is kubefwd. Note that this requires root or sudo to allow temporary editing of /etc/host.

# Port forward all service in namespace1
kubefwd svc -n namespace1

Extract and decode a secret's value

# Get the value of the postgresql password
kubectl get secret -n namespace1 my-postgresql -o jsonpath="{.data.postgres-password}" | base64 --decode

Copy secret from namespace1 to namespace2

kubectl get secret my-secret --namespace namespace1 -o yaml | sed "/namespace:/d" | kubectl apply --namespace=namespace2 -f -

Create an Ubuntu pod

A one liner to create an Ubuntu pod that will just wait forever.

# Create the pod
cat <<ZZZ | kubectl apply -f -
apiVersion: v1
kind: Pod
metadata:
  name: my-ubuntu-pod
spec:
  containers:
  - name: my-ubuntu-container
    image: eldada.jfrog.io/docker/ubuntu:22.04
    command:
    - 'bash'
    - '-c'
    - 'while true; do date; sleep 60; done'
ZZZ

# Shell into the pod
kubectl exec -it my-ubuntu-pod bash

# Delete the pod once done
kubectl delete pod my-ubuntu-pod

Start a shell in a temporary pod

Note - Pod will terminate once exited

# Ubuntu
kubectl run my-ubuntu --rm -i -t --restart=Never --image ubuntu -- bash

# CentOS
kubectl run my-centos --rm -i -t --restart=Never --image centos:8 -- bash

# Alpine
kubectl run my-alpine --rm -i -t --restart=Never --image alpine:3.10 -- sh

# Busybox
kubectl run my-busybox --rm -i -t --restart=Never --image busybox -- sh

Get formatted list of containers and container images

Useful for listing all running containers in your cluster

# Example 1 - just the container names
kubectl get pods -A -o jsonpath='{..containers[*].name}' | tr -s ' ' '\n'
# With sorting and unique names
kubectl get pods -A -o jsonpath='{..containers[*].name}' | tr -s ' ' '\n' | sort | uniq

# Example 2 - container images and tags
kubectl get pods -A -o=jsonpath='{..containers[*].image}' | tr -s ' ' '\n'
# With sorting and unique names
kubectl get pods -A -o=jsonpath='{..containers[*].image}' | tr -s ' ' '\n' | sort | uniq

# Example 3 - pod and its container images
kubectl get pods -A -o=jsonpath='{range .items[*]}{"\n"}{.metadata.name}{":\n"}{range .spec.containers[*]}{.name},{.image}{"\n"}{end}{end}'

# Example 4 - pod and its container images with their resources requests (cpu and memory)
kubectl get pods -A -o=jsonpath='{range .items[*]}{"\n"}{.metadata.name}{":\n"}{range .spec.containers[*]}{.name},{.image}{.resources.requests.cpu},{.resources.requests.memory}{"\n"}{end}{end}'

Look into a few more examples of listing containers

Get list of pods sorted by restart count

• Option 1 for all pods (Taken from kubectl cheatsheet)

kubectl get pods -A --sort-by='.status.containerStatuses[0].restartCount'

• Option 2 with a filter, and a CSV friendly output

kubectl get pods -A | grep my-app | awk '{print $5 ", " $1 ", " $6}'  | sort -n -r

Get current replica count on all HPAs (Horizontal Pod Autoscaling)

kubectl get hpa -A -o=custom-columns=NAME:.metadata.name,REPLICAS:.status.currentReplicas | sort -k2 -n -r

List non-running pods

kubectl get pods -A --no-headers | grep -v Running | grep -v Completed

Top Pods by CPU or memory usage

# Top 20 pods by highest CPU usage
kubectl top pods -A --sort-by=cpu | head -20

# Top 20 pods by highest memory usage
kubectl top pods -A --sort-by=memory | head -20

# Roll over all kubectl contexts and get top 20 CPU users
for a in $(kubectl ctx); do echo -e "\n---$a"; kubectl ctx $a; kubectl top pods -A --sort-by=cpu | head -20; done

Debugging Pods and Nodes

This section is based on debugging pods using ephemeral containers and kubectl node debug

Pod Debugging

# Attach an ephemeral container to an existing container in a pod for debugging
kubectl debug -it my-pod --image=ubuntu --target=my-container

Node Debugging

# Debug a node with a new pod attached to it
# IMPORTANT to delete the pods after exiting it. It will not be deleted automatically (although it will be in the "Completed" state)
kubectl debug node/<mynode> -it --image=ubuntu

Helm

Helm template

View the templates generated by helm install. Useful for seeing the actual templates generated by helm before deploying.
Can also be used for deploying the templates generated when cannot use Tiller

helm template <chart>

Debug helm install

• Debug a helm install. Useful for seeing the actual values resolved by helm before deploying

helm install --debug --dry-run <chart>

Rolling restarts

Roll a restart across all resources managed by a Deployment, DaemonSet or StatefulSet with zero downtime
IMPORTANT: For a Deployment or StatefulSet, a zero downtime is possible only if initial replica count is higher than 1!

# Deployment
kubectl -n <namespace> rollout restart deployment <deployment-name>

# DaemonSet
kubectl -n <namespace> rollout restart daemonset <daemonset-name>

# StatefulSet
kubectl -n <namespace> rollout restart statefulsets <statefulset-name>

Mark Nodes with some roles for visibility (ex. EKS nodes marked with the LifeCycle,NodeType)

• Most use of it can be gained with some GUI client (Lens), still "k get nodes" shows ROLE fields as well

for n in $(kubectl get nodes -o 'jsonpath={.items[*].metadata.name}') ; do
  lb=""
  for a in $(kubectl label --list nodes $n | sort | grep -e NodeType -e lifecycle | cut -d= -f 2); do
    lb="${lb}$a"
  done
  kubectl label nodes $n node-role.kubernetes.io/$lb=
done

A Multi Node Kubernetes cluster in Mac with Kind

To run a multi node Kubernetes cluster in Mac with Kind, do the following (assuming Docker Desktop is already installed)

• Install kind as described in kind installation
• Start a local, three worker nodes cluster using the kind-config.yaml config file

kind create cluster --config yaml/kind-config.yaml --name demo

Delete the cluster with

kind delete cluster --name demo

Metrics Server in Kubernetes on Docker Desktop or Kind for Mac

To get around issue with certificates in your local Docker Desktop or Kind Kubernetes

Install a metrics-server

kubectl apply -f https://github.com/kubernetes-sigs/metrics-server/releases/latest/download/components.yaml

Patch the metrics-server Deployment with the fix

kubectl patch deployment metrics-server -n kube-system --patch-file yaml/metrics-server-patch.yaml

OR Edit the metrics-server deployment directly and add --kubelet-insecure-tls to the args key:

spec:
  containers:
  - args:
    - --cert-dir=/tmp
    - --secure-port=443
    - --kubelet-preferred-address-types=InternalIP,ExternalIP,Hostname
    - --kubelet-use-node-status-port
    - --metric-resolution=15s
    - --kubelet-insecure-tls

Resources

Most of the code above is self experimenting and reading the docs. Some are copied and modified to my needs from other resources...

• https://kubernetes.io/docs/reference/kubectl/cheatsheet/
• https://medium.com/flant-com/kubectl-commands-and-tips-7b33de0c5476
• https://github.com/robscott/kube-capacity