My forth day started learning Kubernetes

Job

• Short lived tasks
• 建立一個或多個Pod並確保他們成功結束
• Job是一個一個建立的，想要同時建立多個Job，加入平行化功能。

Job yaml

CronJob

• An extension of the Job
• Provides a method of executing jobs on a cron-like schedule
• UTC only

CronJob yaml

確認CrobJob是否成功

• 查看History
  • 最後三個成功的Job會被留存
  • 最後一個失敗的Job會被保留
  • sucessfultJobsHistoryLimit可以讓你不保留成功的History (set to zero)

Rolling Updates

• maxSurge 在更新途中，可以多建立幾個Pod用來更新。
• maxUnavailable 在更新途中，可以允許幾個Pod是不可用的。

Blue-Green Deployment

Challenge

• Does not solve thenew database schema problem entirely
• You have to over provision the cluster size

Services

• IP是隨機的，當Pod被Replace之後還能正常連接其他Pod，需要依賴Service
• 使用static IP, static DNS name, [servicename].[namespace].svc.cluster.local

Service 用 Selector 配對Pod

透過Service訪問指令Pod，Service還提供了Load Balance的功能。

Service分為三種類型: ClusterIP(Default), NodePort, LoadBalancer

ClusterIP

• Default service that with Cluster internal visibility.
• port 是Service可以被外部連線的路徑，TargetPort是Pod內部可以被連線的路徑。
• 使用Round Robin演算法做Load Balanced ，可以設定Session affinity(黏性對話)。
• 用途是提供一個持久化的方法給同Cluster內的其他Pod呼叫使用。

ClusterIP yaml

NodePort

• extend the ClusterIP service with internal and external visibility.
• 如果不指定對外port(30000~32767), k8s會自己幫你指定一個。
• Node必須要有公開的IP address，使用Node IP + nodePort 連接到你的 service

圖示

LoadBalancer L4 & Ingress L7

• Make you expose your applications outside of the cluster
• LoadlBalancer是給Cloud Provider，Cloud Provider會提供一組IP，連接到你的port

🧱 Docker Desktop 的特例
Docker Desktop 沒有真實的 Load Balancer，也沒有外部 Node IP。為了讓開發體驗簡化，它模擬 LoadBalancer 行為：
當你建立一個：

type: LoadBalancer
ports:
- port: 8080
  targetPort: 80

Docker Desktop 檢測到後，會建立一個轉發規則：

localhost:<port> → ClusterIP:<port>

也就是說：
在 Docker Desktop 環境裡，「Cluster 內部的 port（8080）」被綁定到了本機 localhost:8080。
所以你打 localhost:8080，其實等於從外部直接連進 ClusterIP:8080。

Storage & Persistence

• Persistent Volume(PV)
  • Cluster wide & managed by administrator
• Persistent Volume Claims(PVC)
  • One to one mapping to a persistent volume
  • PVC can mount one or more pod on it.
  • All conctainers within the pod share the same value.

Persistent Volume & Persistent Volume Claims

Recliam Policy

• Delete the data upon pod deleton.
• You can set the retain policy to keep your data.

Access Mode

• ReadWriteMany The volume can be mounted as read-write by many pods
• ReadOnlyMany The volume can be mounted read-only by many pods
• ReadWriteOnce The volume can be mounted as read-write by a single pod, and the other pods are in read-only mode. The first one have the authoriy to implement write.

圖示

Persistent Volume state

• Available A free resource that is not yet bound to a claim.
• Bound The volume is bound to a claim
• Released The claim has been deleted, but the resource is not yet reclaimed by the cluster.
• Failed The volume has failed it automatic reclamation.

StorageClass

• represent a class storage offered by the admin
• A abstraction on top of an external storage resource.
• No need to set a capacity
• Eliminates the need for the admin to pre-provision a persistent volume.

StorageClass yaml

Application Settings

ConfigMap

• Decouple and externalize configuration.
• static meaning that if you change values, the container will have to be restarted to get them.
  

To solve the static problem, using Volume instead

Secret

• Stored as base64 encoded string
• Not secure
• To upgrade security, you should try to implement RBAC authorization Policy. or store secrets elsewhere

Observalibilty

• K8s會在Pod異常時重新建立一個Pod、但是如果是Container失敗他就不會做任何事。
• Readiness Probes可以知道container啟動的時間，可以告訴Kubernetes先等待，才接受流量。
• Liveness Probes 可以知道code是否正在運行，也可以重啟container

1. StartupProbe 請 k8s 等待 10s後才對/health 發出 HTTP calls.並最多嘗試三次。
2. ReadinessProbe 請 k8s 在 5 秒後開始檢查是否準備好，每十秒使用 TCP socket 連線測試。失敗會停止接受 pod traffic
3. LivenessProbe 請k8s在 15 秒後開始檢查是否活著，每二十秒使用 TCP socket 連線測試。 失敗會重啟pod

Dashboards

• Lens
• K9s

Scaling

• Horizontal Pod AutoScaling use the K8s Metrics Server
• Pods must have requests and limits defined
• The HPA checks the Metrics Server every 30s.
• Scale according to the min and max number or replicas defined
• Cooldown / Delay
  • Prevent racing condition
  • Once a change has been made, HPA waits
  • By default, the delay on scale up events is 3 minutes, and the delay on scale down event is 5 minutes.

yaml

常見Kubernetes 指令

	Command	Description
1	kubectl create job [jobName] --image=[imageName]	The imperative way
2	kubectl get job	List jobs
3	kubectl describe job [jobName]	Get info
4	kubectl delete job [jobName]	Delete a job
5	kubectl create cronjob [jobName] --imgage=[imageName] --schedule="*/ * * * *" -- bin/sh -c "date;"	The imperative way
6	kubectl get cf	List CrobJobs
7	kubectl describe cf [jobName]	Get info
7	kubectl delete cf [jobName]	Delete a cronJob
	Rolling Update
1	kubectl rollout status	Get the progress of the update
2	kubectl rollout history deployment [deployment]	Get the history of the deployment
3	kubectl rollout undo [deploymentName]	Rollback a deployment
4	kubectl rollout undo [deploymeny] --to-revision=[revision#]	Rollback to a revision number
	Service
1	kubectl expose po [podname] --port=80 --target-port=8080 --name=frontend	Create a service to expose a pod
2	kubectl expose deploy [deployName]	Create a service to expose a deployment
3	kubectl get svc	Get the service list
4	kubectl get svc -o wide	Get extra info
5	kubectl describe svc [servicename]	Describe the service
6	kubectl delete svc [servicename]	Delete the service using it’s name
	NodePort cheatsheet
1	kubectl expose po [podname] --port=80 --targetPort=8080 --type=NodePort	Create a service to expose a pod(You can not set nodePort!!!)
2	kubectl expose deploy [deplyName] --port=80 --targetPort=8080 --type=NodePort --name=frontend	Create a service to expose a deployment
	PV&PVC cheatsheet
1	kubectl get pv	Get the PV list
2	kubectl get pvc	Get the PVC list
3	kubectl describe pv [pvname]	Describe the PV
4	kubectl describe pvc [pvcname]	Describe the PVC
5	kubectl delete pv [pvname]	Delete the PV
6	kubectl delete pvc [pvcname]	Delete the PVC
7	kubectl get sc	List the StorageClass
8	kubectl describe sc [className]	Describe the StorageClass
	ApplicationSetting
1	kubectl create cm [name] --from-file=config.txt	Create configMap from file
2	kubectl create cm [name] --from-file=config/	Create configMap from folder
3	kubectl get cm	List the ConfigMaps
4	kubectl get cm [name] -o YAML	Save a ConfigMap in a YAML file
5	kubectl create secret generic [secretName] --from-literal=STATE=Michigan	The imperative way
6	kubectl get secrets	List the secrets
	HPA
1	kubectl autoscale deploy [name] --cpu-percent=50 --min=3 --max=10	The imperative way
2	kubectl get hpa	Get the auto scaler status
1	kubectl delete hpa [name]	Delete the HPA