Skip to content
This repository was archived by the owner on Dec 25, 2023. It is now read-only.

Kubernetes-workload.md

Latest commit

 

History

History
359 lines (252 loc) · 7.02 KB

Kubernetes-workload.md

File metadata and controls

359 lines (252 loc) · 7.02 KB

Deployment, Jobs and Scaling

Deployment

Deployment describes a desired state of pods.

.yaml -> Deployment object -> Deployment controller -> Node (Pod1, Pod2, ...)

  • Ways to create deployments
kubectl apply -f [DEPLOYMENT_FILE]
kubectl run [DEPLOYMENT_NAME] \
--image [IMAGE]:[TAG] \
--replicas 3 \
--labels [KEY]=[VALUE] \
--ports 8080 \
--generator deployment/apps.v1 \
--save-config
  • Use kubectl to inspect your Deployment
kubectl get/describe deployment DEPLOYMENT_NAME [-o wide/yaml]
  • Scaling a Deployment manually
kubectl scale deployment DEPLOYMENT_NAME -replicas=5
  • Autoscaling a Deployment
kubectl autoscale deployment DEPLOYMENT_NAME --min=5 --max=15 --cpu-precent=75

Updating deployments

kubectl apply -f DEPLOYMENT_FILE

kubectl set image deployment DEPLOYMENT_NAME IMAGE IMAGE:TAG

kubectl edit deployment/DEPLOYMENT_NAME
  • rolling updates
kind: deployment
spec:
  strategy:
    rollingUpdate:
      maxSurge: 5
      maxUnavailable: 30%
  • blue-green deployments

when updating, a completely new deployment is created. Once new application is ready, all traffic is shifted to new version.

  • Canary deployment

Based on blue-green method, but the traffic is gradually shifted to the new version.

  • rolling back a Deployment
kubectl rollout undo deployment DEPLOYMENT_NAME

kubectl rollout undo deployment DEPLOYMENT_NAME --to-version=2

kubectl rollout undo deployemnt DEPLOYMENT_NAME --revision=2

Managing deployments

Three different states: Progressing State, Complete State, Failed State

  • Pausing a Deployment
kubectl rollout pause deployment DEPLOYMENT_NAME
  • Resuming a Deployment
kubectl rollout resume deployment DEPLOYMENT_NAME
  • Monitoring a Deployment
kubectl roolout status deployment DEPLOYMENT_NAME
  • Deleting a Deployment
kubectl delete deployment DEPLOYMENT_NAME

Jobs and CronJobs

.yaml file -> job object -> job controller -> pod

  • Parallel Jobs
kind: Jon
spec:
  completions: 3
  parallelism: 2
  • Inspecting a Job
kubectl describe job JOB_NAME

kubectl get pod -l [job-name=my-app-job]
  • Scaling a Job
kubectl scale job JOB_NAME --replicas=VALUE
  • Failing a Job
kind: Job
spec:
  backoffLimit: 4
  activeDeadlineSeconds: 300
  • Deleting a Job
kubectl delete -f JOB_NAME

kubectl delete job JOB_NAME

kubectl delete job JOB_NAME --cascade false
  • CronJob
kind: CronJob
spec:
  schedule: "*/1 * * * *"
  startingDeadlineSeconds: 3600
  concurrencyPolicy: Forbid
  suspend: True
  successfulJobsHistoryLimit: 3
  failedJobHistoryLimit: 1
  jobtemplate: {}
  • Create a CronJob
kubectl apply -f FILE
  • inspect a CronJob
kubectl describe cronjob NAME
  • delete a CronJob
kubectl delete cronjob NAME

Controlling pod placement

  • Affinity and anti-affinity
kind: Pod
spec:
  affinity:
    nodeAffinity:
      requiredDeringSchedulingIgnoreDuringExecution:
        nodeSelectorTerm:
          - matchExpressions:
            - key: accelerator-type
            operator: In
            values:
              - gpu
              - tpu
  • Taints and tolerations
# taint a to repel Pods
kubectl taint nodes node1 key=Value:NoSchedule key2=Value2:NoSchedule
toleration:
- key: "key"
  operator: "Equal"
  value: "value"
  effect: "Noschedule"

Getting software into your cluster

  • How to get software

    1. Build it yourself, and supply your own YAML
    2. Use Helm to install software into your cluster

Create -> Version -> Share -> Publish

  • Helm architecture

Helm client (helm) -> helm server (Tiller) -> Kubernetes APIserver (inside Kubernetes cluster)

Kubernetes Engin Networking

Service to expose applications that are running within Pods. LoadBalancers to expose Services to external clients.

Pod Networking

Each pod has unique IP address. Connect to each other use Pod route networking namespace.

Services

A Kubernetes Service creates endpoints, label selector to select pods then distribute workload.

  • Ways to find a Service in GKE

    1. Environment variables
    2. Kubernetes DNS
    3. lstio

Service types and Load Balancers

  • ClusterIP

ClusterIP Service has a static IP address, Pods use this static IP address to communicate the service.

kind: Service
spec:
  type: ClusterIP
  selector:
    app: Backend
  • NodePort

When creating NodePort Service, ClusterIP service is automatically created in the process. The service can be reached from outside of the cluster using the IP of any Node. traffic will be directed to a service on port 80 and further redirected to one the backend pods on target port.

Can be useful to expose a service through an external laodbalancer that setup and manage yourself. Need to deal with node management, making sure there are no pod collisions.

kind: Service
spec:
  type: NodePort
  selector:
    app: Backend
  ports:
    - protocol: TCP
      nodePort: 30100
      port: 80
      targetPort: 9376
  • LoadBalancer

Loadbalancer forwards the traffic onto the nodes for the service.

kind: Service
spec:
  type: LoadBalancer
  • An engineering tradeoff: Double-hop or imbalance

The network LoadBalancer chooses a random node in the cluster and forwards the traffic to it. then in order to balance the traffic, the iptable will randomly select to pods to handle the incoming traffic. And the pod may be or may not be the same node of the iptable. If the pod not in the same node, moreover the pod need to send response back via iptable node, which is double-hop.

kind: Service
spec:
  type: LoadBalancer
  externalTrafficPolicy: Local # avoid double-hop - lowest latency

Ingress Resource

apiversion: extensions/v1beta1
kind: Ingress
spec:
  backend:
    serviceName: demo
    serviceport: 80
  • Updating an Ingress
kubectl edit ingress INGRESS_NAME

kubectl replace -f FILE

Container-Native Load Balancing

Network Security

Persistent Data and Storgae

Volumns

StatefulSets

ConfigMaps

Secrect