Introduction

Basic Terms : System Parts

• Kubernetes : 전체 orchestration system
• K8s “k-eights” or Kube 라 부름
• Kubectrl : CLI to configure Kubernetes and manage apps
• Node : Kubernetes cluster에 있는 단일 서버
• Kubelet : Kubernetes agent running on nodes
• Control Plane : Set of containers that manage the cluster
  • API server, scheduler, controller manager, etcd 등을 포함
  • 때때로 master라고 불림

Kubernetes Container Abstractions

• Pod : one or more containers running together on the Node
  • Basic unit of deployment. Containers are always in pods
• Controller : For creating/updating pods and other objects
  • Many types of Controllers inc. Deployment, ReplicaSet, StatefulSet, DaemonSet, Job, CronJob, etc.
• Service : network endpoint to connect to a pod
• Namespace : Filtered group of objects in cluster
• Secrets, ConfigMaps and more

Three Management Approaches

• Imperative commands : run, expose, scale, edit, create, deployment
  • Best for dev/learning/personal projects
  • Easy to learn, hardest to manage over time
• Imperative objects : create -f file.yml, replace -f file.yml, delete …
  • Good for prod of small environments, single file per command
  • Store your changes in git-based yaml files
• Declarative objects : apply -f file.yml or dir\, diff
  • Best for prod, easier to automate
  • Harder to understand and predict changes
• Most Important Rule:
  • Don’t mix the three approaches
  • Store yaml in git, git commit each change before you apply
  • This trains you for later doing GitOps(where git commits are automatically applied to clusters)

  Building Your YAML Files

• Kind : we can get a list of resources the cluster supports
  • kubectl api-resources
• Notice some resources have multiple API’s (old vs. new)
• apiVersion : We can get the API versions the cluster supports
  • kubectl api-versions
• metadata : only name is required
• spec : where are the action is at!

Storage in Kubernetes

• Storage and stateful workloads are harder in all systems
• Containers make it both harder and easier than before
• StatefulSets is a new resource type, making POds more sticky
• Bret’s recommendation : avoid stateful workloads for first few deployments until you’re good at the basics
  • Use db-as-a-service whenever you can

Volumes in Kubernetes

• Creating and connecting Volumes : 2 types
• Volumes
  • Tied to lifecycle of a Pod
  • All container in a single POd can share them
• PersistentVolumes
  • Created at the cluster level, outlives a Pod
  • Seperates storage config from Pod using it
  • Multiple Pods can share them
• CSI plugins are the new way to connect to storage

Ingress

• None of our Service types work at OSI Layer 7 (HTTP)
• How do we route outside connections based on hostname or URL?
• Ingress Controllers (optional) do this with 3rd party proxies
• Nginx is popular, but Traefix, HAProxy, F5, Envoy, Istio, etc.

CRD’s and The Operator Pattern

• You can add 3rd party Resources and Controllers
• This extends Kubernetes API and CLI
• A pattern is starting to emerge of using these together
• Operator : automate deployment and management of complex apps
• e.g. Databases, monitoring tools, backups, and custom ingresses

Higher Deployment Abstractions

• All out kubectl commands just talk to the Kubernetes API
• Kubernetes has limited built-in templating, versioning, tracking, and management of your apps
• There are now over 60 3rd party tools to do that, but many are defunct
• Helm is the most popular