Architecture
Kubernetes (k8s)
Open-source container orchestration system for automating software deployment, scaling, and management. Originally developed by Google, open-sourced in 2014, maintained by CNCF.
glossary: canary deployment, rolling-updates, load-balancing, fault-tolerance, service discovery, configuration management, self-healing.
Ref: https://kubernetes.io/docs/concepts/overview/what-is-kubernetes/
Architecture
Ref: https://kubernetes.io/docs/concepts/overview/components/
Components:
kubectl - we interact with this in CLI, its a tool
Cluster- collection of a control plane (Master in above image) and one or more worker nodes
Control plane:
- API server - exposes control plane to CLI and only component which interacts with worker nodes directly
- etcd - persistent key-value store contains cluster meta-data
- scheduler - assigns pods to nodes, just like a CPU scheduler
- controller manager - manages and maintains worker node, multiple controllers handle diff tasks, clubbed under one name. Functioning: Runs control loops (as in robotics) to continuously match the specified state to actual state of all the objects in the system.
Node:
- kube-proxy - contains networking info about containers on worker
- kubelet - it makes sure that containers are running in a pod
- container runtime - docker, containerd, etc…
Pod
Smallest unit deployable and manageable by k8s. It represents a running process on the cluster.
Pod = container + unique IP + storage volumes
Each worker node can have multiple pods and each pod can have multiple containers. Better to keep single container per pod. Gets it own IP address everytime it goes up.