What is it?
Kubernetes main efforts is to maintain a given desired state, which is defined by a YAML file that describes the state itself, and not the operations used to get to this state. Mostly all components in Kubernetes Architecture try to maintain this given state.
How does it work?
Kubernetes consistently watches how it’s working, creating the observed state. When the observed state differs from the desired state, for example, a Pod crashed, all components work together to bring up another Pod, and then reconciling the observed state to be equal as the desired one.
The desired state can be defined and described using the Deployment object, expressing what the system should look and behave like. Here’s an example of a Deployment object, that defined 3 replicas of a application at all times:
apiVersion: apps/v1
kind: Deployment
metadata:
name: my-app
spec:
replicas: 3
selector:
matchLabels:
app: my-app
template:
metadata:
labels:
app: my-app
spec:
containers:
- name: my-container
image: my-image