Skip to content

Latest commit

 

History

History
134 lines (126 loc) · 4.82 KB

KUBERNETES.md

File metadata and controls

134 lines (126 loc) · 4.82 KB

Kubernetes 101

Docker vs. Kubernetes

Docker helps you package applications into images, and execute them in containers. Kubernetes is a robust platform for containerized applications. It abstracts away the underlying network infrastructure and hardware required to run them, simplifying their deployment, scaling, and management.

Kubernetes from the container up

Pods

The basic unit of a Kubernetes deployment is the Pod. A Pod encapsulates one or more containers. For example, the basket Pod specifies two containers:

deployments.yaml

The first container runs the eshop/basket.api image:

spec:
 containers:
 - name: basket
   image: eshop/basket.api
   env:
   - name: ConnectionString
     value: 127.0.0.1

Note the ConnectionString environment variable: containers within a Pod are networked via localhost. The second container runs the redis image:

- name: basket-data
 image: redis:3.2-alpine
 ports:
 - containerPort: 6379

Placing basket and basket-data in the same Pod is reasonable here because the former requires the latter, and owns all its data. If we wanted to scale the service, however, it would be better to place the containers in separate Pods because the basket API and redis scale at different rates.

If the containers were in separate Pods, they would no longer be able to communicate via localhost; a Service would be required.

Services

Services expose Pods to external networks. For example, the basket Service exposes Pods with labels app=eshop and component=basket to the cluster at large:

services.yaml

kind: Service
metadata:
 ...
 name: basket
spec:
 ports:
 - port: 80
 selector:
   app: eshop
   component: basket

Kubernetes's built-in DNS service resolves Service names to cluster-internal IP addresses. This allows the nginx frontend to proxy connections to the app's microservices by name:

nginx.conf

location /basket-api {
   proxy_pass http://basket;

The frontend Pod is different in that it needs to be exposed outside the cluster. This is accomplished with another Service:

frontend.yaml

spec:
 ports:
 - port: 80
   targetPort: 8080
 selector:
   app: eshop
   component: frontend
 type: LoadBalancer

type: LoadBalancer tells Kubernetes to expose the Service behind a load balancer appropriate for the cluster's platform. For Azure Container Service, this creates an Azure load balancer rule with a public IP.

Deployments

Kubernetes uses Pods to organize containers, and Services to network them. It uses Deployments to organize creating, and modifying, Pods. A Deployment describes a state of one or more Pods. When a Deployment is created or modified, Kubernetes attempts to realize that state.

The Deployments in this project are basic. Still, deploy.ps1 shows some more advanced Deployment capabilities. For example, Deployments can be paused. Each Deployment of this app is paused at creation:

deployments.yaml

kind: Deployment
spec:
 paused: true

This allows the deployment script to change images before Kubernetes creates the Pods:

deploy.ps1

kubectl set image -f deployments.yaml basket=$registry/basket.api ...
kubectl rollout resume -f deployments.yaml

ConfigMaps

A ConfigMap is a collection of key/value pairs commonly used to provide configuration information to Pods. The deployment script uses one to store the frontend's configuration:

deploy.ps1

kubectl create configmap config-files from-file=nginx-conf=nginx.conf

This creates a ConfigMap named config-files with key nginx-conf whose value is the content of nginx.conf. The frontend Pod mounts that value as /etc/nginx/nginx.conf:

frontend.yaml

spec:
 containers:
 - name: nginx
 ...
 volumeMounts:
 - name: config
   mountPath: /etc/nginx
 volumes:
 - name: config
   configMap:
     name: config-files
     items:
     - key: nginx-conf
       path: nginx.conf

This facilitates rapid iteration better than other techniques, e.g. building an image to bake in configuration.

The script also stores public URLs for the app's components in a ConfigMap:

deploy.ps1

kubectl create configmap urls --from-literal=BasketUrl=http://$($frontendUrl)/basket-api ...

Here's how the webspa Deployment uses it:

deployments.yaml

spec:
 containers:
 - name: webspa
   ...
   env:
     ...
     - name: BasketUrl
       valueFrom:
         configMapKeyRef:
           name: urls
           key: BasketUrl

Further reading