Scale out, with Kubernetes (k8s)
•
3 likes•1,183 views
Presentation is in english, besides first slide. In presentation I represented basics concepts from docker and kubernetes, each part ends with short example. On my github: https://github.com/arekborek/k8s-jdd you can find additional information.
Report
Share
Scale out, with Kubernetes (k8s)
- 1. Arkadiusz Borek LJUG Scale out, a mówili, że się nie da
- 3. Microservices? You probably heard a lot already!
- 5. Old way: shared machines ● No isolation ● No namespacing ● Common libs ● High coupled apps and OS source@kubernetes.io
- 6. Old way: shared app servers ● No isolation ● No namespacing ● Common libs source@kubernetes.io
- 7. Old way: many virtual machines ● Some isolation ● Expensive and inefficient ● Still highly coupled to the guest OS ● Hard to manage source@kubernetes.io
- 9. Containers: Process isolation ● CPU (PIDs) ● Memory ● Network interfaces ● Filesystem
- 10. Containers: Security ● Chroots determines what parts of the filesystem a use can see ● Control - CPU, Memory, IO via cgroups ● Limits what a user can do: mount, kill, chown ● No access to host !!!
- 11. Containers: Other features ● New approach to packaging ● Lightweight ● Runs on any machine Physical, virtual, cloud ● Write once – run everywhere! Development, Test, Staging, Production
- 14. Containerize Option #1 Docker file Containerize Option #2 Maven plug-in made by spotify Containerize Option #3 Docker hub / GitHub
- 16. Docker file /hellonode - Dockerfile - server.js
- 17. server.js var http = require('http'); var handleRequest = function(request, response) { console.log('Received request for URL: ' + request.url); response.writeHead(200); response.end('Hello World (' + process.env.HOSTNAME + ':V1) !'); }; var www = http.createServer(handleRequest); www.listen(8080);
- 18. Dockerfile FROM node:4.4 EXPOSE 8080 COPY server.js . CMD node server.js
- 19. Build Docker Container docker build -t gcr.io/$PROJECT_ID/hello-jdd:v1 .
- 20. Run Docker Container docker run -d -p 8080:8080 --name hello-jdd gcr.io/$PROJECT_ID/hello-jdd:v1
- 21. Test for ((;;)) do curl http://localhost:8080; printf 'n'; sleep 1; done
- 22. What next ?? ● What if there are more containers? ● Scheduling – where should my containers run ? ● Lifecycle and health: keep my containers running ● Discovery: where are my container now ? ● Monitoring: What’s happening with my container ? ● Scaling: Making job bigger or smaller
- 23. How to handle it ?
- 26. Enter Kubernetes (k8s) ● Greek for “Helmsman”, also the root of the word “Governor” ● Container orchestrator, run containers ● Support multiple cloud and bare-metal environments ● Inspired and informed by Google’s experiences and internal systems ● Open source, written in Go
- 28. Enter Kubernetes (k8s) ● Kubernetes master – order others ● Kublet – ask how many resources you have if enough run machine ● Distributed process scheduler ● Cluster of machines as one ● All your machines via kubernetes can be treats as one single machine
- 29. k8s basis concepts ● Pod
- 30. Pod ● Group of containers ● One atomic resource that can be managed by Kubernetes ● Live and die together ● Shared network interface
- 31. k8s basis concepts ● Pod ● Pod networking
- 32. Pod networking ● Pod IPs routable, docker default is private IP ● Pods can reach each other without NAT, even across nodes ● No brokering of port numbers ● This is fundamental requirement, several SDN solutions
- 33. k8s basis concepts ● Pod ● Pod networking ● Replication Controllers
- 34. Replication Controller ● Allow start pods multiple times ● Check state if one of you pod died, RC run next instance of container for you
- 35. k8s basis concepts ● Pod ● Pod networking ● Replication Controllers ● Service
- 36. Service ● Provide access to your N pods ● LB – front of pods ● Only route traffic to pods ● Selector-label concept
- 37. k8s basis concepts ● Pod ● Pod networking ● Replication Controllers ● Service ● Labels & Selectors
- 38. Labels & Selectors ● Label anything ● Name-value pair ● Make your own
- 39. Example of kube
- 40. Push image gcloud docker push gcr.io/$PROJECT_ID/hello-jdd:v1
- 41. Replication Controller kubectl create -f rc-v1.yaml
- 42. apiVersion: v1 kind: ReplicationController metadata: name: hello-jdd-v1 spec: replicas: 1 template: metadata: name: hello-jdd-v1 labels: app: hello-jdd-v1 env: jdd tier: backend version: v1 visualize: "true" spec: containers: - name: hello-jdd image: "gcr.io/strong-art-145220/hello-jdd:v1" imagePullPolicy: Always ports: - containerPort: 8080
- 43. Service Allow external traffic kubectl create -f service.yaml
- 44. apiVersion: v1 kind: Service metadata: labels: env: jdd visualize: "true" name: hello-jdd spec: ports: - port: 8080 #The port that this service should serve on. selector: #Label keys and values connect RC and Service env: jdd tier: backend type: LoadBalancer
- 45. Test :) for ((;;)) do curl http://localhost:8080; printf 'n'; sleep 1; done
- 46. Scale out kubectl scale rc hello-jdd-v1 --replicas=2
- 47. Roll out an upgrade gcloud docker push gcr.io/$PROJECT_ID/hello-jdd:v2 kubectl create -f rc-v2.yaml kubectl scale rc hello-jdd-v1 –replicas=1 kubectl scale rc hello-jdd-v2 –replicas=2 kubectl scale rc hello-jdd-v1 --replicas=0
- 48. --rollback kubectl scale rc hello-jdd-v1 --replicas=1 kubectl scale rc hello-jdd-v2 --replicas=1 kubectl scale rc hello-jdd-v1 --replicas=2 kubectl scale rc hello-jdd-v2 --replicas=0
- 49. Some useful cmds kubectl get rc kubectl get pods kubectl logs <POD_NAME> kubectl cluster-info kubectl config view kubectl describe rc <RC_NAME> kubectl get -o json pod
- 50. Some useful ... Staging vs. production Use label – deploy in the same infrastructure Service discovery Kubernetes API Or… DNS Lookups! Use Docker Machine In the cloud – faster network to download images
- 51. Some useful ... Don’t Log to Container Filesystem! Log to a volume… docker -v /tmp/log:/log Or Send it elsewhere! Clean up disk spaces Every image, layer and even container litters Use Spring Profile One container – run on Docker Machine, Kubernetes, and App Engine!
- 52. THE END !!