Managing Stateful apps on Kubernetes

This talk was presented at All Things Open 2018. The presentation is an introduction to Kubernetes objects used to manage stateful apps. It also includes an introduction to containers, container orchestration and Kubernetes itself.

Transcript

1. Managing Stateful Apps In Kubernetes
3. @rajashree_28 mrajashree I’m Rajashree
5. 1. Containers and Container Orchestration 2.Kubernetes 3.Stateless vs stateful apps 4.Running stateful apps on k8s
6. 1. Containers and Container Orchestration 2.Kubernetes 3.Stateless vs stateful apps 4.Running stateful apps on k8s
7. Containers
8. Application packaging • Source code, dependencies, env vars • Application portability • Performance consistency • No more “it works on my machine”
10. OS level virtualization • Containers share host Operating System • Lightweight as opposed to a VM • One host can run multiple containers • Resource isolation • Increased efficiency
11. Managing containers Container Node Container Container • Scheduling across nodes • Availability
12. EC2 Container EC2 EC2 Client Managing containers Container Container Container Container Container Container Container Container Run 3 containers Run 3 containers Run 3 containers
13. Managing containers • Which machine is my container running on? • Are there sufficient resources for my app? • How do containers communicate with each other? • How do I manage load balancing? • How do I ensure healthcheck/monitoring?
14. Container orchestration
15. Client EC2 Container Container Container EC2 Container Container Container EC2 Container Container Container Run 9 containers
16. 1. Containers and Container Orchestration 2.Kubernetes 3.Stateless vs stateful apps 4.Running stateful apps on k8s
17. Kubernetes! • Manage containerized applications • Open source • Scalability • Handle upgrades
18. Pod • Smallest deployable unit • One or more containers • “one-container-per-pod” • Unique IP address Container Pod 10.10.1.1 Node
19. Deployment • Pods are ephemeral • Deployments manage pod life-cycle • Ensure actual state same as desired • Provide easy scaling, upgrades for apps
20. k8s cluster Pod 1 10.10.1.1 10.11.1.1 10.11.1.2 Pod 2 Pod 3 Node 1 Node 2 Deployment selector=
 app: nginx labels=
 app: nginx labels=
 app: nginx labels=
 app: nginx
21. 1. Containers and Container Orchestration 2.Kubernetes 3.Stateless vs stateful apps 4.Running stateful apps on k8s
22. Stateless • No persisted data • No application state • Pods are replaceable • Examples: edge API server, nginx
23. Stateful • Persisted data • Server/node identity is irreplaceable • Examples: mysql, cassandra
24. Why run stateful apps on k8s? 1. Depends on use case 2. Leverage containerization benefits 3. Manage stateless and stateful apps together 4. Ability to use same infrastructure provider 5. Easy scalability 6. Affordability
25. 1. Containers and Container Orchestration 2.Kubernetes 3.Stateless vs stateful apps 4.Running stateful apps on k8s
26. K8s objects for statefulness • Persistent Volumes (PV) • Persistent Volume Claims (PVC) • Storage Class • StatefulSet
27. Persistent volume • Storage resource in k8s • Volume plugins • Created by cluster admin • Lifecycle independent of pod
28. Persistent Volume Claims • Request to claim a storage resource • Abstracts storage details of PV from user • User only specifies size and access mode • Access modes: RWO, RWM, ROM
29. Static Provisioning Cluster admin Creates PV PV size: 10 GB AccessMode: RWO User Creates PVC PVC requests: size: 3 GB (min) AccessMode: RWO Available PVController Bound
30. Limitations? • User created PVC requesting 3 Gib • But there was only one PV with 10 Gib • PV got bound to PVC with excess volume • User will get a PV-PVC binding only if admin has created enough PVs
31. Storage Class • Resource for classifying storage • Specifies provisioner, parameters, reclaimPolicy • Created by cluster admin • User must know name of Storage Class
32. Dynamic Provisioning Cluster admin Creates 
 Storage Class Name: standard
 Provisioner: aws-ebs Parameters: type=gp2 ReclaimPolicy: Delete User Creates PVC PVC storageClass: standard size: 3 GB (min) AccessMode: RWO Creates PV PV size: 3 GB AccessMode: RWO
33. Dynamic Provisioning • Efficient resource utilization • Only needs storageClass and PVC creation • PV created dynamically
34. Pod in stateful app • Stores data independently • Should have separate PV • Needs separate PVC for separate PV
35. • Lets say you want 10 pods for your app • Do you create 10 PVCs? • What about when you want to scale up? • Can something create PVCs dynamically too? Pod in stateful app
36. StatefulSets • Stable unique persistent volume for each pod • Creates PVC using a volume claim template • Stable unique network identifier for all pods • Ordered deployment and scaling • Automated rolling updates
37. Use Case - Cassandra • Highly available, scalable • Uses replication while storing data • All nodes are equal • Cassandra nodes can identify each other
38. EC2 EC2 cassandra-0 cassandra-1 cassandra-2 PVC 0 PV 0 PV 1 PV 2 PVC 1 PVC 2 StatefulSet
39. 1. Containers and Container Orchestration 2.Kubernetes 3.Stateless vs stateful apps 4.Running stateful apps on k8s
40. Thank you!