Persistent Storage with Kubernetes in Production Which solution and why?

Transcription

1 Persistent Storage with Kubernetes in Production Which solution and why? Kubecon + CloudNativeCon, Dec 8, 2017 Cheryl Hung, Product Manager

2 Cheryl Product manager at StorageOS 2

3 Objectives Why is state tricky? How do I evaluate storage? What storage options are there for Kubernetes? 3

4 Objectives Why is state tricky? How do I evaluate storage? What storage options are there for Kubernetes? Anti-objective: Should I use a database, message queue, NoSQL for my app? 4

5 Why is state so tricky? 5

6 Why do I need storage?

7 Why do I need storage? 7

8 Problem 1: No pet storage StorageOS Ltd. All rights reserved. 8

9 Problem 2: Data needs to follow StorageOS Ltd. All rights reserved. 9

10 Problem 3: Gravity can be a problem 10

11 Problem 4: Storage isn t just storage App Binaries App data Config Backup 11

12 Kubernetes Storage Model: Persistent Volumes and Claims

13 Kubernetes Storage Model: Persistent Volumes and Claims Registers PVs in the pool NFS PV iscsi PV NFS PV GCE PV Administrator Pool of Persistent Volumes

14 Kubernetes Storage Model: Persistent Volumes and Claims Registers PVs in the pool NFS PV iscsi PV NFS PV GCE PV Administrator Pool of Persistent Volumes Claims a PV from the pool claim Developer

15 Kubernetes Storage Model: Persistent Volumes and Claims Registers PVs in the pool NFS PV iscsi PV NFS PV GCE PV Administrator Pool of Persistent Volumes Pod Claims a PV from the pool claim Developer References claim in pod claim

16 Dynamic provisioning with Storage Classes Registers Storage Classes fast NS PV slow GCE PV= Administrator Storage Classes

17 Dynamic provisioning with Storage Classes Registers Storage Classes fast NS PV slow GCE PV= Administrator Storage Classes Claims a PV from the pool claim Developer

18 Dynamic provisioning with Storage Classes Registers Storage Classes fast NS PV slow GCE PV= Administrator Storage Classes fast PV Pod Claims a PV from the pool claim Developer References claim in pod claim

19 Eight Principles of Cloud Native Storage 19

20 What is Cloud Native? Horizontally scalable No single point of failure Resilient and self healing Minimal operator overhead Decoupled from the underlying platform 20

21 1. Application centric 1 Storage should be presented to and consumed by applications, not by operating systems or hypervisors 21

22 2. Platform agnostic 2 The storage platform should be able to run anywhere. Upgrades and scaling is non-disruptive. 22

23 3. Declarative and composable 3 Storage resources should be declared and composed just like all other resources required by applications and services. 23

24 4. API Driven 4 Storage resources and services should be easy to be provisioned, consumed, moved and managed via an API. 24

25 5. Natively secure 5 Storage services should integrate and inline security features such as encryption and RBAC. 25

26 6. Agile 6 The platform should be able to move application data between locations, dynamically resize volumes for growth, take point in time copies of data for data retention or to facilitate rapid recovery of data. 26

27 7. Performant 7 The storage platform should be able to offer deterministic performance in complex distributed environments. 27

28 Block vs file vs object Block Storage Data stored in a fixed-size blocks in a rigid arrangement ideal for enterprise databases File Storage Data stored as files in hierarchically nested folders ideal for active documents Object Storage Data stored as objects in scalable buckets ideal for unstructured big data, analytics and archiving

29 8. Consistently available 8 The storage platform should manage data distribution with a predictable, proven data model to ensure high availability, durability, consistency of application data. 29

30 Storage Landscape 30

31 Meet Jane A DevOps engineer in a bank She wants to migrate a Postgres database to containers

32 Centralized file system NFS Server Share: /Share/ TCP/IP Network NFS Client 1: mount /share/ into/home/data/nfsshared/ NFS Client 2: mount /share/ into/mnt/data/nfsshared

33 Centralized file system Single point of failure Hard to scale horizontally No Docker/Kubernetes integration

34 Proprietary storage array

35 Proprietary storage array Deterministic performance Vendor lock in No thin provisioning Hard to scale horizontally Expensive and long lead times

36 Distributed storage File (ino, ono) oid Objects hash(oid) & mask pgid PGs CRUSH(pgid) (osd1, osd2) OSDs (grouped by Failure domain)

37 Distributed storage Horizontally scalable Hardware agnostic Write amplification Failures are expensive Complicated to set up

38 Public cloud SAN EC2 Instance EBS Volume EBS Volume EBS Volume

39 Public cloud SAN Horizontally scalable Consistent and performant Attach/detach times on physical block devices is slow Expensive, vendor lock in Compliance

40 API framework MESOS / Marathon Docker (Swarm mode) Kubernetes API framework Interface (Native, Flex, DVDI, CSI)

41 API framework Docker/Kubernetes integration No guarantees for performance, replication, failover, scalability

42 Orchestrated storage (StorageOS)

43 Orchestrated storage (StorageOS) Platform agnostic

44 Orchestrated storage (StorageOS) Platform agnostic Horizontally scalable

45 Orchestrated storage (StorageOS) Platform agnostic Horizontally scalable Database (ie. block)

46 Orchestrated storage (StorageOS) Platform agnostic Horizontally scalable Database (ie. block) Docker/K8s integration

47 Orchestrated storage (StorageOS) Platform agnostic Horizontally scalable Database (ie. block) Docker/K8s integration High availability

48

49 Orchestrated storage (StorageOS)

50 Create a fast StorageClass Registers Storage Classes fast NS PV GCE PV= Administrator Storage Classes

51 Claim a persistent volume Registers Storage Classes fast NS PV slow GCE PV= Administrator Storage Classes Claims a PV from the pool claim Developer

52 Claim a persistent volume Registers Storage Classes fast NS PV slow GCE PV= Administrator Storage Classes fast PV Pod Claims a PV from the pool claim Developer References claim in pod claim

53 High availability with StorageOS

54 Orchestrated storage (StorageOS) Horizontally scalable Consistent and performant Platform agnostic Synchronous replication Volume is limited to the size of one node

55 To Recap 55

56 Storage is tricky so think about your storage needs! Eight principles of cloud native storage 1. Application centric 5. Natively secure 2. Platform agnostic 6. Agile 3. Declarative and composable 7. Performant 4. API driven 8. Consistently available

57 K8S Storage SIG / CNCF Storage WG

58 StorageOS resources Browser-based tutorials my.storageos.com/main/tutorials Quickstart storageos.com/kubernetes Slack channel slack.storageos.com We re hiring!

59 Thanks Slides at oicheryl.com