Considerations in Cloud DevOps

Presented at BNY Mellon Silicon Valley Innovation Center, 10/22/2014

Video: https://www.youtube.com/watch?v=q-c20Z1ZN-I&list=UUODym48KciXOkpU_U62Ut6g

Transcript

1. Considerations in Cloud DevOps. Ben Whaley @iAmTheWhaley
2. Mobile Scale
3. 0 900000000 1800000000 2700000000 3600000000 PCs Mobile Phones Connected Clients 3.5 billion 1.8 billion Source: Andreessen Horowitz
4. Growth in Connected Client Devices Units (millions) 0 400 800 1,200 1,600 Jun 07 Jun 08 Jun 09 Jun 10 Jun 11 Jun 12 Jun 13 Tablets iPhone & Android PCs Source: Andreessen Horowitz
5. 0 400000000000 800000000000 1200000000000 1600000000000 Photos Taken 2014 1999 1.5 trillion 80 billion Source: Andreessen Horowitz
6. ~1.3 Million Apps
7. Revisions measured in days or weeks.
8. IoT
9. Mobile is powered by APIs.
10. Mature Mobile • Use the cloud to enable rich native apps • Blur the line between apps and the web • Hyperlocal via iBeacons & NFC • Mobile != Smartphone
11. Business Benefits • APIs engage customers, partners • Data pipelines for business analytics • Improved availability • Elasticity for • Cost savings • Capacity on demand • Focus on core competencies
12. Systems
13. Systems • n+2 redundancy n n+1 n+2 Load Balancer
14. Systems • n+2 redundancy • Compute is ephemeral n n+1 n+2 n+3 n+4 n+5 Load Balancer
15. Systems • n+2 redundancy • Compute is ephemeral • Infrastructure is code resource "digitalocean_droplet" "www-1" {! image = "ubuntu-14-04-x64"! name = "www-1"! region = "nyc2"! size = "512mb"! private_networking = true! ssh_keys = [! "${var.ssh_fingerprint}"! ]! }!
16. Systems • n+2 redundancy • Compute is ephemeral • Infrastructure is code • Less is more
17. Systems • n+2 redundancy • Compute is ephemeral • Infrastructure is code • Less is more • Ubiquitous monitoring
18. Systems • n+2 redundancy • Compute is ephemeral • Infrastructure is code • Less is more • Ubiquitous monitoring • Service discovery
19. Applications
20. Applications • Loosely coupled
21. Applications • Loosely coupled • Deeply instrumented
22. Applications • Loosely coupled • Deeply instrumented • Lean
23. Applications • Loosely coupled • Deeply instrumented • Lean • Built to be disabled
24. Applications • Loosely coupled • Deeply instrumented • Lean • Built to be disabled • Stateless
25. Applications • Loosely coupled • Deeply instrumented • Lean • Built to be disabled • Stateless • Robustly implemented “Be conservative in what you send, be liberal in what you accept”
26. Applications • Loosely coupled • Deeply instrumented • Lean • Built to be disabled • Stateless • Robustly implemented • Continuously integrated
27. Delivery
28. Delivery Commit Test Build Deploy
29. Delivery Build server Commit Site Deploy
30. Delivery
31. Delivery Challenges
32. Delivery Challenges 1. Zero downtime
33. Zero Downtime Green/Blue www.example.com Old Code
34. Zero Downtime Green/Blue www.example.com Old Code New Code
35. Zero Downtime Rolling Upgrade www.example.com Old Old Old
36. Zero Downtime Rolling Upgrade www.example.com New Old Old
37. Zero Downtime Rolling Upgrade www.example.com New New Old
38. Zero Downtime Rolling Upgrade www.example.com New New New
39. Zero Downtime Replacement www.example.com Old Old Old
40. Zero Downtime Replacement www.example.com New New New
41. Delivery Challenges 1. Zero downtime 2. Distributing credentials
42. Distributing Credentials 1. Seed credentials in the OS image Pros Cons Simpler bootstrap process Must load credential to image initially May be more difficult to modify Credentials in plain text in stored image
43. Distributing Credentials 1. Seed credentials in the OS image! 2. Commit encrypted config data to DVCS repository Pros Cons Data stored alongside code in version control Distributing decryption key to nodes varies by tool Easily updated Key distribution
44. Distributing Credentials 1. Seed credentials in the OS image! 2. Commit encrypted config data to DVCS repository! 3. Asymmetric encryption Pros Cons Public/Private key pair may already exist Requires mature, possibly automated cert infrastructure Easy rotation and revocation Limited implementations No shared secret
45. Delivery Challenges 1. Zero downtime 2. Distributing credentials 3. Database migrations
46. Database Migrations 1. NoSQL vs Relational!
47. Database Migrations 1. NoSQL vs Relational! 2. Multi-stage! a. Make backward compatible DB changes ! b. Run new code that is both backward and forward compatible! c. Migrate old data to new schema! d. Remove code that understands old schema! e. Clean up old schema
48. Delivery Challenges 1. Zero downtime! 2. Distributing credentials! 3. Database migrations! 4. Distributing the code! 5. Updating the config! 6. Rollbacks
49. Thanks! ! Ben Whaley ! bwhaley.com @iAmTheWhaley