Unshackling the database!

Using Redis as the primary data storage to ensure rapid growth scalability

Transcript

1. Unshackling the database! Using Redis as the primary data storage to ensure rapid growth scalability
2. $ whoami • marc.brandenburg • 3 years in Japan • 3 months at SmartNews ◦ Fundamental Architecture Squad • Interests: Kotlin, DDD, Clean Architecture
3. Agenda • Project case study ◦ Improving an existing system • Workflow ◦ Analysis ◦ Design ◦ Implementation ◦ Deployment
4. Existing System
5. A/B test system Control Test
6. Use cases System Project team User Analytics 1) create new tests 2) fetch test identifiers 3) analyze test results
7. Quality Attributes • Performance ◦ 200ms response time of news feed refresh • Scalability ◦ 30k+ requests / second ◦ 1.5M+ unique users during peak time • Availability ◦ 99.99% of requests return a response
8. Analysis
9. Current Architecture Analytics Admin API Database 1) query 2) read 3) compute 4) write 5) return
10. Problem Analysis Analytics Admin Database Not optimized for read-heavy load API 1) query 2) read 3) compute 4) write 5) return
11. Problem Analysis Analytics Admin Database Response waits for database writes API 1) query 2) read 3) compute 4) write 5) return
12. Problem Analysis Analytics Admin Database Database serves too many masters API 1) query 2) read 3) compute 4) write 5) return
13. Problem Analysis Analytics Admin Database Logic spread over multiple project API Logic isn’t flexible enough 1) query 2) read 3) compute 4) write 5) return
14. Design
15. Proposed Architecture Analytics Admin API Database 1) query 2) read 3) compute 5) write 4) return Cache Stream 6) write
16. Proposed Architecture Analytics Admin API Database 1) query 2) read 3) compute 5) write 4) return Cache Stream 6) write Write admin changes into fast cache
17. Proposed Architecture Analytics Admin API Database 1) query 2) read 3) compute 5) write 4) return Cache Stream 6) write Update asynchronously
18. Proposed Architecture Analytics Admin API Database 1) query 2) read 3) compute 5) write 4) return Cache Stream 6) write Refactor into single code base Follow principles of clean architecture
19. Research • Redis seems like a good choice for the cache ◦ How to optimize memory consumption? ◦ How to ensure data durability?
20. Memory consumption • The intset encoding for SETs
21. Data durability • Append-only file persistence
22. Feedback Session • Short feedback loop crucial ◦ Improvements from team members ◦ Updates about my research
23. Implementation
24. Tech stack • Kotlin • Coroutines • Spek Framework • Spring Boot 2
25. Entities Entities
26. Entities Entities Pure functions Table-driven tests Core business logic Changes are infrequently
27. Use Cases Delivery Entities Admin Sync
28. Use Cases Delivery Entities Admin Sync Asynchronous data flow between components External dependencies defined as interfaces Modelling the applications BDD test scenarios
29. Gateways Delivery Entities Admin Sync Kinesis Redis SQL
30. Gateways Delivery Entities Admin Sync Kinesis Redis SQL Implementations of use case interfaces Semi-manual test
31. Deployment
32. Performance tests • Locust • Determine correct instance sizing • Setup monitoring dashboards/alerts • Observe any resource leakage
33. Canary deployment • New Relic • Observe on a subset of production users • Identify any final issues
34. Outcome • Slashed peak response time by up to 50% • Removed rate limiting database protection
35. http://about.smartnews.com/careers/ We’re Hiring!!