Stateful Interaction In Serverless Architecture With Redis: Pyounguk Cho
0 likes468 views
This presentation discusses how to bring stateful behaviors to serverless architecture using Redis. It introduces the problem of enabling statefulness in serverless applications and proposes using Redis as a solution. Key considerations for the Redis-based architectural approach are discussed, including topology, high availability and scalability, and Redis configuration tuning. A demo is then presented to illustrate "Redis in Serverless" in action.
Stateful Interaction In Serverless Architecture With Redis: Pyounguk Cho
- 1. PRESENTED BY Stateful Interaction in Serverless Architecture with Redis Pyounguk Cho Oracle, Director of Product Management
- 2. PRESENTED BY 1 Problem Space How to bring stateful behaviors to Serverless Architecture 2 Redis-based Solution Proposed solutions with Redis and considerations 3 Demo & Summary “Redis in Serverless” in action Agenda:
- 4. PRESENTED BY Software Development Drives Business Agility Development Process Application Architecture Deployment Packaging Application Infrastructure 2000-072008-152016+ Waterfall Monolithic Physical Server Datacenter N-Tier Virtual Server HostedAgile Plan Code Build Test Release Deploy Operate Monitor DevOps Microservices Containers Cloud
- 5. PRESENTED BY Serverless Architecture Only pay for compute resources actually consumed. Truly Cloud-Elastic Team can rapidly deploy and upgrade applications maintaining service availability DevOps Enabled Scales automatically to handle incoming requests Fully Autonomous 5
- 6. PRESENTED BY Introducing Oracle Functions Autonomous Platform auto-scales functions No servers to provision, manage Pay Per Use Pay for execution, not for idle time No Lock-in Built on open-source Fn Project and Docker Oracle Functions Functions-as-a-Service Oracle Cloud Integrated Container Native Multi-tenant Secure Open Source Engine 6
- 7. PRESENTED BY How Does it Work? Push container to registry Configure function trigger Code runs only when triggered Pay for code execution time only 7
- 8. PRESENTED BY Common Use Cases Web, Mobile, IoT Backends Real-time Stream, File Processing Batch ProcessingIntegrate and Extend SaaS 8
- 9. PRESENTED BY Example: Integrate and Extend SaaS ORACLE CLOUD ERP Cloud Analytics Visualize dataEnrich data and send to Analytics service Use data for analytics and insights Inventory create / update Service Cloud Trigger functionsSupport incident create / update Functions 9
- 10. PRESENTED BY Match made in cloud! • Cloud-native – Elastic • Container-native – Redis is most-dockerized OSS • High performant – In-memory – Write vs. read-only • Redis is a great choice for stateless scenarios – Instant data update & access – Asynchronous processing via pub-sub • Scaling Redis matters even more in Serverless Why Redis in Serverless? Master Replica Master Replica Master Replica
- 11. PRESENTED BY Let it be stateful! • “Stateful + Serverless” is an oxymoron? – Serverless ≄ Stateless – Serverless architecture is more natural for stateless applications – Similar to TCP over IP • Multiple function invocations need to be tied together as a “session” • Use cases – Stateful skills for digital assistants • Alexa/Google Assistant • Chatbot – Stateful APIs – Loosely-coupled transactional application (shopping cart) Why Stateful in Serverless
- 12. PRESENTED BY How to enable statefulness • Logically similar to server-side HTTP session – Session ID/token • Client-driven • Function-initiated – Session data • # of entries • Max size – Client-provided data (aka, cookies) • Slightly different from server-side HTTP session – Function does not keep session information – Function needs to externalize session tracking to external components such as database or Redis Stateful Serverless Architecture
- 13. PRESENTED BY Solution with Redis
- 14. PRESENTED BY ORACLE CLOUD Stateful Use Case: Web and Mobile Backends Identity Process dataTrigger functions Functions Web, Mobile Apps API Platform Mobile Data persisted Database 1 4 Loyalty program
- 15. PRESENTED BY How to enable statefulness with Redis • Data structure – Simple session data: Hash – Complex session data: Hash + list/set/hash – Multiple boolean session data : Hash + bitmap • Session initiation – Unique sessionid as the key to hash • Session expiration/termination – For session timeout: Redis TTL – For Termination: Delete session entries – Eviction policies • Volatile-ttl: when session timeout is in place • Noeviction or allkeys-lru: otherwise (more memory efficient) Redis for Stateful Serverless
- 16. PRESENTED BY Architectural considerations • Topology – Request routing • Primary for read/write function endpoints • Read-replicas nodes for read-only function endpoints – Redis cluster • To handle a large data set • To segregate session data from application data • HA & Scalability – Throttling or lack thereof (Function and/or Redis) – How to handle a surge in function requests (how to make Redis “cloud-elastic”) – High number of short, transient duration connections (horizontal scaling matters) – Persistent client connections for “hot function endpoints” • Cardinality – A separate session Hash for each function application comprising multiple endpoints Redis for Stateful Serverless (Cont.)
- 17. PRESENTED BY Redis configuration tuning • Should be optimized for highly chatty interactions – Every function invocation needs to read session data back (no session stickiness) – # of max connections: high – For high workload (tcp-backlog): high – Connection timeout (timeout): low – Client status (tcp-keepalive): high (for busy function endpoints) or low (otherwise) • Replication – Synchronous: For few write operations – Asynchronous Redis for Stateful Serverless (Cont.)
- 19. PRESENTED BY Redis can be complimentary to Serverless architecture • Can enhance stateless design • Can enable stateful design Redis can be added to Serverless architecture naturally for stateful use cases • Consider serverless-specific architectural implications • Leverage relevant Redis capabilities Summary
- 20. Thank you!
- 21. PRESENTED BY