Architecture of scalable and resilient apps with GraphQL & event-driven serverless

Presents an architecture in which asynchronous serverless functions are triggered from an event system listening to database changes, where the functions execute business logic and update the database back with status and a GraphQL server streaming these changes back to client through subscriptions.

Using:
- [Hasura GraphQL Engine](https://github.com/hasura/graphql-engine)
- PostgreSQL on Amazon RDS
- AWS Lambda

Transcript

1. shahidh_k AWS Community Day Bangalore Architecture of scalable and resilient apps with GraphQL, Amazon RDS and AWS Lambda 1
2. shahidh_k 2 I’m Shahidh
3. shahidh_k The Problem 3
4. shahidh_k Traditional architecture App Database API Payment microservice Restaurant microservice Delivery microservice 4 1. Scalable? 2. Resilient?
5. shahidh_k Moving towards scale & resilience 5
6. shahidh_k Async architecture App Database API Event system Payment processing Restaurant approval Delivery assignment Async microservices 6
7. shahidh_k Serverless 7
8. shahidh_k Serverless
    $ cat index.js exports.function = (req, res) => { res.json({"message": "Hello World"}); } 8 $ serverless functions deploy hello_world --trigger-http https://serverless.functions.net/hello_world $ curl https://serverless.functions.net/hello_world {"message": "Hello World"}
9. shahidh_k Scalable! • No OPS • No VMs to manage • Don’t need to worry about scaling • Just write the code • Deploy • Scale for “free” Graphic Source 9
10. shahidh_k Resilient? • Serverless functions are ephemeral • No knowledge of state - async • Trigger them on events • Add retries to the mix • We have resiliency 10
11. shahidh_k Example: Food ordering app Column | Value -------------------+--------------------------------------- id | a705a43d-c55b-4a8a-ac16-80679bd02404 created_at | 2018-09-23 13:24:32.159024+00 user_id | eca786b6-73ae-44c7-b34f-1757af554464 is_validated | false is_paid | false is_approved | false is_agent_assigned | false 11
12. shahidh_k Order state machine with serverless is_validated | false is_paid | false is_approved | false is_agent_assigned | false is_validated | true is_paid | false is_approved | false is_agent_assigned | false is_validated | true is_paid | true is_approved | false is_agent_assigned | false is_validated | true is_paid | true is_approved | true is_agent_assigned | false is_validated | true is_paid | true is_approved | true is_agent_assigned | true Checks payment New order Validate order Restaurant approval Agent assignment Completed 12
13. shahidh_k Architecture Recap • Microservices became serverless functions • Serverless functions are scalable by nature • Database events trigger async serverless functions • Resiliency by event source tool instead of in the API code • Moved specific failure handling logic to generic failure handling 13
14. shahidh_k The cost 14 App Database API Event system Payment processing Restaurant approval Delivery assignment Generic Event System Event system Async Serverless Fns Payment processing Restaurant approval Delivery assignment
15. shahidh_k High level overview Database Event system Payment processing Restaurant approval Delivery assignment Async serverless functions A Database API layer Payment microservice Restaurant microservice Delivery microservice Synchronous logic Asynchronous logic 15
16. shahidh_k However, there’s still one problem
17. shahidh_k GraphQL 17
18. shahidh_k REST vs GraphQL Product Brand Category GET PUT POST PATCH DELETE GET PUT POST PATCH DELETE GET PUT POST PATCH DELETE ProductWithBrand 18
19. shahidh_k GraphQL changes the way we think about UI + API • UI First modelling • Makes API usage more natural • Maps UI components to GraphQL queries • Faster iterations - no back and forth b/n frontend-backend 19
20. shahidh_k GraphQL Subscriptions Consume async info from the backend without any worries It’s as if REST had a WATCH verb Datastore App 20
21. shahidh_k Let’s say I want to build... 21
22. shahidh_k Without GraphQL • Polling from the client • Websockets - custom API • Community fragmentation • Various patterns in different frameworks • More work for the developer 22
23. shahidh_k With GraphQL • Standardised spec • All GraphQL Clients are compatible with all GraphQL Servers • Client implementations available for all major platforms and frameworks • Subscriptions: access to async info • Less work for the developer 23
24. shahidh_k GraphQL Subscriptions App Datastore Process payment Validate order Agent assignme nt Restauran t approval New Order Complete 24
25. shahidh_k However, the cost is... 25 App Database API Event system Payment processing Restaurant approval Delivery assignment Generic Event System Event system Async Serverless Fns Payment processing Restaurant approval Delivery assignment GraphQL Server Subscriptions
26. shahidh_k Traditional vs GraphQL + Serverless Architecture 26
27. shahidh_k Traditional architecture App Database API layer Payment microservice Restaurant microservice Delivery microservice 27
28. shahidh_k Moving towards scale & resilience App Database API Event system Payment processing Restaurant approval Delivery assignment Async microservices 28
29. shahidh_k GraphQL & Serverless architecture 29 App Database Payment processing Restaurant approval Delivery assignment Mutation/Subscriptions Async serverless functions
30. shahidh_k Demo Time! 30
31. shahidh_k Food ordering app 31 https://serverless-demo-aws.hasura.app/order-app
32. shahidh_k Food ordering app - scalable 32 https://serverless-demo-aws.hasura.app/analytics-app
33. shahidh_k Food ordering app - resilient 33 https://serverless-demo-aws.hasura.app/console
34. shahidh_k 34 Thanks Questions?