Painful Serverless - Using AWS lambda as an http service - PyCON CA 2016

Serverless doesn’t mean anything.
It is a fancy marketing word from cloud vendors.
It means you don’t have to administrate servers.
Function as a service is a better description: FaaS!
Anyway, it promises low operational costs, infinite scalability and less management.
In reality, it strongly challenges the way you design, implement and deploy your application.
This talk presents one of the ways we use at http://dialogue.co

Transcript

1. Painful serverless Using AWS lambda as an http service Hadrien David
3. I am Hadrien! • @hadr • Software engineer • Montreal Python organizer • Python since 2008 • AWS since 2008 • Infra + APIs
5. AWS Lambda • Run code (js, py, java) • Pay compute time - (per 100ms chunk) • Zero server administration • Infinite scaling
6. Constraints • <100 parallel executions • 128M - 1.5GB mem • <300 seconds • Python 2.7
8. Use cases • Chat bots • Parallel computation • Data/Stream processing • HTTP Services
9. AWS API Gateway • HTTP front door • Pay as you go • 1000 req/s with burst up to 2000 • Automatic provisioning! “With a few clicks in the AWS Management Console…”
10. Clicks?
12. Chalice • “Python Serverless Microframework for AWS” • Flask-ish API • https://github.com/awslabs/chalice
15. What we do with it at Dialogue
16. Push notifications
17. HTTP API • Register device push token ◦ POST /devices • Send push to user ◦ POST /pushes
19. Components
20. Register device
21. Notify user
22. Batteries included
23. Summary API Gateway + Lambda can be a challenge but the effort is worth it.
24. References • Serverless Architectures - Mike Roberts • WTF is operations? - Charity Majors • Operational best practices #serverless - Charity Majors • The Twelve-Factor App
25. Thanks! Question? We’re hiring!