Lambda @ Hooroo

The what, the why, the how, the challenges, and the learnings of AWS Lambda. A real world example of how it has gone from a 10% time project to production at Hooroo.

Transcript

1. λ @ Hooroo Jacob Evans & Jackie Camomot Qantas Developer Meetup Sydney, 27 July 2017
2. This page is intentionally left blank
3. Jacob Evans Developer, Hooroo Github: dekz jacob@hooroo.com Jackie Camomot Developer, Hooroo Github: jacki3lene jackie@hooroo.com
4. λ a.k.a Serverless
5. What is it? - A bit of code - Triggered by some something happening (an event) - Executed in the cloud (no dedicated server)
6. What else is there? FaaS BaaS Serverless Cloud Functions PaaS Microservices Lambda Openwhisk Azure Functions Firebase Parse Google Functions FaunaDB Event Driven MBaaS
7. Why should I use it? - You do not care about the servers. - You do not interact with the OS. - You do not think about capacity. - You do not think about scale. - There are very few dials to tweak. - Pay for fulfillment, not provisioning. - It is composable.
8. How do I use it? - You provide your Node.js, Java, or Python code - Configure: - Event sources (what triggered me?) - Permissions (what things can I access?) - Handler function (what should I invoke when it happens?) - Memory (how much gusto should it have?)
9. λ @ Hooroo: Booking Alerts
11. As Customer Support, I need a notification when a booking failed so that I can resolve the issue with the customer.
14. We can’t use Splunk because - It needs to be real time - Complicated and resource consuming query - Dynamic data in the email content - Event logs are truncated - Splunk is difficult to test - Conditional complexity
15. What about a service? - Bookings don’t fail often - Infrastructure would sit there idle - Increased setup time - Need to choose - Instance size - Auto scale groups - Auto scale conditions - AMI
18. Amazon SES Bookings publishes an event Customer Support Booking Alerts in AWS Lambda λ Sends an email when event type is `booking_failed`
19. Challenges
21. Deployment and Secure Config
23. - Offline development (on the plane over here) - Development environment (tail -f, hot code reload) - Frameworks (active development; things break) And more challenges ...
24. Learnings & Protips - Write your function code as stateless, keep lambda at the boundary of your core code - Function code should do only one thing (be simple) - Execution time is limited to 5 mins - Cold starts (sometimes seconds)
25. Pets Cattle Rabbits Microbes
26. - Reacting to something (an event) - Infrequent events - Async - 100’s of milliseconds (i.e not High Frequency Trading - Max 5 minutes (i.e no Video Encoding) - Connecting things together (AWS Step functions, IFTTT) - Scale quickly (1000’s of headless tests - Chromeless) Niche
28. Q & A
29. Thanks Questions: jackie@hooroo.com | jacob@hooroo.com Complaints: no-reply@hooroo.com