Serverless
Beyond PaaS

Introduction

Tusa Viktor @ Xively (LogMeIn)

Twitter: @talien

Github: talien

Recap

• IaaS -> PaaS -> Serverless
• Request handling(API Gateway) +
  Business Logic (Lambda)
• Frontend heavy, backend is third party (Backend-as-a-Service), or very lightweight (Function-as-a-Service)

What happened
in the last year?

• Competitors (Azure, Google, IBM)
• Frameworks (serverless.io)
• API GW usage plans, more and
  more languages

Common offerings

• Pay what you use
• Event-driven, IoT focused
• Stateless execution context
• No operational maintenance
• Easy deployment

Azure Functions

• Tries to group your functions into a logical app
• Memory allocation is per-app not per-function
• Two modes: "classic" and "dynamic"
  • Classic is like a PaaS ans billed the same way
  • Dynamic is billed like AWS Lambda.
• Runs on top of Azure WebJobs.
• C#, F#, Node.JS, PHP, Python

IBM OpenWhisk

• Open Source runtime
• Completely stateless, very similar to Lambda
• Python, Node.js, Swift, Docker
  • Docker program should communicate through stdin and stdout in JSON.
• Similar limits as Lambda, however Lambda is more flexible
• Blocking and async invokation.

Google Cloud Functions

• Still in Alpha
• Only Node.js
• Very simple, very lightweight
• Similar to Lambda, less event trigger types, but no limit on execution duration

AWS Lambda

• Python, NodeJS, Java8
• Oldest, most integrated platform
• Granular IAM policies - per-endpoint access control on cloud resources
• API Gateway gives a full-fledged API frontend in front of Lambda.
• "cold start"

Lambda + API GW changes

• Python support (yay!)
• Increase execution duration limit to 5 min
• Scheduling
• VPC support
• Stage variables
• Usage Plans
  • Throttling, quotas and usage metrics on API key level

Frameworks

• Last year, lack of automation was the biggest problem
• This year: two different approach
  • Cloud Management: cloudformation (AWS), terraform 
  • Frameworks: Serverless, Zappa, Chalice, Apex, etc.
• Environment management and deployment included in most of the frameworks

Serverless

• Started as JAWS
• Now the most mature framework over AWS Lambda + API Gateway 
• Plan for support Google Cloud Functions, Azure Functions and OpenWhisk
• Extendable, large community

Serverless #2

• Written in Node.js, supports Node and Python
• 1.x is a total reimplementation with API breaks (as usual in the Node community)
• Highly dependent on CloudFormation.

Zappa

• Only supports Python, targets AWS
• Focuses on turning existing Django/Flask applications into serverless
• Super simple to use, no deep AWS knowledge needed

Chalice

• Focuses on Python
• Created by AWS
• Uses very similar syntax for routing like Flask, generates API GW resources from routing declaration during deployment.
• IAM policy auto-generation

Apex

• Node.js, Python, Java, Go only in AWS
• Functionality is similar to serverless, Go support is a crude hack. 
• Uses Terraform instead of Cloudformation
• Written in Go

Summary

• Frameworks are still in early stage (except for serverless)
• If we have a complex API, we need to deploy before we can try it out (losing the advantage of Docker).
• Still a bit hard to debug.
• But it still seems to be the future of Web and we are closing in.

Thank you!

Questions?