AWS Summit - SFO - Day - 1 Notes

Written on April 18, 2017

AWS organized the AWS Summit at San Francisco on 18th and 19th of April. It was an eye-opener for me on the devOps progress and the adoption of cloud based work-flows. I had take some notes and thought of sharing it. This is the notes from Day 1 of the event. Hope it is of use to you!

Amazon Summit SFO logo

This is part 1 of the blog post. For notes from Day 2, please see the post AWS Summit - SFO - Day - 2 Notes.

The State of Serverless Computing

Serverless Computing

Built on function as a service. Building blocks are container, function as a service (FAAS) and then serverless computing

Use cases:

  • Static and dynamic web apps
  • Mobile apps
  • IoT
  • Big data
  • Chatbots
  • Media & log processing

Value prop for Lambda:
** Pay only when the function runs, scaling is automated. **.
Next step evolution of using AWS.

Occupy the cloud - Distributed Computing for the 99% talks about how AWS is bringing Cloud Computing to everyone.

  • 600 concurrent functions
  • Expedia 1.2 billion lambda requests per month as part of CI/CD pipeline

Autodesk case study - get the PPT

very good stats on how devOps can save time and money

  • SAM - serverless CI/CD

    Code -> AWS Codecommit -> AWS Codebuild -> AWS Codebuild -> AWS Cloudformation Code -> Commit -> Build-> Test ->Deploy to prod Code deploy pipeline

AWS Cloud X-Ray: debugging tool

Building Serverless Web Applications

@leozh - Solutions Architect 

Basic Lambda concepts

  • no servers to provision or manage - C#, NodeJs, Python, Java
  • scales with usage
  • never pay for idle
  • availability and fault tolerance built in

Core concepts:

  • Event driven
  • Continuous scaling
  • pay by usage

Use case: Create a thumbnail of a just updated image. Github example here:

New features:

  • Allows for creating dev/prod - by pulling in the latest
  • Key-value pairs (SAM)

Amazon API Gateway

  • authenticate and authorize
    • sigv4, oAuth, Amazon Cognito, custom auth ==> per method authorization
  • unify multiple microservices
  • DDoS protection / Scaling: Use cloudfront on top if API gateway, tie WAF rules, use Shield to protect origin
  • Stage variables have environment variables (like prod, dev). based on it, point to beta or prod end point.

Design patterns

  • Monolithic: don’t use!
    • One large lambda function. Easier to comprehend.
    • Single handler - so single one for GET/PUT/POST, etc.
  • Microservices: provides a lot of freedom to make changes and impact small sections.
    • Each method has one lambda function. one each for GET, PUT, POST
    • Easier to code and focus on one smaller component. Focus shifts to integration testing.
    • Use X-ray for debug, Use SAM to manage Lambda functions

« get image from PPT »

SAM is an extension of CloudFormation. CloudFormation is basically infrastructure as code. Supports JSON or YAML.

  • Claudia.js - nodejs for this purpose
  • Chalice - python

Code example at Github.

Amazon has also built a pipeline automation for CI/CD. Here’s the refarch:

Serverless Orghestration with AWS Step functions

Ability to connect multiple functions into a work-flow using GUI. Essentially, chaining Lambda or any kind of function to follow a strict logical ordering.

To learn about how to build app in the modern world, read this:

Coordination must have

  • scales out
  • doesn’t lose state
  • deals with errors/timeouts
  • easy to build out.

** AWS Step function - State machine as a service **

  • Define in JSON
  • Visualize in console
  • monitor executions

Sample state machine

State machine

7 states for lambda step function Task, choice, parallel, wait, fail, succeed, pass

Once the state is executing, you can go back and see the execution in UI. It will show the specific inputs, the states reached and final output.

DevOps Best Practice
If you build a JSON spec, then:
1. publish the spec
2. provide examples
3. create a LINT and publish that as well.

Building CI/CD Pipelines for Serverless Applications

Serverless computing space:

Typical work-flow:

Event Source -> Function (lambda) -> Services (anything)

Lambda supported on

  • Node.js
  • Python
  • Java
  • C#

e.g.: deploy code through slack or even Alexa

  • Cont Integration - anytime code is in main-line, it will built and it is valid
  • Cont Delivery - automatically build, test and after a manual approval, push to prod.
  • Cont Deployment - automatically build, test and push to prod - no manual gating.

Amazon Cloudformation is a good way to implement CI/CD. However, it can get verbose. An alternative is to use Serverless Application Model (SAM).

  • Have different accounts for Dev, Stage, and Prod
  • Having a single account with different stacks is ok for smaller team. For more details, refer AWS Organizations. Basically, control access to AWS environments by policy rather than trying to handle everything manually.

AWS Codebuild

Can be used as a managed build service. Pay by minute. Can provide docker images for build. cheaper than using an EC2 based jenkins or build pipeline since it is charged by hour.


Altassian has developed a local EC2 stack. Use it to test locally before moving to EC2. This is cool!

Deep Dive on CICD and Docker

  • John Pignata - Startup solutions architect


CI/CD Objective

How can we quickly and reliably deliver good ideas to our customers?

« Learnings slide, CI/CD slide - get it when published. »

ECR - docker repository in CI/CD scenario.

AWS CodeBuild

Build and test projects. How does the testing work? 4 phases:

  • install: get the right test libraries, etc
  • pre-build: log into the right AWS ECR etc
  • build: run the actual build process, SASS
  • post_build: build the final jar, clear up temp files, push to registry
Tag the code image with something that links back to development. Avoid using simplistic names like "LATEST", "PROD".

Amazon EC2 Container Service (Amazon ECS)

Managed service for setting up docker images.

  • Able to specify the min health required, maximum health. ECS will straddle the deployment based on these numbers.
  • Can even perform canary deploy
  • Blue/Green deployment: Having 2 different app versions (blue / green). Basically launch for short time, measure and stabilize or revert back.
    • Common way is with DNS swap.
    • New feature: Host based routing.

AWS CodePipeline

Provides ability to model and implement the full CI/CD pipeline. It can talk to external tools and can even integrate with Lambda functions.

Reference architecture:


Deep Dive: Log analytics with Amazon Elasticsearch Service

Amazon Elasticsearch is a managed service around Elastic Search. {Used by Expedia}

Use “Data Pattern” slide for sizing of the EL cluster.

Usage data source -> Amazon Kinesis Firehose -> Amazon Elasticsearch service -> Kibana

Take the best practice slide

Kinesis Firehose

send streaming data between applications.

  • serverless
  • error handling
  • s3 backup

Sign the requests on the fly using a local proxy aws-es-kibana