Today’s cutting-edge companies have software release cycles measured in days instead of months. This agility is enabled by the DevOps practice of continuous delivery, which automates building, testing, and deploying all code changes. This automation helps you catch bugs sooner and accelerates developer productivity. In this session, we’ll share the processes that Amazon’s engineers use to practice DevOps and discuss how you can bring these processes to your company by using a new set of AWS tools (AWS CodeCommit, AWS CodePipeline, and AWS CodeDeploy). These services were inspired by Amazon's own internal developer tools and DevOps culture.

1. © 2016, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
James Saull, Solutions Architect – Amazon Web Services
Emerson Loureiro, Senior Software Engineer - Gilt Groupe
July 2016
DevOps on AWS:
Deep Dive on Continuous Delivery
and the AWS Developer Tools

2. https://secure.flickr.com/photos/mgifford/4525333972
Why are we
here today?

3. Software moves faster today
Software creation and distribution is
easier and faster than ever:
• Startups can now take on giants with little to
no funding ahead of time
• Getting your software into the hands of
millions is a download away
• Your ability to move fast is paramount to your
ability to fight off disruption

4. What tools do you need to move fast?
Releasing software in this new software driven world
requires a number of things:
• Tools to manage the flow of your software development
release process
• Tools to properly test and inspect your code for defects
and potential issues
• Tools to deploy your applications

5. First, we need to
understand a little bit about
software release processes
https://www.flickr.com/photos/jurvetson/5201796697/

6. Release processes levels
Source Build Test Production
Continuous integration
Continuous delivery
Continuous deployment

7. Continuous Delivery Benefits
Improve developer
productivity
Find and address
bugs quickly
Deliver updates fasterAutomate the software
release process

8. A look back at
development at
Amazon..
https://secure.flickr.com/photos/pixelthing/15806918992/

9. 2001
Development transformation at Amazon: 2001-2009
2009
monolithic
application + teams
microservices + 2 pizza teams

10. In 2009, we
ran a study to
find out where
inefficiencies
might still exist

11. We were just waiting.
WaitWrite
Code WaitBuild
Code WaitDeploy
to Test
Deploy
to
Prod
Weeks
Mins Days Mins Days Mins Days Mins

12. Automated actions and
transitions; from check-
in to production
Development benefits:
• Faster
• Safer
• Simplification &
standardization
• Visualization of the
process
Pipelines

13. This has continued to work out really well:
In 2014:
• Thousands of service teams across Amazon
• Building microservices
• Practicing continuous delivery
• Many environments (staging, beta, production)
50 million deploys

14. Continuous delivery service for fast and
reliable application updates
Model and visualize your software release
process
Builds, tests, and deploys your code every time
there is a code change
Integrates with 3rd party tools and AWS
AWS CodePipeline

15. Source
Source
GitHub
Build
JenkinsOnEC2
Jenkins
Deploy
JavaApp
Elastic Beanstalk
Pipeline
Stage
Action
Transition
CodePipeline
MyApplication

16. Source
Source
GitHub
Build
JenkinsOnEC2
Jenkins
Deploy
JavaApp
Elastic Beanstalk
NotifyDevelopers
Lambda
CodePipeline
MyApplication
Parallel actions

17. Source
Source
GitHub
Build
JenkinsOnEC2
Jenkins
Deploy
JavaApp
Elastic Beanstalk
NotifyDevelopers
Lambda
TestAPI
Runscope
CodePipeline
MyApplication
Sequential actions

18. 8. Retrieve build artifact
EC2 instance
CodePipeline
Source
Source
GitHub
Build
JenkinsOnEC2
Jenkins
Deploy
JavaApp
Elastic Beanstalk
Source Artifact
S3
Build Artifact
S3
5. Get source artifact
1. Get Changes
6. Store build artifact
3. Poll for Job
4. Acknowledge Job
7. Put Success
9. Deploy build artifact
Elastic Beanstalk
Web container
Java App
MyApplication

19. We have a strong partner list, and it’s growing
Source Build Test Deploy

20. AWS service integrations
Source Invoke Logic Deploy
AWS Elastic BeanstalkAmazon S3
AWS CodeDeploy
AWS Lambda
AWS CodeCommit

21. DEMO
https://www.flickr.com/photos/seattlemunicipalarchives/12504672623/

22. Deploying your
applications
https://secure.flickr.com/photos/simononly/15386966677

23. Automates code deployments to any instance
Handles the complexity of updating your
applications
Avoid downtime during application deployment
Deploy to Amazon EC2 or on-premises
servers, in any language and on any operating
system
Integrates with 3rd party tools and AWS
AWS CodeDeploy

24. appspec.yml Example
version: 0.0
os: linux
files:
- source: /
destination: /var/www/html
permissions:
- object: /var/www/html
pattern: “*.html”
owner: root
group: root
mode: 755
hooks:
ApplicationStop:
- location: scripts/deregister_from_elb.sh
BeforeInstall:
- location: scripts/install_dependencies.sh
ApplicationStart:
- location: scripts/start_httpd.sh
ValidateService:
- location: scripts/test_site.sh
- location: scripts/register_with_elb.sh

25. appspec.yml Example
version: 0.0
os: linux
files:
- source: /
destination: /var/www/html
permissions:
- object: /var/www/html
pattern: “*.html”
owner: root
group: root
mode: 755
hooks:
ApplicationStop:
- location: scripts/deregister_from_elb.sh
BeforeInstall:
- location: scripts/install_dependencies.sh
ApplicationStart:
- location: scripts/start_httpd.sh
ValidateService:
- location: scripts/test_site.sh
- location: scripts/register_with_elb.sh
• Remove/add instance to ELB
• Install dependency packages
• Start Apache
• Confirm successful deploy
• More!
• Send application files to one
directory and configuration
files to another
• Set specific permissions on
specific directories & files

26. v2 v2 v2 v2 v2 v2
one at a time
half at a time
all at once
v2 v2 v2 v1 v1 v1
v2 v1 v1 v1 v1 v1 Agent Agent
Dev Deployment group
OR
Prod Deployment group
Agent
AgentAgent
Agent Agent
Agent
Choose Deployment Speed and Group

27. https://www.flickr.com/photos/spacex/16510243060/
Launching to
Production

28. Launching to Production
After you’ve built and tested your code and
hopefully gone through a few pre-production
deploys, its time for the real thing!
You’ll want think about:
• Impact to customers
• Impact to infrastructure
• Impact to business
How can we track these and communicate
deploys?

29. Extend AWS CodePipeline Using Custom Actions
Update tickets Provision resources
Update dashboards Send notifications Security scan
Mobile testing

30. Source
Source
GitHub
Build
JenkinsOnEC2
Jenkins
Deploy
Action
Custom Action
JavaApp
Elastic Beanstalk
MyApplication
CodePipeline
Deploy
JavaApp
Elastic Beanstalk Job Worker
3. Perform Job
1. Poll for Job
2. Acknowledge Job
4. Put Success
Amazon
EC2
With custom actions,
the job worker drives the interaction
between AWS CodePipeline
and other applications or services

31. Source
Source
GitHub
Build
JenkinsOnEC2
Jenkins
Deploy
MyAction
AWS Lambda
JavaApp
Elastic Beanstalk
2. Perform Job
1. Invoke Lambda function
3. PutJobSuccessResult
MyApplication
CodePipeline With AWS Lambda-based actions,
AWS CodePipeline
drives the integration with Lambda,
which then connects with other
applications or services
AWS
Lambda

32. Micro-services
Deployment at
Gilt with AWS
Emerson Loureiro,
Senior Software
Engineer, Gilt

33. Gilt...
●Sales go live everyday at
noon

34. Micro-services

35. Move fast

36. Phased rollout

37. Rollback just as fast

38. Deployment tooling is important for us...
Can we do it with AWS?

39. 1st Generation (DiscoDeploy)
● SBT plugin
awsAccountName: gilt-xxxxxxx
instanceType: m3.xlarge
servicePort: 9000
instanceCount: 2
jvm:
- -Dnewrelic.config.file=/opt/gilt/service/conf/newrelic.yml
- -Dnewrelic.config.agent_enabled=true
- -Dlogger.resource=logback.production.xml
- -Dconfig.resource=application.production.conf
- -Xms12G
- -Xmx12G
- -XX:NewSize=6G
- -XX:MaxNewSize=6G
...

40. 1st Generation (DiscoDeploy)
Docker
Registry
ASG
EC2 EC2 EC2 EC2
ELBsbt deploy-setup
sbt release-docker
sbt deploy-service
sbt deploy-scale

41. ●Good to begin with
1st Generation (DiscoDeploy)
Docker
Registry
ASG
EC2 EC2 EC2 EC2
ELB

42. 2nd Generation (Ionblaster)
●Based on CloudFormation
[
{
"service": "SERVICE",
"domain": "SERVICE.gilt.com",
"instances": { "min": 3, "max": 3 },
"type": "t2.small"
"run": {
"image": "gilt/SERVICE",
"cmd": ["-Dhttp.port=9000"],
"args": ["-p", "9000:9000"]
},
"health_path": "_internal_/healthcheck"
...
}
]

43. ●Based on CloudFormation
sbt docker:publish
ionblaster new SERVICE 0.0.1
2nd Generation (Ionblaster)
0.0.1
DNS

44. 2nd Generation (Ionblaster)
●Based on CloudFormation
sbt docker:publish
ionblaster new SERVICE 0.0.2
ionblaster traffic SERVICE 0.0.2 10
0.0.1
DNS
0.0.2

45. 2nd Generation (Ionblaster)
●Based on CloudFormation
ionblaster traffic SERVICE 0.0.2 100
0.0.1
DNS
0.0.2

46. 2nd Generation (Ionblaster)
●Based on CloudFormation
ionblaster kill SERVICE 0.0.1 DNS
0.0.2

47. ●Traffic migration could be better
●No logical grouping among stacks
2nd Generation (Ionblaster)
DNS
0.0.2

48. ●Based on Elastic Beanstalk
ionroller release SERVICE 0.0.1
3rd Generation (Ionroller)
0.0.1
DNS
Application

49. ●Based on Elastic Beanstalk
ionroller release SERVICE 0.0.2
3rd Generation (Ionroller)
0.0.1
DNS
0.0.2
Application

50. ●Based on Elastic Beanstalk
ionroller release SERVICE 0.0.2
3rd Generation (Ionroller)
0.0.1
DNS
0.0.2
Application

51. ●Based on Elastic Beanstalk
ionroller release SERVICE 0.0.2
3rd Generation (Ionroller)
0.0.1
DNS
0.0.2
Application

52. ●Based on Elastic Beanstalk
ionroller release SERVICE 0.0.2
3rd Generation (Ionroller) DNS
0.0.2
Application

53. What we didn’t like about those...

54. “We could solve this now, or, just wait six
months, and Amazon will provide a solution.”
Andrey Kartashov, Distinguished Engineer,
Gilt

55. ● Based on CodeDeploy + CloudFormation + AWS CLI (+ ECR)
make create stack
Development Canary
4th Generation (Mobile Service Tools - MST)
Production

56. ● Based on CodeDeploy + CloudFormation + AWS CLI (+ ECR)
make publish
make deploy-development
make deploy-production-canary
make deploy-production
Development Canary
4th Generation (Mobile Service Tools - MST)
Production

57. ● Based on CodeDeploy + CloudFormation + AWS CLI (+ ECR)
make deploy-production
Development Canary
4th Generation (Mobile Service Tools - MST)
Production

58. ● Based on CodeDeploy + CloudFormation + AWS CLI (+ ECR)
make deploy-production
Development Canary
4th Generation (Mobile Service Tools - MST)
Production

59. ●Mostly AWS tooling
●Quick deployments
Development Canary
4th Generation (Mobile Service Tools - MST)
Production

60. ● Boilerplate scripts
● Docker registry problems
● No cross-account support
● Shared CloudFormation template
Development Canary
4th Generation (Mobile Service Tools - MST)
Production

61. ● Boilerplate scripts
● Docker registry problems
● No cross-account support
● Shared CloudFormation template
4th Generation (Mobile Service Tools - MST)
Base
template

62. ●Based on CodeDeploy + CloudFormation + S3
nova stack create ENVIRONMENT
Development Canary
5th Generation (Nova)
Production
Environment

63. ●Based on CodeDeploy + CloudFormation + S3
nova deploy ENVIRONMENT development
nova deploy ENVIRONMENT canary
nova deploy ENVIRONMENT production
Development Canary
5th Generation (Nova)
Production
Environment

64. ●Versioned CloudFormation templates
5th Generation (Nova)
V2V1

65. Happiness

67. FIN, ACK
We’ve seen a quick run through today of the benefits of
continuous delivery on our software release process:
• Continuous integration (build/test) helps shrink our
feedback loop greatly
• We can get our software out in front of our users much
more rapidly
• By moving faster we can actually ensure better quality
• CodePipeline allows for integration with almost any
service or tool you can think of!
• Plus visualization of what’s going on!

68. Resources to learn more:
• DevOps: https://aws.amazon.com/devops/
• Continuous integration: https://aws.amazon.com/devops/continuous-integration/
• Continuous delivery: https://aws.amazon.com/devops/continuous-delivery/
• CodePipeline
• https://aws.amazon.com/codepipeline/
• https://aws.amazon.com/documentation/codepipeline/
• CodeDeploy
• https://aws.amazon.com/codedeploy/
• https://aws.amazon.com/documentation/codedeploy/
• https://github.com/awslabs/aws-codedeploy-samples
• Code Services Starter Kit: http://bit.ly/AWSCodeStarterKit
But wait, there’s more!

69. Please remember to rate this
session under My Agenda on
awssummit.london

70. Thank you!