SkyDock is an open source solution for building & hosting Docker images at scale in the cloud.  Designed to solve real-world problems faced in traditional CI systems that limit throughput and hinder your ability to deliver software quickly and reliably. From a Skyscanner Engineering perspective this is a great opportunity to showcase some of the cutting edge work that we are doing with Docker, Ansible and AWS. It also highlights that we are solving problems at scale from a both a technical and organisational perspective.

1. SkyDock
How (and why) to roll your own Docker SaaS
Ryan Crawford & Grant Henderson
DevOps Meetup, May 2015

2. Who are we?
Squad of 4 engineers + 1 tech manager
Live within the Engineering Services “tribe”
Work across engineering squads in an enablement capacity
Release Engineering Services
We’re not Release Managers, and not Build Masters

3. What do we do?
Tooling
Own and operate tools that enable the
business to release software at scale
Swarm on release blockers & pain points
Provide adoption support for tools & tech
Focus on self-serviceability

4. What do we do?
Enablement
Coach Continuous Delivery best practices
Develop frameworks that empower teams
to adopt tech and deliver products
without reinventing the wheel
Provide adoption support for tools & tech
Focus on self-serviceability

5. What do we do?
R&D
Research emerging tech
Build proof-of-concepts to validate
Transition from PoC to wider rollout
Think about future adoption support &
self-serviceability

6. What Is SkyDock?
• It’s essentially your own DockerHub.com
• Scalable CI system and Registry for Docker images
• It’s mostly integration work, with a little innovation

7. Why bother when there is a DockerHub.com?
VCS is internal
• Want to avoid making this externally visible (susceptible to exploits)
• Want to avoid intellectual property hosted outside our network
IT Governance
• Need to retain control over user management

8. What about the on-premises version?
What about Docker Hub Enterprise?
• Not fully featured
• Licensing model is preventative at scale
• Cost is based on number of containers built, as well as the number
of containers you deploy
• Same applies for competitors like Quay.io

9. Why not just add Docker to TeamCity?
Unlimited Capacity
• Built on open source tech
• Elastically scaled pool of build agents
• The only cost is your cloud footprint
Deterministic Builds
• Short-lived ‘phoenix’ build agents
• Avoid cross-project environment corruption breaking your builds

10. Are there other reasons?
Scale & Resiliency
• Docker client registry is horizontally scalable
• Meta-data & artifacts persisted across AWS Availability Zones in S3
Developer Freedom
• The core build environment is simply Docker, and Docker abstracts
• Build with any software you like
• No need to request change from a centralised function

11. SkyDock Architecture

12. SkyDock Architecture

13. SkyDock Architecture

14. SkyDock Architecture

15. SkyDock Architecture

16. SkyDock Architecture

17. How We Build It
Build a hardware stack in AWS Deploy the software
Everything in Source Control
Everything Automated

18. Provisioning the Infrastructure
Automate everything – no manual changes through the AWS console
Wait for ~20 mins
(It takes a while to
create the RDS
database)

19. Deploying the Software Stack

20. Creating Baseline Jenkins Slave AMIs
1. Create an instance
2. Provision Instance3. Create AMI from Instance

21. From 0 to 60 in Two Commands
Step 1: Provision the required AWS infrastructure
run-playbook aws-skydock-cf.yml --verbose --extra-vars "skydock_stack_revision=phase01
skydock_search_db_user=******* skydock_search_db_pass=******* skydock_cf_sandbox=prod
build_number=1-0-5 aws_cf_access_key=******* aws_cf_secret_key=*******"
Step 2: Deploy and configure the software stack
aws-playbook -i inv-aws-prod aws-skydock-site.yml --verbose --extra-vars "skydock_search_db_user=*******
skydock_search_db_pass=******* skydock_registry_s3_access_key=******* skydock_registry_s3_secret_key=*******
docker_private_registry_internal_hostname=*******"

22. Demo Flow
1. Run an Ansible job to create the hardware tier using
CloudFormation.
2. Provision an “Ansible Jumpbox” in AWS to deploy the
application tier.
3. Run an Ansible job to provision the full application tier.

23. Demo
<< INSERT VIDEO HERE >>

24. Demo Summary
We just built and deployed:
• 1x Ansible jumpbox
• 3x Docker Registry web application servers
• 2x Docker Registry UI browsers
• 2x AWS elastic load balancers
• 1x AWS RDS MySQL database
• 1x Jenkins Master (capable of spinning up it’s own agents)
• A lot of security groups…

25. What Next?
Top 5 SkyDock TODOs
1. AWS – auto-scaling for the Registry (both clients & UI servers)
2. Jenkins Master – configure the Jenkins master at provisioning time using scripts and
artifacts from version control (remove requirement for manual actions)
3. Jenkins Resiliency – improved monitoring & backup / restore
4. Decentralise Jenkins – provide “turn-key” Jenkins instances (satellites) for individual
squads to use and customise instead of using the SkyDock Jenkins master
5. Registry – migrate to registry version 2.0 (and Docker 1.6)

26. thank you