Video and slides synchronized, mp3 and slide download available at URL http://bit.ly/2dljkbl. Adrian Trenaman provides detailed examples of how HBC Digital has used Docker and where container technology has paid off and, pragmatically, what aspects of the technology haven't panned out as they thought they would. Filmed at qconnewyork.com. Adrian Trenaman works as SVP Engineering at HBC Digital. He leads the engineering and infrastructure teams for Gilt in New York and Dublin. He is an experienced, outspoken software engineer, communicator and leader with over 20 years of experience working with teams throughout Europe, US and Asia in diverse industries such as financial services, telecoms, retail, and manufacturing.

1. How containers have panned out
Adrian Trenaman, Raconteur & SVP Engineering, Gilt / HBC Digital
Q-Con, New York, June 2016
@gilttech @adrian_trenaman @hbc_tech

2. InfoQ.com: News & Community Site
• 750,000 unique visitors/month
• Published in 4 languages (English, Chinese, Japanese and Brazilian
Portuguese)
• Post content from our QCon conferences
• News 15-20 / week
• Articles 3-4 / week
• Presentations (videos) 12-15 / week
• Interviews 2-3 / week
• Books 1 / month
Watch the video with slide
synchronization on InfoQ.com!
https://www.infoq.com/presentations/
hbc-containers

3. Presented at QCon New York
www.qconnewyork.com
Purpose of QCon
- to empower software development by facilitating the spread of
knowledge and innovation
Strategy
- practitioner-driven conference designed for YOU: influencers of
change and innovation in your teams
- speakers and topics driving the evolution and innovation
- connecting and catalyzing the influencers and innovators
Highlights
- attended by more than 12,000 delegates since 2007
- held in 9 cities worldwide

4. “What competitive advantage did
containers give you?”

5. Gilt: luxury designer brands at discounted prices

6. we shoot the product in our studios

7. we receive, store, pick, pack and ship...

8. we sell every day at noon...

9. stampede...

10. this is what the stampede really looks like...

12. m > n
This is fundamentally a packing problem.
We have n machines, and we have m services to deploy.

13. 1
It’s also an isolation problem
Any given service / team / engineer shouldn’t be able to take out someone else’s
work in production.

15. It’s also an impedance mismatch problem.
Developers often think of machines as something that’s all theirs, magically
provided by the hardware fairy.

16. LXC
Leveraging LXC in Tokyo for Gilt Japan

17. Rack 1
Load Balancer
DB (CLXC
)
Email
Email
16xCPU, 128GB RAM,
900GB Disk.
Ubuntu 12.04 (→ 16.04)
~220 CLXC
in total.
Rack 1
Load Balancer
20-40 CLXC
20-40 CLXC
20-40 CLXC
20-40 CLXC
20-40 CLXC
20-40 CLXC
DB (CLXC
)
Email
Email

18. ✔ Scalable, performant use of machine resources.
✔ Solves the impedance mismatch: developers see ‘a machine’
✔ Limits the damage a single engineer can do.
✔ Infra/Devops engineer embedded into a tightly knit engineering team
❌ Static infrastructure
❌ Potential for resource hogging
LXC @ Gilt Japan

19. Immutable
Deployment
With
Docker

20. Prod
Core idea #1: dark canaries, canaries, release, roll-back.
Dark Canary
1.0.0
Instance_0
1.0.0
Instance_1
1.0.0
Instance_n
1.0.0
Dark Canary
1.0.1
Canary
1.0.1
Instance_0
1.0.1
Instance_1
1.0.1
Instance_2
1.0.1

21. Core idea #2: One container per host / EC2 instance
<<EC2 Instance>>
docker
<<container>>
Docker
registry

22. ION-Roller - https://github.com/gilt/ionroller
ION-Roller
(orchestrates
everything)
Elastic Load
Balancer (ELB)
Auto Scaling
Group (ASG)
Instance_0 - v1.0.0
Instance_1 - v1.0.0
Instance_2 - v1.0.0
Instance_0 - v1.0.1
Instance_1 - v1.0.1
Instance_2 - v1.0.1
Auto Scaling
Group (ASG)
Docker
registry

23. ✔ Immutable deployment :)
✔ DNS + ELB traffic migration :)
❌ Slow to set up / tear down environments :(
❌ Potentially expensive under continuous deployment :(
❌ Open-source, but in-house. ‘A snowflake in the making’ ❅
ION-Roller deployment:

25. 6
“We could solve this now, or, just wait six months, and Amazon will provide a
solution”
Andrey Kartashov, Distinguished Engineer, Gilt.

26. github.com/gilt/nova- deployment patterns
Instance_0 - v1.0.0
Instance_1 - v1.0.0
Instance_2 - v1.0.0
Live Traffic
Instance_3 - v1.0.0 Canary
Instance_4 - v1.0.0
Dark
Canary
Elastic Load
Balancer (ELB)
http://hello-world-nova.common.giltaws.com
Elastic Load
Balancer (ELB)
http://hello-world-nova-dark.common.giltaws.com

27. github.com/gilt/nova - creating environments
nova.yml
templates
$> nova stack create production
CloudFormation
CodeDeploy

29. github.com/gilt/nova- deployment
Instance_0 - v1.0.0
Instance_1 - v1.0.0
Instance_2 - v1.0.0
Live Traffic
Instance_3 - v1.0.0 Canary
Instance_4 - v1.0.0
Dark
Canary
Elastic Load
Balancer (ELB)
live
Elastic Load
Balancer (ELB)
dark
$> nova deploy common DarkCanary
1.0.1
Instance_4 - v1.0.1
$> nova deploy common Canary 1.0.1
Instance_3 - v1.0.1
$> nova deploy common Production
1.0.1
Instance_0 - v1.0.1
Instance_1 - v1.0.1
Instance_2 - v1.0.1
CodeDeployS3bundle

30. ✔ No docker registry (shock! gasp!) :)
✔ Less boilerplate code :)
✔ Immutable deployment (on mutable infrastructure) :)
✔ Leverage AWS tooling :)
？ Next up? Integrate with Code Pipeline :?
Nova deployment:

31. Fighting bit rot, chaos-monkey style
With long running mutable AMIs, it’s possible for bit-rot to creep in.
Think security vulnerability.
Novel approach: every day, kill and restart your oldest AMI randomly.
✔ Pick up latest AMI with fixes
✔ Fail early, noisily and loudly if there’s a problem without a production outage.
Vulnerability in container? Cut a new release against a fixed base-image.

32. Explorations in ECS

33. Sundial - running batch jobs with Docker & ECS
✔ Job dependencies (allows us to break large jobs into smaller jobs)
✔ Ease of viewing logs and debugging failures
✔ Automatic rescheduling of failed tasks within a job
✔ Isolation between jobs
✔ Low cost of setup and maintenance, as few moving parts as possible for Infra
teams to manage
http://github.com/gilt/sundial

34. Sundial: processes
A process in Sundial is a grouping of tasks (jobs) with dependencies between
them.
Schedule: Either manually triggered, continuous schedule, or cron schedule
Overlap strategy: if previous iteration hasn’t completed, do we
Wait
Terminate previous iteration
Run in parallel
When a process kicks off, all tasks with no dependencies kick off.
When a task finishes, any tasks blocked by that task will kick off.

36. ECS is getting really attractive...
We’re prototyping using for customer-facing services on our mobile team:
✔ Less configuration / moving parts than MST/Nova
✔ Automatic rollout
✔ Easy integration with IAM, CloudWatch, ECR
But:
❌ IAM roles at instance level not container level
❌ Tension between CF stack templates and deployment updates
❌ ELBs require fixed ports: we want to define the listening port.

37. Docker as Build Platform

38. Using docker as a
local build platform
The problem: keeping up
with different versions /
combinations of build
tools is crazy hard.
Why not use Docker for
build, using a versioned
build container?
docker-machine
Build Container
docker

39. Lesson #1
Containers have let us separate what we deploy (JVM, RoR, …)
from how and where we deploy it (mst, nova, EC2, Triton)
and This Is Good.

40. Lesson #2
It’s still a wild-west in terms of how containers are deployed.
Different teams have different needs - be sensitive to that.

41. Lesson #3
Seek immutability in the container, not in the stack.

42. Lesson #4
The competitive advantage: containers let us deploy quickly, frequently and safely
to production, which help us innovate faster.
That’s it.

43. #thanks @adrian_trenaman
@gilttech @hbc_tech

44. Watch the video with slide synchronization
on InfoQ.com!
https://www.infoq.com/presentations/hbc-
containers