A Docker Swarm Love Story

Presentation chronicles Adgorithmics decision making process and experience with selecting a container orchestration tool.

Transcript

1. A Docker Love Story You either die a monolithic codebase or live long enough to see yourself become microservices
2. About me Brian de Heus - Lead Engineer @ Adgorithmics Twitter: @briandotjp Website: brian.jp
3. I stole the container illustration from Container Co, a NZ company
4. Our monolithic infrastructure Client Message queue Application that does practically everything Persistent Storage
5. Suddenly we have customers
6. Our microservice infrastructure Client Message queue API Gateway Analytics Service Remote Sync Service Order Insertion Service Heuristic Event Service Persistent Storage
7. Problem #1 - How do we scale? gce.beefy.g1
8. Problem #2 - Where do we store our configs? v2.0.0 v2.0.0-a v2.0.0-b
9. Problem #3 - How do we balance our loads?
10. Don’t hate -- orchestrate
12. Mesosphere - what is it • We haven’t figured out what it is • Is DCOS Mesos? Is Mesos Mesosphere? Where is Jimmy Hoffa? Do we really need Marathon? • Dependency on Zookeeper
13. Mesosphere - Orchestration • Can run tasks, not just services • Master schedules containers to run on worker nodes • Configurable Readiness- and Health checks
14. Mesosphere - configuration • Set environment variables in the configuration • Mount as volumes
15. Mesosphere - load balancing • External dependencies only • Mesos DNS • Marathon-LB
16. Kubernetes • Run microservices before you can run microservices • It’s feature packed, has everything we want • Dependency on ETCD • Too much documentation scattered across their website
17. Kubernetes - Orchestration • Can run tasks as well as services • Scheduler / Controller-manager
18. Kubernetes - Load Balancing • Kube-proxy (node only) • External load Balancer for ingress traffic
19. Kubernetes - Configuration • Configmaps • Will be mounted, but no need to distribute across all worker nodes like with persistent volumes
20. Docker Swarm (mode) • Simple: docker swarm init • Has most things we want • Comes with built-in, undocumented Chaos Monkey™ • Little resources available for learning from other people their mistakes • Good documentation
21. Docker Swarm - Orchestration • Built-in, install docker and you’ll have swarm • Master node schedules containers on worker nodes (and the master node)
22. Docker Swarm - Configuration • Environment Variables
23. Docker Swarm - Load balancing • Built-in, across all nodes
25. Lessons #1 Containerize your application, not infrastructure
26. Lessons #2 Start small -- you’re probably not thinking small enough
27. Lessons #3 Have a contingency plan ready. Orchestration will fail.
28. Lessons #4 Stay away from hosted solutions.
29. Persistent Storage Consul Load Balancer Client Message Queue
30. Fin