Download as PDF, PPTX
Uploaded byDocker, Inc.
Docker on Docker
This document discusses Docker Inc's transition to running their infrastructure on Docker EE (Docker Enterprise Edition). Some key points: - Previously Docker Inc ran their infrastructure (Docker Hub, Docker Store, etc) on a collection of Docker hosts managed imperatively without orchestration. - They transitioned to running everything on Docker EE for the benefits of declarative orchestration using Docker Swarm, access controls, and resource management. - The transition was done by changing the orchestration layer only, without changing applications, containers, logs, metrics, etc. This allowed them to gradually transition services to Docker EE. - Running as a customer of their own product has provided Docker Inc valuable insights into best practices, feature requests
More Related Content
Docker on Docker
- 1. Docker on Docker Add picture here BrettInman Infrastructure Engineer Marcus Martins Senior Manager - Orchestration Docker Inc Docker Inc
- 2. Docker on Docker ●Where we were ● How we moved to Docker EE ● What we learned ● How we’re making Docker EE better
- 3. SaaS at Docker ●Docker Hub ● Docker Store ● Docker Cloud > 1B pulls a month > 500K builds a month > 5000 requests / sec
- 4. Philosophy ● High Availability ●Self-healing ● Self-service ● Build it - own it
- 5.
- 6. Previous Infrastructure Infra provides: ●Group of Docker Hosts per team ● Docker TLS Certs ● Centralized logging ● Centralized metrics ● Service discovery / routing
- 7. How dev usedthe old setup Dogfooding Docker since 2013
- 8. How dev usedthe old setup The good parts: ● Fully containerized ● Docker as the only interface ● No SSH access
- 9. How dev usedthe old setup The bad parts: ● Multiple deployments tools ● Imperative deployments
- 10. How dev usedthe old setup The bad parts: ● Hosts as pets ● Manual resource management
- 11.
- 12. Docker EE Add picture here ●Docker Swarm ● User Management ● Resource Access Control
- 13.
- 14.
- 15. Focus Changed: ● Orchestration Did notchange: ● Code ● Containers ● Logs ● Metrics ● Service Discovery ● Routing
- 16.
- 17. Self-service transition ● Goal:give service teams control ● Use both deployments at will ● Swarmkit mesh networking - magic, but: ○ Hard cutover transition for us ○ Can’t pull metrics from individual containers
- 18. Our Routing ● Registrator ●Consul ● Consul-template ● Haproxy
- 19. Host Port Publishing --publishmode=host Enables classic engine behavior of exposing container port on host for Swarm services
- 20. Automation ● Bootstrap lock ●Discovery - Load balancer or tags
- 21.
- 22. Resources Constraints Human isno longer the scheduler
- 24. Reservation = Limit docker service create --reserve-memory640m --limit-memory 640m
- 25. System Reservation docker service create --name system-reservation --reserve-memory 1G --limit-memory 1G --mode global --init ubuntu sleep infinity
- 26. Cloud Permissions ● IAMis at host level - not container ● Each team gets an autoscaling group ● Leverage UCP Collections
- 27.
- 28. Declarative version: '3.3' services: api: image:example/api:2.1.2 ports: - mode: host protocol: tcp target: 80 environment: DEBUG: False ENVIRONMENT: product ... ● Compose file ● Deploy with `docker stack deploy` ● All stacks defined in source control
- 29. Updates version: '3.2' services: api: image:example/api:2.1.2 deploy: replicas: 20 … update_config: delay: 10s parallelism: 2 order: stop-first max_failure_ratio: 0 failure_action: rollback ● Rolling updates ● Automated rollback
- 30. Access Control version:'3.2' services: api: image: example/api:2.1.2 deploy: replicas: 2 … labels: com.docker.ucp.access.label:/Api … ● Using Docker EE Collections
- 31. What’s better? Decouple hostfrom application ● Host replacement ● Host failures ● Build it own it - developers don’t own hosts
- 32.
- 33.
- 34. What else? ● Easyto scale application ● Standardization ● Docker-native
- 35. Where we arenow
- 36. Now ● All ofDocker SaaS is running on Docker EE ● 80 worker nodes ● 60 swarm services ● 1000 tasks ● Multiple deployments a day
- 37.
- 38. Best Practices Topics including: ●Managing Resources ● Provisioning and Automation ● Logging and Monitoring Coming soon to success.docker.com
- 39. Product guidance ● In-housecustomer at scale in production ● Architecture ● 35+ feature requests ● Host port publishing ● More soon!
- 40. Canary ● Running everyDocker EE Release Candidate ● 50+ bugs filed ● Customer Zero
- 41. Docker EE with Kubernetes inproduction
- 42.
- 43. Docker EE Hosted Demo Addpicture here docker.com/trial ● Free 4 Hour Demo ● No Servers Required ● Full Docker EE Cluster Access