Containers in Production: It’s Like Orchestrating Cats
summary
Company runs ~50 services across hundreds of machines and spends six months to roll the first container to production. Uses multistage builds to shrink a .NET image from ~2 GB to 180 MB (1/10 the size). Chooses Kubernetes over Docker Swarm/ECS for multi‑cloud portability and plans to use managed clusters; mentions Helm for package charts. Moves logging off disk to logstash over TCP, cutting CPU parsing by >50%. Warns about metric cardinality (Graphite/whisper allocates ~80 MB per metric) and avoids big stateful stores (15 TB ES) on containers.
AI-written summary · about
chapters (24)
- 0:00Start
- 3:54How To Build a Docker Container
- 4:16Docker File
- 5:48Multistage Builds
- 12:15Three Ways To Run Things on Kubernetes
- 13:07Deployment Files
- 15:03Health Checks
- 21:15Helm for Kubernetes
- 21:40Helm
- 24:20Docker Swarm
- 24:31Docker Compose Files
- 25:55Managed Services
- 26:24Azure-Kubernetes
- 27:12Logging Changes
- 29:37Where Does the Log Shipper Sit
- 31:09Do You Put the Log Shippers inside every Container
- 31:42Log Stash on every Host
- 37:39Monitoring
- 39:33How Do You Collect Metrics
- 40:06Metrics Application
- 53:31Auto Scaling
- 54:53Managing State
- 59:01Stateful Sets
- 1:01:52Questions
related talks
43mContainers Change Everything
38mThe Container Operator’s Manual
46mHow Box Runs Containers in Production with Kubernetes
08mKeynote: Cloud Native in the Enterprise: Production, Best Practice and Innovation
46mAround & After Kubernetes: The Principles and Ideas that Guide Us
38mThe Container Operator's Manual
29mFrom EC2 to Kubernetes Takeaways from a Brownfield Migration
50mJourneys To Cloud Native Architecture: Sun, Sea & Emergencies