Scaling Organizations Using Kubernetes
summary
Organization grows from 5 to ~250 engineers and doubles microservices yearly, spiking to tens of thousands of instances across 50 data centers and the default process-heavy ops model breaks. Kubernetes becomes a distributed OS: teams build Geronimo (packages + automated deploy/rollback via the Kubernetes API, stores cryptographic hashes in annotations, uses namespaces for atomic upgrades) and ticklish (a yum-like system that composes systems and resolves cross-service dependencies), add immutable PXE-boot images, CI/CD with metric-driven auto-rollback, and reduce cross-team communication from N^2 to manageable levels.
AI-written summary · about
chapters (35)
- 0:00Intro
- 1:32Kubernetes Architecture
- 1:49The Wonderful Things We Do
- 2:07Our Story
- 2:30Customer Demand Spike
- 3:09Customer Demand Spike 2
- 5:13One Year Later
- 5:55New Hires
- 6:17The Problem
- 6:45Mathematical Solution
- 8:21Why is that important
- 8:43Nonlinear relationship
- 9:27Velocity
- 9:47Port Infrastructure
- 10:14Organizational Units
- 10:40Puppet
- 11:22DevOps
- 12:04Puppet Modules
- 13:10Scaling Our Work
- 13:24Is It Enough
- 13:57Distributed Operating System
- 14:47Microservices
- 15:20Cuddle
- 15:49Estrace
- 16:38Deploying New Software
- 17:33Geronimo
- 18:22Cube API
- 19:51Group Install
- 22:07Configuration Deployment
- 23:01Auto Deploy
- 23:38Self Deploy
- 24:31Provisioning
- 25:39Continuous Deployment
- 28:20The Platform
- 29:51Data Centers
related talks
35mKubernetes journey
03mFrom 5 to 1,300+ Clusters: Declarative Scaling for Private Kubernetes
16mRunning Services at Google Scale
36mOptimizing Resource Usage in Kubernetes
33mScaling k8s Deployments: Achieving Elite DORA Metrics
31mThe 20 Year Platform – bringing together Kubernetes, 12-Factor and Functions
21mPlatform Perseverance: Taming 1,000 Kubernetes Clusters
36mUsing K8s to Power Microservices at Scale: The GitHub Way