Kubernetes — A Year in Production
summary
A platform automates installing Kubernetes, Kafka and ELK across Azure and on‑prem data centers using a Terraform‑style infrastructure layer and a Python CLI and has been running in production for a year. It covers concrete production failures and fixes: DNS resolution issues breaking Kafka broker discovery on customer networks, cron jobs creating ~64k orphan pods by exhausting the Pod CIDR, painful rolling upgrades from Kubernetes 1.11 to 1.13 requiring intermediate steps, and Java GC/Out‑Of‑Memory behavior causing node outages, with described mitigations.
AI-written summary · about
chapters (12)
- 0:00Wprowadzenie i motywacja
- 2:51Dlaczego Kubernetes na produkcji
- 5:39Architektura platformy na Kubernetes
- 13:06Pierwsze problemy produkcyjne
- 17:40Problemy z DNS i siecią
- 18:22Bezpieczeństwo i aktualizacje
- 20:34Orphan jobs i limity zasobów
- 24:37Benchmarkowanie i hardening
- 27:06Problem z downem klastra
- 31:18Monitoring i rozwiązania
- 33:05Porady i wnioski końcowe
- 34:31Pytania i odpowiedzi
related talks
40mFrom dev to prod. One Year of Using Kubernetes in Production
35mKubernetes journey
37mRunning Kubernetes in a Manufacturing Line
27mRunning Kubernetes in Production A Million Ways to Crash Your Cluster
45mLearning How to Pronounce Kubernetes to Production in 3 Months!
20mKubernetes 101
40mThe Good, the Bad, and the Ugly of Managing Production Systems with Kubernetes
53mImplementing Kubernetes on with confidence