The Good, the Bad, and the Ugly of Managing Production Systems with Kubernetes
summary
Concrete failures and fixes from dozens of Chinese enterprise deployments: 70% on‑premise, 30% on Alibaba Cloud; an Ansible offline installer and local Docker registry for air‑gapped sites; logging alternatives that keep file output and integrate with Fluentd; monitoring moved to Prometheus + OpenTSDB with an alert manager to avoid single points of failure; a highly available ingress using keepalived + HAProxy/nginx and an ingress‑claim sharding model to prevent reload storms; Cyclone CI/CD, Harbor registry, and TensorFlow-on-Kubernetes examples.
AI-written summary · about
chapters (20)
- 0:00Intro
- 1:48Our customers
- 2:27Kubernetes in China
- 3:28Positives
- 4:57Negatives
- 5:51Deployment
- 7:09Integration
- 8:07Logging
- 10:22Enhancements
- 10:40Empty Directory
- 12:05Monitoring
- 14:40Rationale
- 15:36External Load Balancing
- 18:52HiAvailable Ingress
- 24:36Scheduling
- 25:38Ingress Claims
- 26:15Ingress Claim Controller
- 27:31High Availability Master
- 28:49CICD Cyclone
- 37:05Summary
related talks
42mMastering Kubernetes: Challenges, Strategies, and Best Practices
22mAnatomy of a production grade Kubernetes cluster
27mRunning Kubernetes in Production A Million Ways to Crash Your Cluster
41mKubernetes — A Year in Production
42mKubernetes on-premises
27mA Safari of Kubernetes and its Natural Habitat
35mNo More Chaos: Audit and Inspect Kubernetes at Scale
37mRunning Kubernetes in a Manufacturing Line