You can't have a cluster*!$& without a cluster!
summary
Explains why Kubernetes’ built-in watch primitive and controller/scheduler model deliver real resilience while also adding measurable complexity and risk. Describes a production outage where removing a memory limit caused an auth service to fail within 30 seconds and a Solando rollback lesson: add CI/CD integration testing. Details a CVE that lets callers escalate from the API server to the kubelet for remote command injection, and shows how observability, conformance tests and automation actually fix these failures.
AI-written summary · about
chapters (9)
- 0:00Start
- 4:08What about the cloud?
- 4:20Modern Cloud Computing: Reselling electricity at a premium
- 5:08Modern Cloud Computing: The Stack
- 7:39Kubernetes: Management
- 7:54Kubernetes: Internals
- 10:00Kubernetes: API
- 13:05Once upon a time...
- 13:35What Happened?
related talks
35mKubernetes, Lessons Learnt
29mKubernetes Failure Stories and How to Crash Your Clusters
31mThe Myth of the Monocluster
33mKubernetes Security: Horror Stories to Help Secure Your Cluster
27mRunning Kubernetes in Production A Million Ways to Crash Your Cluster
44mPreventing Controller Sprawl From Taking Down Your Cluster - When a Scalable Pattern Stops Being Scalable
35mKubernetes Security - Challenge and Opportunity?
49mCluster Wrangling: How to make Kubernetes clusters secure and usable