English
English posts
Kubernetes is not YAML first. It is boundary, state, and signal under pressure.
Mainline
KAI Road of Kubernetes
KAI Road of Kubernetes 01 — What a Kubernetes Pod is, and why it does not manage containers directly
Start with the core Kubernetes abstraction: what a Pod actually manages, and why Kubernetes schedules Pods instead of orchestrating containers directly.
Core idea A container is an execution unit. A Pod is an orchestration unit.
KAI Road of Kubernetes 02 — What a Deployment is, and why Pods should not be babysat manually
A Pod is not enough. A Deployment captures the desired replica count, Pod template, selector, and rollout behavior so Kubernetes can keep a workload converged.
Core idea A Pod is one landing of a workload. A Deployment is the rule that keeps the workload converged.
KAI Road of Kubernetes 03 — What a Service is, and why Pod IPs should not be your interface
Pods can be replaced by Deployments at any time. A Service gives a changing group of Pods a stable network entry point through DNS, ClusterIP, selectors, EndpointSlices, and ready backend endpoints.
Core idea Pods change and Pod IPs drift. A Service turns a group of Pods into a stable network entry point.
Production
Production notes
Concepts
Technical concepts
GitOps review checklist before Argo CD syncs production
A compact pre-sync checklist for GitOps changes so an innocent YAML edit does not become surprising production behavior in Kubernetes.
Core idea GitOps applies intent; it does not prove the rollout math is safe.
Kubernetes deployment guardrails: probes, rollout settings, and fast triage
A compact baseline for readiness, liveness, rollout settings, and fast triage before a Kubernetes deployment turns into an outage.
Core idea Readiness routes traffic; liveness decides restart. Treat them as separate contracts.
Agent observability in production: spans, tool events, and failure memory
Basic logging is not enough for agent systems in production; you need traces, tool-level outcomes, and a short path from failure back to prompt or policy.
Core idea Basic logs are not enough; agent systems need traces, tool events, and failure memory.