LEARN. PRACTICE. GROW.Concept and mechanism
Kubernetes maintains objects describing team intent. A controller observes differences and attempts to bring actual state toward desired state. If a Deployment requires three replicas, deleting a Pod does not change that intent; a replacement can appear. The new object need not retain the previous UID or address. The control plane coordinates state and placement while nodes execute workloads. A temporary API failure does not mean every process immediately stops, but it affects management and reconciliation. During an incident, distinguish control failure, execution failure, and functional application failure before choosing an intervention.
Guided application
A Pod groups containers sharing networking context. A process listening on localhost can communicate with another container in the same Pod without an external Service. This does not mean a container in another Pod shares that localhost. For reproducible releases, pin approved content by digest and verify registry availability; a tag can move. Namespaces organize names and support scoped controls, but are not themselves a firewall. At APS handover, record the responsible controller, artifact reference, namespace, and health signals. The team can then identify which configuration to change and which effects to observe.
If a deleted Pod reappears, inspect its controlling workload before repeating deletion. To reduce replicas, change intent through the approved process.
Common pitfalls
Replacement treated as the same Pod; healthy API treated as healthy application; tag treated as immutable content; namespace treated as complete isolation.
Related topics: Capacity, scheduling, and workload types · Services, DNS, and network policies
Observe the difference between declared intent, cluster state, and service outcome.
Reference: Controllers · KCNA current four-domain curriculum; edition date unconfirmed