LEARN. PRACTICE. GROW.Concept and mechanism
A Deployment manages a Pod template and revision progression. Deleting Pods without correcting the template can reproduce the same failure. Before rollback, inspect state, history, and previous-version compatibility with current configuration and data. ConfigMaps provide nonsecret configuration, but the consumption mechanism matters: environment-variable values are loaded at container creation and do not automatically change in the existing process. Controlled Pod replacement may be needed. Do not generalize that behavior to every projected volume or application with its own reload mechanism.
Guided application
The scheduler compares requests with allocatable capacity on eligible nodes. Low CPU usage does not imply room for new requests. Tolerations permit passing matching taint restrictions but do not guarantee placement or override affinity or resource shortages. For CPU-percentage HPA, requests provide the calculation baseline and the metrics pipeline provides usage. Check both when the value appears unknown. Keep readiness criteria aligned with serving ability and observe the service during scaling, replacement, and rollback. More replicas do not automatically fix a saturated external dependency.
A ConfigMap changes MODE, but the old process retains its previous value. Plan a controlled rollout and confirm the configuration observed by the new execution.
Common pitfalls
Using Pod deletion as a template fix; reducing requests without measurement; assuming a toleration selects a node; confusing image rollback with data reversal.
Related topics: Services, policies, and names · Volumes and data lifecycle
Desired configuration needs capacity and health criteria allowing it to run.
Reference: Deployments · CKA Kubernetes v1.35