Kubernetes nodes reserve a portion of their total memory for system processes and the kubelet. The ‘allocatable’ memory, which is what is available for Pods, is typically less than the total physical memory. A request for 500Mi might fail if the allocatable memory on any node is below this threshold, even if the total is 1Gi.

The correct command is `kubectl logs -c `. The `-c` flag specifies the container name within the Pod. Option A would fail or show logs from an arbitrary container if the Pod is multi-container. Option B uses incorrect syntax, and Option D uses a non-existent `get logs` subcommand.

The manifest is valid. The container requests a guaranteed minimum of 250m CPU and 64Mi memory. It has a limit of 500m CPU and 128Mi memory, meaning it can burst its CPU usage up to that limit. Limits can be higher than requests. The container only faces termination if it exceeds the memory *limit*, not the request.

`kubectl set image` is the standard, declarative command for updating the image of a Deployment, ReplicaSet, or Pod. It triggers a rolling update by modifying the Pod template spec. While options A, C, and D are technically possible, option B is the most direct, reliable, and commonly used method for this specific task.

For sensitive data like database credentials, a Secret is the appropriate resource. While similar to a ConfigMap, Secrets are specifically designed for this purpose, offer slightly better protection (e.g., base64 encoding, optional encryption at rest), and can be mounted as a volume file or exposed as environment variables securely.

The core function of a liveness probe is to determine if a container is running. If it fails (based on its `failureThreshold`), the kubelet kills the container, which is then restarted subject to the Pod’s `restartPolicy`. It does not directly evict the Pod or just change its display status.

The `completions` field defines the total number of successful Pod completions needed for the Job to finish. The `parallelism` field defines the maximum number of Pods running concurrently. Therefore, `completions: 10` and `parallelism: 3` correctly models 10 tasks with up to 3 running at once. The `backoffLimit` is for retries on failure, which is also good practice but not the primary solution to the concurrency/completion requirement.

The `type: ClusterIP` field explicitly defines this as a ClusterIP service. This is the default type, creating a stable virtual IP address inside the cluster that load balances traffic to Pods matching the selector `app: api-server` on port 8080, accessible internally via port 80 on the Service IP.

The Downward API allows Pods to expose their own metadata to running containers. The field `status.hostIP` is the IP address of the node to which the Pod is assigned. Mounting it as a file in a volume is a standard and straightforward method. `spec.nodeName` is the node’s name, not its IP. Querying the API is more complex and unnecessary for this simple need.

Kubernetes nodes reserve a portion of their total memory for system processes and the kubelet. The ‘allocatable’ memory, which is what is available for Pods, is typically less than the total physical memory. A request for 500Mi might fail if the allocatable memory on any node is below this threshold, even if the total is 1Gi.0