Skip to main content

Network I/O Monitoring Extension

Runtime Environment - Monitoring Tab

Metrics Server provides CPU/memory usage, but not network I/O metrics. If you need per-container network usage, install the Network I/O Monitoring Extension.

When Do You Need This Extension?
  • When you need to view per-container network traffic
  • When you suspect a network bottleneck and want per-Pod traffic analysis

Monitoring Components​

Metrics Server vs Monitoring Extension​

Let's compare the role of each tool.

  • Metrics Server: Provides CPU/memory resource metrics. Used by HPA and the kubectl top command. All CPU/memory usage displayed in KIOPS is collected through Metrics Server.
  • cAdvisor: The lightweight option. Deployed as a DaemonSet on each node to collect container resource and network metrics. This is the recommended option.
  • Prometheus Stack: The standard option. Installs the kube-prometheus-stack chart with Helm, enabling metric collection, querying, and long-term storage. For advanced users.
Single Option Selection

In the KIOPS Network I/O Monitoring Extension modal, you can install only one of cAdvisor or Prometheus Stack via a radio button. There is no flow to install both together.

Prerequisites​

  • A Kubernetes cluster must be registered in KIOPS.
  • Helm must be installed on the cluster master node.
  • Sufficient cluster resources (for Prometheus Stack).

Permission Notice: If you cannot access this feature, please request permission from your organization manager.

Step 1: Navigate to Real-time Monitoring Tab​

  1. From the [Runtime Environments] page, select the target cluster.
  2. Click the Real-time Monitoring tab on the cluster detail page.
  3. Click the Network I/O Monitoring Extension button to open the modal.

Step 2: Preflight Check (12 items)​

When the modal opens, the following 12 items are automatically checked.

  1. connection: Whether SSH to the master node is possible
  2. commands: Whether required commands (kubectl, helm, etc.) exist
  3. kubernetes_version: Whether the Kubernetes version is supported
  4. helm_installed: Whether Helm is installed (required for Prometheus Stack)
  5. available_memory: Whether available memory is sufficient
  6. available_disk: Whether available disk space is sufficient
  7. namespace_permission: Permission to create namespaces
  8. cluster_permission: Permission to create cluster resources
  9. network_access: Network access for pulling container images
  10. storage_class: Whether a usable StorageClass exists
  11. node_count: Node count
  12. existing_installation: Whether a previous installation exists
Recommended Option Auto-Selected

Based on the recommended_option returned by the preflight check, the radio button default is automatically set. In general, cAdvisor is recommended for resource-constrained or simpler environments, while Prometheus Stack is recommended when sufficient resources and Helm are in place.

Step 3: Choose Installation Option​

After preflight check passes, choose one of the following radio options.

  • Install cAdvisor (lightweight, recommended): The lightweight option. Deploys cAdvisor as a DaemonSet on each node to collect container metrics.
  • Install Prometheus Stack (standard, advanced): The standard option. Uses Helm to install the kube-prometheus-stack chart, which includes Prometheus, Grafana, and Alertmanager.
Only Requirements Shown

Each option's required resources (recommended memory/disk, etc.) are shown only as requirements text. There is no UI to directly enter CPU/memory/storage values. There is also no UI for retention period or alert rules (Slack/Email/Webhook).

Step 4: Execute Install (5 phases)​

When you click Start Install, the backend executes the following 5 phases.

  1. prepare: Preliminary preparation tasks
  2. namespace: Create the monitoring-dedicated namespace
  3. helm_repo: Add the Helm repository (for Prometheus Stack)
  4. install: Install components (cAdvisor DaemonSet or Helm chart)
  5. verify: Verify the installation
Phase Differences By Option

Depending on the option, some phases may be skipped or behave differently. For example, cAdvisor may not run the helm_repo phase, while Prometheus Stack runs all 5 phases.

Background Execution Supported​

If you close the modal during installation, the backend job keeps running. When you return to the Monitoring tab later, you can see the progress and result.

Check Installation Status​

Status Display​

You can check the status of each component in the Monitoring tab:

  • Not Installed: The monitoring component has not been installed yet.
  • Installing: Installation is in progress. Typically takes 1-3 minutes.
  • Running: The component is operating normally.
  • Error: An error occurred. Try reinstalling if necessary.

Real-time Metric Verification​

Metrics available in KIOPS after installation:

Node Level (Metrics Server + Extension):

  • CPU usage (%) — provided by Metrics Server
  • Memory usage (GB / %) — provided by Metrics Server
  • Network I/O — provided by the monitoring extension

Pod Level:

  • Per-container CPU/memory usage — provided by Metrics Server
  • Per-container network traffic — provided by the monitoring extension
  • Restart count
CPU/Memory Comes From Metrics Server

Regardless of whether the monitoring extension (cAdvisor / Prometheus Stack) is installed, CPU/memory usage is collected via Metrics Server. Installing Metrics Server first is therefore recommended.

Troubleshooting​

Preflight Failures​

  • helm_installed fails (when Prometheus Stack is selected): Helm is not installed on the master node. Ask a cluster administrator to install Helm and try again. Alternatively, choose cAdvisor to install without Helm.
  • available_memory / available_disk fails: Insufficient resources. Switch to the cAdvisor (lightweight) option or free up cluster resources.
  • storage_class fails: No usable StorageClass. A StorageClass must be created on the cluster first.

cAdvisor Installation Failure​

  • When DaemonSet deployment fails: Insufficient resources on nodes to schedule cAdvisor Pods. Check node resource availability.
  • When image pull fails: Check cluster node internet connectivity or private registry settings.

Prometheus Stack Installation Failure​

  • Helm chart install failure: On the master node, inspect with helm repo list and helm list -A.
  • When OOM (Out of Memory) occurs: Node memory is insufficient. Add a larger node or switch to the cAdvisor option.

Metrics Not Being Collected​

  • Cause 1: cAdvisor/Prometheus Pod is not in Running state.
  • Cause 2: Blocked by a network policy.

Resolution:

  1. Check Pod status: kubectl get pods -n monitoring
  2. Check logs: kubectl logs <pod-name> -n monitoring