3.4. Nodes (Node Management)

Overview

The Nodes page manages the nodes of every member cluster registered with the MCM control plane in one place. You can take in and monitor node status, resource usage, and role information per cluster at a glance.

What it is for:

• Monitoring multi-cluster nodes together
• Filtering and searching nodes per cluster
• Checking node status and resource usage
• Identifying and managing GPU nodes
• Diagnosing and analysing node problems

How the Screen Is Laid Out

The node management page comes in two forms, a table view and a grid view.

1. Table View

The default form, showing the node list as a table.

The Elements at the Top of the Page

Element	Description
Cluster filter	Shows only the nodes of a particular cluster
View switch buttons	Switch between grid view and table view
Refresh button	Refreshes the node list
Search box	Searches by node name

Table Columns

Column	Description	Sortable
Name	The node hostname	✅
Cluster	The cluster the node belongs to	✅
Status	Ready / Not Ready	✅
Role	master, worker, control-plane, etcd, infra and so on	✅
Taints	The number of taints set on the node	✅
Internal IP	The node's internal IP address	✅
CPU	CPU usage and utilisation (with a progress bar)	✅
Memory	Memory usage and utilisation (with a progress bar)	✅
Pods	Current pods / maximum pods	✅
GPU	GPU information (shown as NVIDIA where present)	✅
Kubelet Version	The Kubelet version	✅
Age	The time since the node was created	✅

The Format Resource Usage Is Shown In

usage / allocatable (utilisation%)

For example 2.1 Core / 72 Core (2.9%)

• Actually used: 2.1 cores
• Allocatable: 72 cores
• Utilisation: 2.9%

2. Grid View (Cards)

Shows the nodes as cards so they can be taken in visually.

How a Node Card Is Laid Out

Each node card shows:

Element	Description
Node name	The node hostname
Status badge	Ready (green) / Not Ready (red)
Cluster	The name of the cluster the node belongs to
Role badges	master, worker, control-plane, etcd and so on
GPU badge	For a GPU node, shows NVIDIA and the GPU count
Internal IP	The node's internal IP address
Kubelet Version	The Kubelet version
OS Image	The operating system (Rocky Linux 9.5, for example)
CPU	Usage / allocatable (utilisation%)
Memory	Usage / allocatable (utilisation%)
Pods	Current + Karmada pods / maximum (utilisation%)
Taints	The taints (shown for a not-ready node)

3. The Cluster Filter

Lets you filter the view to the nodes of particular clusters.

• One or more clusters can be chosen from the dropdown
• Several clusters can be chosen at once with the checkboxes
• Clearing the filter shows the nodes of every cluster

Node Detail

Clicking a node card or a table row opens the detail panel.

The Detail Panel Header

Element	Description
Node name	The hostname of the chosen node
Status badge	Ready / Not Ready
Cluster badge	The cluster the node belongs to
Refresh button	Refreshes the node information

The Tabs

The Overview Tab

Shows the node's basic information, resource usage, and system information.

Basic information:

Item	Description
Name	The node hostname
Cluster	The cluster the node belongs to
Status	Ready / Not Ready
Internal IP	The node's internal IP address
Hostname	The node hostname
Unschedulable	Whether scheduling is possible (Yes/No)
Pod CIDR	The pod CIDR assigned to the node
Age	The time since the node was created
Created	When the node was registered

Resources:

Item	What is shown
CPU	Usage / allocatable (utilisation%), with a progress bar
Memory	Usage / allocatable (utilisation%), with a progress bar
Pods	Current / capacity (utilisation%), with the failed pod count
GPU	The GPU count, or "No GPU"

System information:

Item	Description
OS Image	The operating system image (Rocky Linux 9.5, for example)
Kernel Version	The Linux kernel version
Container Runtime	The containerd or docker version
Kubelet Version	The Kubelet version
Kube Proxy Version	The Kube Proxy version
Operating System	OS/architecture (linux/amd64, for example)
Machine ID	The node machine ID
System UUID	The node system UUID

The Conditions Tab

Shows the node's status conditions:

• Ready: whether the node is healthy
• MemoryPressure: whether memory is short
• DiskPressure: whether disk is short
• PIDPressure: whether PIDs are short
• NetworkUnavailable: whether there is a network problem

The Taints Tab

Shows the taints set on the node:

• The taint key
• The taint value
• The taint effect (NoSchedule, NoExecute, PreferNoSchedule)

The Labels Tab

Shows the list of Kubernetes labels applied to the node.

The Annotations Tab

Shows the list of annotations applied to the node.

The Pods Tab

Shows the list of pods running on that node.

The Events Tab

Shows the Kubernetes events related to the node.

Node Status

Status Types

Status	Description	How it is shown
Ready	The node is working normally	A green badge
Not Ready	The node has a problem	A red badge

What a Not-Ready Node Looks Like

A not-ready node shows these traits:

• The status badge is red
• Resource usage may show as 0
• The node.kubernetes.io/unreachable taint is applied automatically
• NoSchedule and NoExecute effects block new pods being scheduled

Node Roles

Role	Description
master	A Kubernetes master node
control-plane	A node running control plane components
etcd	An etcd cluster member node
worker	A node running workloads
infra	A node for infrastructure components

Note: one node can hold several roles (master, control-plane, and etcd, for example).

GPU Nodes

A node with GPUs is shown as follows:

• Table view: shown in the GPU column as "N NVIDIA" (N being the GPU count)
• Grid view: a GPU badge beside the role badges
• Detail panel: the GPU count in the GPU section

Worked Examples

Scenario 1: Checking Nodes per Cluster

1. Click the cluster filter dropdown
2. Choose the cluster you want to check
3. Only that cluster's nodes are listed
4. Check the status and resources of each node

Scenario 2: Diagnosing a Not-Ready Node

1. Find the nodes with a "Not Ready" status in the list
2. Click that node to open the detail panel
3. Check which condition is False on the Conditions tab
4. Check the automatically applied taint on the Taints tab
5. Check the related events on the Events tab
6. Resolve the node problem on that cluster

Scenario 3: Monitoring Resource Usage

1. Sort by the CPU or Memory column in the table view
2. Identify the nodes with high utilisation (visible from the progress bars)
3. Click a node for its detail if needed
4. Check which pods use a lot of resources on the Pods tab
5. Consider redistributing workloads or growing the nodes if needed

Scenario 4: Managing GPU Nodes

1. Identify the GPU nodes by checking or sorting the GPU column
2. Check the resource usage of the GPU nodes
3. Check the GPU information on the node detail page if needed
4. Use it when placing GPU workloads

Scenario 5: Searching for a Node

1. Type the node name in the search box
2. The matching nodes are filtered
3. Click the node you want for its detail

Related Concepts

Node Taints

A taint is the mechanism that stops particular pods being scheduled on a node:

• NoSchedule: stops new pods being scheduled
• NoExecute: evicts existing pods as well
• PreferNoSchedule: avoids scheduling where possible (not enforced)

A not-ready node automatically receives the node.kubernetes.io/unreachable taint.

Node Resources

• Allocatable: the resources actually available to pods
• Capacity: the node's total resource capacity
• Usage: the resources currently in use

Karmada Pods

In the pod figure, the number after the "+" is the number of pods Karmada manages:

15 + 4 / 110 (17%)

• 15: ordinary pods
• 4: Karmada-related pods
• 110: the maximum pod count
• 17%: overall utilisation

---

Next Steps

• Deployments -- managing multi-cluster Deployments
• DaemonSets -- managing multi-cluster DaemonSets

---