Docs

Volcengine Kubernetes Engine

Managing nodes

Managing nodes of managed clusters

The existing clusters can be managed on the node management page. This topic introduces the scenarios, impacts, and procedures of operation related to node management.

Operation of a single node

Taking nodes offline

During routine cluster maintenance, nodes may need to go offline to ensure data security in scenarios such as upgrades, node scale-out, maintenance downtime, node reclaim, and change operations. For more best practices, see Safely taking nodes offline.

Warning

When a node is taken offline, all pods on that node (except DaemonSet) will be migrated to other nodes in the cluster. The node will then be set to unschedulable status, so new pods cannot be scheduled on it. This action may adversely affect business operations. We recommend performing this action during low-traffic hours.

Taking a node offline will reduce the remaining ResourceQuota of the cluster to which the node belongs. If the cluster does not have sufficient resource reservations, this action will adversely affect the scheduling of new services and the running of existing workloads.

Click the target cluster name, then in the left navigation bar, select Nodes > Node.

Click Go offline in the Actions column of the target node.

In the pop-up window, click Confirm to proceed.

Cordoning nodes

Unlike taking nodes offline, the action of cordoning nodes does not migrate pods to other nodes in the cluster. Instead, the node is set to cordoned and unschedulable status, and the pods already running on the node are not affected.

tip

Cordoning a node will reduce the remaining ResourceQuota of the cluster to which the node belongs. If the cluster does not have sufficient resource reservation, this action will affect the scheduling of new services and the running of existing workloads.

Click the target cluster name. In the left navigation bar, select Nodes > Node.

Click Cordon in the ... menu to the right of the target node.

In the pop-up window, click Confirm to proceed.

Switching to schedulable

You can restore scheduling capability for nodes to allow them to accept new pods. After a node is restored to schedulable status, the remaining ResourceQuota of the cluster to which the node belongs will increase.

Click the name of the target cluster. In the left navigation bar of the cluster management page, click Nodes > Node.

Click Make it schedulable in ... menu to the right of the target node.

In the pop-up window, click Confirm to proceed.

Label management

Assigning different labels to nodes allows you to define various attributes, making batch filtering and similar operations more convenient. You can manage node labels in the Volcengine Kubernetes Engine (VKE) console, including adding, editing, and deleting node labels.

Click the name of the target cluster. In the left navigation bar of the cluster management page, click Nodes > Node.

Click Label Management in the ... menu to the right of the target node to go to the Label tab, where you can manage labels for the node.

Taint management

Applying one or more taints to a node allows the node to reject specific pods, preventing those pods from being scheduled on the node. You can manage node taints in VKE console, including adding, editing, and deleting node taints.

Click the name of the target cluster. In the left navigation bar of the cluster management page, click Nodes > Node.

Click Taint management in ... menu to the right of the target node to go to the Taint tab, where you can manage taints for the node.

ECS terminal

Log in to the node console using ECS Terminal, where you can manage the node using the command line. For details, see Log in to Linux instances via the console.

Click the name of the target cluster. In the left navigation bar of the cluster management page, click Nodes > Node.

Find the target node and click ... > ECS Terminal to connect to the console using the ECS terminal, where you can manage the node.

Viewing monitoring data

VKE allows you to view node-related metrics and dashboards by using the cloud native observability feature. The alerting feature of Volcengine Managed Service for Prometheus (VMP) pushes alerting messages based on alerting rules. This achieves real-time monitoring of node health status.

Click the name of the target cluster. In the left navigation bar of the cluster management page, click Nodes > Node.

Click ... > View monitoring in the Actions column of the target node to go to the node monitoring page of cloud native observability, where you can view the monitoring dashboards related to the node. For more details, see VKE observability.

Removing nodes

The nodes that are no longer needed can be removed, provided that the action will not adversely affect the ongoing business.

warning

This action will cause the pods on the node to be automatically migrated to other nodes. We recommend doing it during low-traffic hours.

For a node of the Pay-as-you-go or spot instance type, the removal operation allows you to destroy and release the node. For a node of the subscription type, the action only excludes it from the node list. If the corresponding Elastic Compute Service (ECS) instance is required to be released, see Canceling Elastic Compute Service subscription.

If you destroy and release a node, the data on the system volume and data volume will be deleted. This action cannot be undone. You need to proceed with caution.

If the node pool where the node resides has enabled Automatic binding of Elastic IP Addresses, the bound IP addresses will be released along with the deletion and release of the node.

If you do not destroy and release the node, the corresponding ECS instance will start to reinstall the operating system and clear the data on the system volume of the instance. When reinstalling the operating system, ECS will forcibly shut down Bare Metal instances and some High Performance Computing Optimized GPU instances. For details, see Shutting down instances.

If Buy affinity group (beta) has been enabled for the node pool to which the target node belongs, the number of nodes must be an integer multiple of the affinity group size. Therefore, you cannot remove a single node. You can remove nodes by using the following methods:

Remove entire groups of nodes in a batch, with an affinity group as the minimum unit. For example, if the affinity group size of the current node pool is 2, you need to remove 2 nodes in the same affinity group at the same time. To remove multiple affinity groups, you need to select all nodes in each group. For the instructions on how to remove nodes in a batch, see the Remove nodes in a batch section.

Scale in the node pool. For details, see Scaling node pools.

Click the name of the target cluster. In the left navigation bar of the cluster management page, click Nodes > Node.

Click Remove in ... menu to the right of the target node.

In the pop-up window, select the options as needed. Then, click Remove to remove the node.

Parameter	Description
Synchronously destroy nodes of the pay-as-you-go or spot instance type	Whether to destroy nodes synchronously when removing them? Enable: Nodes of pay-as-you-go and spot instance types will be destroyed when removed. Disable: Nodes will not be destroyed when removed.
Migrate all pods on the node except DaemonSet and set the node to unschedulable status before removal.	Whether to drain the node first before removal. Enable: drains the node before removal, which means migrating all pods on it (except DaemonSet) elsewhere and setting it to unschedulable status. Disable: does not drain the node before removal.

Diagnosis

The diagnosis of nodes can help users identify potential risks that may exist within a VKE cluster, including abnormal node status, excessive node workload, node network failures, and node kernel errors.

Click the target cluster name. In the left navigation bar, select Nodes > Node.

Select Diagnosis in ... menu to the right of the target node.

In the new page, create diagnostic tasks for this node. For details, see Node diagnosis.

Batch management of nodes

VKE allows batch management of nodes, making it easy to quickly perform operations across multiple nodes.

tip

Virtual nodes cannot be managed in batch.

Click the name of the target cluster. In the left navigation bar of the cluster management page, click Nodes > Node.

Select the nodes you want and click Batch operation on the page, and then select the operation as needed.

The following content includes descriptions for each option in the Batch operations menu.

Batch operation and maintenance

You can create batch O&M jobs for ECS nodes. Batch Job is a native automated O&M tool provided by Volcano Engine. It allows you to conveniently run O&M commands on one or more targets without passwords, logins, or bastion hosts. For details, see Batch jobs.

Before performing batch operation and maintenance on nodes, install the batch job client on each node. For details, see Installing batch job clients.

Select the nodes you want and click Batch operation > Batch operation and maintenance on the page.

In the pop-up window, click Confirm to proceed.

Batch set to schedulable

This option can switch the nodes that have been cordoned or taken offline to the schedulable status, making them available again to accommodate new pods. This action will increase the remaining ResourceQuota of the cluster to which this node belongs.

Select the desired nodes and click Batch operation > Enable scheduling in batches on the page.

In the pop-up window, click Make it schedulable to proceed.

tip

Only the nodes that have been cordoned or taken offline can be set to schedulable status.

Batch cordon

This option can cordon nodes in batch. The existing pods on the nodes will not be affected.

Select the nodes you want and click Batch operation > Batch cordon on the page.

In the pop-up window, click Cordon to proceed.

Batch offline

This option can take nodes offline in batch. For details, see Taking nodes offline mentioned earlier in this topic.

Select the nodes you want and click Batch operation > Batch offline on the page.

In the pop-up window, click Go offline to proceed.

Batch removal

This option can remove nodes in batches. For details, see Removing nodes mentioned earlier in this topic.

Select the nodes you want and click Batch operation > Batch removal on the page.

In the pop-up window, click Remove to proceed.

Parameter	Description
Synchronously destroy nodes of the pay-as-you-go or spot instance type	Whether to destroy nodes synchronously when removing them? Enable: Nodes of pay-as-you-go and spot instance types will be destroyed when removed. Disable: Nodes will not be destroyed when removed.
Migrate all pods on the node except DaemonSet and set the node to unschedulable status before removal.	Whether to drain the node first before removal. Enable: drains the node before removal, which means migrating all pods on it (except DaemonSet) elsewhere and setting it to unschedulable status. Disable: does not drain the node before removal.

Managing nodes on the node pool page

On the Node pool page, you can also manage nodes in the node pool by bringing them online, taking them offline, removing them, or cordoning them, either individually or in batches.

Click the target cluster name. In the left navigation bar, click Node pool.

Select the Node list tab, and then perform operations on the nodes as needed.

Last updated: 2026.07.24 14:51:35