Nodes, cordon and drain

Click Nodes in the sidebar. The table shows Name, Status, Roles, Version, CPU, Memory and Age.

Use CPU & Memory: above the table to show Usage, Requests or Limits. See CPU and memory bars.

Click a node to open it. The Capacity card shows CPU, memory and pods against what the node can hold. Pods on this node lists its pods.

1. Open the node.
2. Click Cordon.
3. Click Cordon in the dialog.

The node is marked unschedulable. Its pods keep running, but no new pods go there. To undo it, click Uncordon, then Uncordon in the dialog.

1. Open the node.
2. Click Drain.
3. Click Drain in the dialog.

Glauco then:

• cordons the node;
• evicts every pod it can, except DaemonSet pods and mirror (static) pods.

A PodDisruptionBudget can refuse an eviction. Glauco skips that pod and goes on. It does not wait for the pods to stop. Drain has no options. For a drain with options, such as --delete-emptydir-data or a timeout, use kubectl drain.