Draining a Node for Maintenance

Scheduling and Placement / Guide 44 of 103 / Intermediate

• Kubernetes1.36.4
• Cluster2 nodes
• Runtimecontainerd 2.2.6
• CNICalico v3.32.1
• TimeAbout 16 min

This guide includes

Use this before taking a node out for maintenance. This matters because kubectl drain refuses twice before it works, and each refusal names the flag it wants - DaemonSet Pods and local data are both things it will not decide for you.

• cordoning first, and seeing that it stops new work while moving nothing
• running a bare drain and reading the refusal instead of reaching for --force
• supplying --ignore-daemonsets and --delete-emptydir-data, the two flags it asked for
• meeting the third case - a Pod with no controller, which nothing will recreate

Before you start

• guide 39 for how pods land on nodes in the first place.
• A cluster with at least one worker you are willing to empty.

1. 1

   Cordon stops new work and moves nothing

2. 2

   Drain refuses, and says why

3. 3

   The two flags it asked for

4. 4

   A pod with no controller is different again

Official sources

• Safely Drain a Node
• kubectl drain
• Taints and Tolerations