talos-node-removal

name	talos-node-removal
description	Use when removing, replacing, or decommissioning a Talos node — covers the correct talosctl reset workflow
user-invocable	false
origin	auto-extracted

Talos Node Removal

Extracted: 2026-06-12 Context: Removing a control plane or worker node (hardware replacement, decommission, OS reinstall)

Problem

Manual node removal requires 4+ steps in the correct order (cordon → drain → etcd remove-member → kubectl delete → shutdown). Wrong order risks etcd split-brain or the node failing to rejoin after reinstall.

Solution

talosctl reset handles everything automatically — cordons, drains, leaves etcd, erases disk, powers off:

# Step 1: Reset the node (handles cordon/drain/etcd-leave/erase/shutdown)
talosctl -n <node-ip> reset

# Step 2: Remove the Kubernetes node object (reset does NOT do this)
kubectl delete node <node-name>

The machine is powered off after reset. To reinstall (e.g. after hardware swap):

# Verify cluster version before applying
talosctl version -n <other-node-ip>

# Boot from Talos ISO, apply existing machine config
talosctl apply-config --insecure -n <node-ip> \
  -f provision/talos/clusterconfig/<node>.yaml

Talos installs onto the new disk and rejoins etcd automatically. No uncordon needed — nodes join uncordoned.

When to Use

• NVMe/disk replacement requiring power-off
• Node decommission
• OS reinstall or Talos version migration on a single node

Key Constraint

etcd requires 2/3 quorum — never reset two control plane nodes simultaneously.

Reference

Official docs: https://docs.siderolabs.com/talos/v1.13/deploy-and-manage-workloads/scaling-down