TL;DR

Upgrading directly from etcd v3.5 to v3.6 can result in 'zombie' members joining the database consensus. To prevent this, upgrade to v3.5.26 first for automatic repair before proceeding to v3.6.

What happened

During upgrades from v3.5 to v3.6, clusters may encounter a bug causing removed nodes (zombies) to reappear and join the cluster. This can be resolved by upgrading to v3.5.26 or later first.

Why it matters for ops

The issue stems from inconsistencies between etcd's v2store and v3store data stores when transitioning membership management, fixed in v3.5.26 updates.

Action items

  • Upgrade clusters to etcd v3.5.26 or later
  • Verify cluster health post-upgrade before moving to v3.6
  • Follow official upgrade paths for a seamless transition

Source link

https://kubernetes.io/blog/2025/12/21/preventing-etcd-zombies/