This patch release enables etcd corruption checks on every etcd ring that is running etcd 3.5 (which applies to all user clusters with Kubernetes 1.22). This change is a recommendation from the etcd maintainers due to issues in etcd 3.5 that can cause data consistency issues. The changes in this patch release will prevent corrupted etcd members from joining or staying in the etcd ring.
To replace a member in case of data consistency issues, please:
- Follow our documentation for replacing an etcd member if you are not running etcd-launcher.
- Delete the
PersistentVolumethat backs the corrupted etcd member to trigger the automated recovery procedure if you are using etcd-launcher.
Please be aware we do not recommend enabling etcd-launcher on existing Kubernetes 1.22 environments at the time. This is due to the fact that the migration to etcd-launcher requires several etcd restarts and we currently recommend keeping the etcd ring as stable as possible (apart from the restarts triggered by this patch release to roll out the consistency checks).
Misc
- For user clusters that use etcd 3.5 (Kubernetes 1.22 clusters), etcd corruption checks are turned on to detect etcd data consistency issues. Checks run at etcd startup and every 4 hours (#9480)