If there's a workload which creates or rotates through a lot of Kubernetes objects (like namespaces) the etcd DB size can easily reach the 8GB limit in 5min between the Kubernetes API automated defrags.

When the limit is reached etcd will arm alarm:NOSPACE which prevents any future attempts by Kubernetes API to compact and defrag the database.

Run a workload which creates and deletes namespaces quickly.

1.Create and delete Kubernetes objects until etcd DB reaches more than 8GB within 5min
2.Etcd will lock up
3.

Etcd arms alarm:nospace, this requires a manual intervention to compact and defrag.

Etcd automatically compacts and defrags, possibly sends an alert to investigate the issue further.