Namespace creation is not a cluster wide synchronous process. A namespace is created by an admin user at a given node. This initial namespace configuration is then gossiped across nodes.
This means 2 bad things are possible:
The node dies before gossiping to the rest of the cluster.
Users simultaneously create namespaces with the same name on different nodes.
We can prevent both of these by creating a system provided root namespace to allow consistent namespace creation.
Since we are going to likely use a root namespace to allow consistent deletion of namespaces we may as well add this to the diskless 1.0 milestone as well. It's not hard to implement.
Namespace creation is not a cluster wide synchronous process. A namespace is created by an admin user at a given node. This initial namespace configuration is then gossiped across nodes.
This means 2 bad things are possible:
We can prevent both of these by creating a system provided root namespace to allow consistent namespace creation.