Clarify namespace sameness wrt policy

After many discussions with customers and implementors, I think we need
to clarify that implementations DO have an axis of freedom around
implementation-defined control of sameness.

E.g. "Sameness applies to <this namespace> only in <these clusters>,
and to <that namespace> in <those clusters>."

Author: thockin

sig-multicluster/namespace-sameness-position-statement.md

@@ -1,38 +1,42 @@
 # Namespace Sameness - SIG Multicluster Position Statement
 
-Author: Jeremy Olmsted-Thompson (**[@jeremyot](https://github.com/jeremyot)**), Google  
-Last Edit: 2020/04/20  
+Last Edit: 2023/04/22
 Status: RELEASED
 
 ## Goal
+
 To establish a normative statement for multi-cluster namespace semantics and
 governance as a building block for further development which will require
 specifying behaviors across clusters.
 
 ## Context
+
 Users are reaching for multi-cluster deployments for a
 [variety of reasons](http://bit.ly/k8s-multicluster-conversation-starter-doc).
-However, Kubernetes treats the cluster boundary as the edge of the universe.
-There are currently no standard practices for how to extend the Kubernetes
-resource model across multiple clusters. Without common patterns we can’t build
-portable tooling to facilitate multi-cluster capabilities and know that behavior
-will be consistent for each user.
+However, Kubernetes has historically treated the cluster boundary as the edge
+of the universe.  Without common patterns we can’t build portable tooling to
+facilitate multi-cluster capabilities and know that behavior will be consistent
+for each user.
 
 ## Scope
+
 A single organization may need multiple, disjoint sets of clusters. They may,
 for example, represent different phases of the development lifecycle (dev,
 staging, prod) or support unrelated projects. Each organization governs their
 own clusters in isolation, so the scope of a namespace can only reasonably be
 declared within the organizational boundary. The scope of namespace identity is
-defined as the union of clusters, governed by a single authority, that are
+defined as the set of all clusters, governed by a single authority, that are
 expected to work together.  An authority is a company, organization, team,
 individual, or other entity which is entrusted to manage these clusters and, in
 particular, to create namespaces in them.
 
 ## Position
-**For a set of related clusters governed by a single authority, all namespaces of
-a given name are considered to be the same namespace. A single namespace should
-have a consistent owner across the set of clusters.**
+
+**Within a set of related clusters which are governed by a single authority (a
+"clusterset"), samely named namespaces in all clusters are considered to be
+related, to have the same owner(s), and to represent the same intention,
+unless otherwise dictated by some overriding policy, including, but not limited
+to, implementation-specific configuration.**
 
 ## Appendix
 
@@ -46,12 +50,12 @@ Consider an organization that runs two clusters, "A" and "B".  They have a
 cluster-ops team which is responsible for keeping those clusters alive and
 healthy.  They also have app-teams, "foo" and "bar" which use those clusters.
 
-Cluster-ops owns the creation of namespaces in both clusters. They can choose
+Cluster-ops owns the creation of namespaces in all clusters. They can choose
 to delegate the ability to create namespaces to foo-team and bar-team, but any
-namespace name that is allocated in "reserved" in all clusters.  How the
+namespace name that is allocated is "reserved" in all clusters.  How the
 delegation works is an area for innovation, but some possible examples
 include:
-  * A self-service portal which allocate a name in a global database and
+  * A self-service portal which allocates a name in a global database and
     creates the namespace on the user's behalf
   * Tooling that allocates a name in a global database and issues a credential
     to the user which is checked in an admission controller
@@ -79,7 +83,7 @@ If there are special RBAC rules that need to be applied in some clusters (e.g.
 clusters in EU have more limited access), the LDAP-to-RBAC sync implementation
 can apply specializations based on whatever criteria it understands.
 
-### Example 3: Multi-cluster services
+### Example 3: Multi-cluster Services
 
 Consider the same organization from previous examples.  They have enabled an
 implementation of multi-cluster Services, which lets them access backends which
@@ -88,14 +92,42 @@ are running on other clusters in the group.
 As in example 2, cluster-ops run a per-cluster metrics service,  It does not
 make sense for clients in cluster A to access the metrics server of cluster B.
 Even though this runs in the same "metrics" namespace in both clusters (and
-thus namespce sameness applies), the implementation of multi-cluster services
+thus namespace sameness applies), the implementation of multi-cluster services
 probably should not be on-by-default.  The details of how multi-cluster
-services will work are an area for innovation, but ideas include:
-  * Opt-in: services must be "exported" to be merged across clusters
-  * Opt-out: services or namespace can be opted out of service merging
+services work are out of scope for this document, but valid approaches could
+include:
+  * Opt-in: services must be explicitly "exported" to be merged across clusters
+  * Opt-out: services or namespaces can be opted out of service merging
   * Different discovery: merged services and "raw" services use different names
     or other discovery mechanisms
 
-However the implementation works, the metrics service is not automatically
-merged across clusters, though the LDAP-to-RBAC sync from example 2 can still
-apply consistent policies.
+However the implementation works, the metrics service is not _necessarily_
+merged across clusters, even if the LDAP-to-RBAC sync from example 2 still
+applies consistent RBAC policies.  Those are independent capabilities.
+
+### Example 4: Authority sameness policy
+
+Consider the same organization from previous examples, but with more clusters:
+"A", "B", "C", and "D".  Business rules dictate that the foo-team and the
+bar-team have a common security posture and may share clusters, but the
+billing-team has stricter requirements and must not share clusters with other
+teams. Cluster-ops wants to assign foo-team's and bar-team's namespaces to
+clusters A and B, and billing-team's namespaces to clusters C and D.  They also
+want to ensure that the opposite is never true - foo-team's and bar-team's
+namespaces cannot be used from clusters C or D, and billing-team's namespaces
+cannot be used from clusters A or B.
+
+What exactly it means to "be used" depends on each specific multi-cluster
+capability.  For multi-cluster Services, the billing-team's endpoints from
+clusters C and D would be eligible for merging, but billing endpoints from
+clusters A and B would not.  This can be used by implementations to add another
+layer of safety - even if cluster A were compromised and the billing-team's
+namespaces was created, the MCS implementation can use the policy described
+above to know that it should NOT merge billing endpoints from cluster A.
+
+As an implementation choice, the authority may offer controls which govern
+which cluster-namespaces are considered for sameness and which are not.
+The details of how these policies might be expressed could include:
+  * In-cluster namespace labels or annotations
+  * In-cluster custom-resources
+  * An external control-plane API