Merge pull request #2799 from fgiloux/slice-doc

openshift-merge-robot · web-flow · commit 64341429a9d7 · 2023-02-21T23:41:52.000+01:00
📖 APIExportEndpointSlice and Partition doc
diff --git a/docs/content/en/concepts/partitions.md b/docs/content/en/concepts/partitions.md
@@ -0,0 +1,72 @@
+---
+description: >
+  How to create shard partitions
+---
+
+# Partition API
+
+`Partitions` and `PartitionSets` build an API that allows service providers and cluster administrators to create shard Partitions. These partitions can then be leveraged:
+- for the deployment of the controllers of the service providers and other components. These APIs provide the information on shard topology required for the scheduling.
+- for grouping APIExport endpoints of multiple shards in common buckets, that is `Partitions`. These buckets can the be used to achieve geo proximity: controllers should communicate with API servers of shards in the same region. They can also be leveraged for scalability and load distribution, that is to adapt the number of deployment/ processes to the load pattern specific to the controllers of a service provider.
+
+## Partitions
+
+The partitioning of `Shards` is done through labels and [label selectors](https://kubernetes.io/docs/concepts/overview/working-with-objects/labels/#label-selectors).
+
+Here is an example of a `Partition`
+
+```yaml
+kind: Partition
+apiVersion: topology.kcp.io/v1alpha1
+metadata:
+    name: cloud-region-gcp-europe-xdfgs
+    ownerReferences:
+    ...
+spec:
+    selector:
+     matchLabels:
+       cloud: gcp
+       region: europe
+     matchExpressions:
+     - key: country
+       operator: NotIn
+       values:
+       - LI
+  ```
+
+`Partitions` can be referenced in [`APIExportEndpointSlices`](./quickstart-tenancy-and-apis.md)
+
+## PartitionSets
+
+`PartitionSets` is  an API for convenience. `PartitionSet` can be used to get `Partitions` automatically created based on dimensions that match the shard label keys. The `Partitions` are created in the same workspace as the `PartitionSet`. They can then be copied to the desired workspace for consumption, for instance, by an `APIExportEndpointSlice`.
+
+Here is an example of a `PartitionSet`
+
+```yaml
+kind: PartitionSet
+apiVersion: topology.kcp.io/v1alpha1
+metadata:
+    name: cloud-region
+spec:
+   dimensions:
+   - region
+   - cloud
+   selectors:
+     matchExpressions:
+     - key: region
+       operator: NotIn
+       values:
+       - Antarctica
+       - Greenland
+     - key: country
+       operator: NotIn
+       values:
+       - NK
+status:
+   count: 10
+...
+```
+
+It is to note that a `Partition` is created only if it matches at least one shard. With the provided example if there is no shard in the cloud provider `aliyun` in the region `europe` no `Partition` will be created for it.
+
+An example of a `Partition` generated by this `PartitionSet` can be found above. The `dimensions` are translated into `matchLabels` with values specific to each `Partition`. An owner reference of the `Partition` will be set to the `PartitionSet`.
diff --git a/docs/content/en/concepts/quickstart-tenancy-and-apis.md b/docs/content/en/concepts/quickstart-tenancy-and-apis.md
@@ -247,7 +247,33 @@ EOF
 cowboy.wildwest.dev/one created
 ```
 
-## Dig deeper into `APIExports`
+## Managing permissions
+
+Besides publishing APIs and reconciliating the related resources service providers' controllers may need access to core resources or resources exported by other services in the user workspaces as part of their duties. This access needs for security reason to get authorized. `permissionClaims` address this need.
+
+A service provider wanting to access `ConfigMaps` needs to specify such a claim in the `APIExport`:
+
+```yaml
+spec:
+...
+  permissionClaims:
+    - group: ""
+      resource: "configmaps"
+```
+Users can then authorize access to this resource type in their workspace by accepting the claim in the `APIBinding`:
+
+```yaml
+spec:
+...
+  permissionClaims:
+    - group: ""
+      resource: "configmaps"
+      state: Accepted
+```
+
+There is the possibility to further limit the access claim to single resources.
+
+## Dig deeper into APIExports
 
 Switching back to the service provider persona:
 
@@ -263,25 +289,61 @@ metadata:
 status:
   ...
   identityHash: a6a0cc778bec8c4b844e6326965fbb740b6a9590963578afe07276e6a0d41e20
-  virtualWorkspaces:
-  - url: https://myhost:6443/services/apiexport/root:wildwest:cowboys-service/wildwest.dev
 ```
 
-We can see that our `APIExport` has two key attributes in its status - its identity (more on this below) and a "virtual
-workspace" URL. You can think of this virtual workspace as behaving just like a workspace or cluster, except it searches
-across all true workspaces for instances of the resource types provided by the `APIExport`. We can use API discovery to
-see what resource types are available via this virtual workspace:
+We can see that our `APIExport` has a key attribute in its status - its identity (more on this below).
+This identity can be used in permissionClaims for referring to non-core resources.
+
+## APIExportEndpointSlice
+
+`APIExportEndpointSlices` allow service provider to retrieve the URL of service endpoints, acting as a sink for them. You can think of this endpoint as behaving just like a workspace or cluster, except it searches across all workspaces for instances of the resource types provided by the `APIExport`.
+An `APIExportEndpointSlice` is created by a service provider, references a single `APIExport` and optionally a `Partition`.
+`Partitions` are a mechanism for filtering service endpoints. Within a multi-sharded kcp, each shard will offer its own service endpoint URL for an `APIExport`. Service provider may decide to have multiple instances of their controller reconciliating, for instance, resources of shards in the same region. For that they may create an `APIExportEndpointSlice` in the same workspace where a controller instance is deployed. This `APIExportEndpointSlice` will then reference a specific `Partition` by its name in the same workspace filtering the service endpoints for a subset of shards. If an `APIExportEndpointSlice` does not reference a `Partition` all the available endpoints are populated in its `status`. More on `Partitions` [here](./partitions,md).
+
+```shell
+$ kubectl apply -f - <<EOF
+kind: APIExportEndpointSlice
+apiVersion: apis.kcp.io/v1alpha1
+metadata:
+    name: cowboys
+spec:
+    export:
+        path: root:wildwest:cowboys-service
+        name: cowboy
+    # optional
+    partition: cloud-region-gcp-europe-xdfgs
+EOF
+apiexportendpointslice.apis.kcp.io/cowboys created
+```
+
+Looking at the status populated by the controller
+
+```shell
+$ kubectl get APIExportEndpointSlice/cowboys -o yaml
+kind: APIExportEndpointSlice
+apiVersion: apis.kcp.io/v1alpha1
+metadata:
+    name: cowboys
+...
+status:
+    endpoints
+        - url: https://host1:6443/services/apiexport/root:wildwest:cowboys-service/wildwest.dev
+        - url: https://host2:6443/services/apiexport/root:wildwest:cowboys-service/wildwest.dev
+...
+```
+
+We can use API discovery to see what resource types are available via the endpoint URL:
 
 ```shell
-$ kubectl --server='https://myhost:6443/services/apiexport/root:wildwest:cowboys-service/wildwest.dev/clusters/*/' api-resources
+$ kubectl --server='https://host1:6443/services/apiexport/root:wildwest:cowboys-service/wildwest.dev/clusters/*/' api-resources
 NAME      SHORTNAMES   APIVERSION              NAMESPACED   KIND
 cowboys                wildwest.dev/v1alpha1   true         Cowboy
 ```
 
 The question is ... can we see the instance created by the consumer?
 
 ```shell
-$ kubectl --server='https://myhost:6443/services/apiexport/root:wildwest:cowboys-service/wildwest.dev/clusters/*/' get -A cowboys \
+$ kubectl --server='https://host1:6443/services/apiexport/root:wildwest:cowboys-service/wildwest.dev/clusters/*/' get -A cowboys \
     -o custom-columns='WORKSPACE:.metadata.annotations.kcp\.dev/cluster,NAME:.metadata.name'
 WORKSPACE                                  NAME
 root:users:zu:yc:kcp-admin:test-consumer   one
@@ -293,13 +355,11 @@ Yay!
 
 Q: Why is there a new `APIResourceSchema` resource type that appears to be very similar to `CustomResourceDefinition`?
 
-A: FIXME
+A: An APIResourceSchema defines a single custom API type. It is almost identical to a CRD, but creating an APIResourceSchema instance does not add a usable API to the server. By intentionally decoupling the schema definition from serving, API owners can be more explicit about API evolution.
 
-Q: Why do I have to append `/clusters/*/` to the `APIExport` virtual workspace URL?
+Q: Why do I have to append `/clusters/*/` to the `APIExport` service endpoint URL?
 
-A: The URL represents the base path of a virtual kcp API server. With a standard kcp API server, workspaces live under
-the `/clusters/` path, so `/clusters/*/` represents a wildcard search across all workspaces via this virtual API
-server.
+A: The URL represents the base path of a virtual kcp API server. With a standard kcp API server, workspaces live under the `/clusters/` path, so `/clusters/*/` represents a wildcard search across all workspaces via this virtual API server.
 
 Q: How should we understand an `APIExport` `identityHash`?
 
@@ -309,19 +369,14 @@ some way of securely distinguishing them.
 Each `APIExport` is allocated a randomized private secret - this is currently just a large random number - and a public
 identity - just a SHA256 hash of the private secret - which securely identifies this `APIExport` from others.
 
-This is important because an `APIExport` makes a virtual workspace available to interact with all instances of a
-particular `APIResourceShema`, and we want to make sure that users are clear on which service provider `APIExports` they
-are trusting and only the owners of those `APIExport` have access to their resources via virtual workspaces.
+This is important because an `APIExport` makes service endpoints available to interact with all instances of a particular `APIResourceShema`, and we want to make sure that users are clear on which service provider `APIExports` they are trusting and only the owners of those `APIExport` have access to their resources via the service endpoints.
 
-Q: Why do you have to use `--all-namespaces` with the apiexport virtual workspace?
+Q: Why do you have to use `--all-namespaces` with the `APIExport` service endpoint?
 
-A: Think of this virtual workspace as representing a wildcard listing across all workspaces. It doesn't make sense to
-look at a specific namespace across all workspaces, so you have to list across all namespaces too.
+A: Think of this endpoint as representing a wildcard listing across all workspaces. It doesn't make sense to look at a specific namespace across all workspaces, so you have to list across all namespaces too.
 
-Q: If I attempt to use an `APIExport` virtual workspace before there are any `APIBindings` I get the "Error from server
-(NotFound): Unable to list ...: the server could not find the requested resource". Is this a bug?
+Q: If I attempt to use an `APIExport` endpoint before there are any `APIBindings` I get the "Error from server (NotFound): Unable to list ...: the server could not find the requested resource". Is this a bug?
 
 A: It is a bug. See <https://github.com/kcp-dev/kcp/issues/1183>
 
-When fixed, we expect the `APIExport` behavior will change such that there will be no virtual workspace URLs until an
-`APIBinding` is created.
+When fixed, we expect the `APIExport` behavior will change such that an empty list is returned instead of the error.
