Quickstart: Tenancy and APIs
A running kcp server. The quickstart is a good starting point.
Set your KUBECONFIG
To access a workspace, you need credentials and an Kubernetes API server URL for the workspace, both of which are stored
The default context in the kcp-provided
admin.kubeconfig gives access to the
root workspace as the
You can use API discovery to see what resources are available in this
root workspace. We're here to explore the
Create and navigate some workspaces
ws plugin for
kubectl makes it easy to switch your
kubeconfig between workspaces, and to create new ones:
$ kubectl ws . Current workspace is "root". $ kubectl ws create a --enter Workspace "a" (type root:organization) created. Waiting for it to be ready... Workspace "a" (type root:organization) is ready to use. Current workspace is "root:a". $ kubectl ws create b Workspace "b" (type root:universal) created. Waiting for it to be ready... Workspace "b" (type root:universal) is ready to use. $ kubectl get workspaces NAME TYPE PHASE URL b universal Ready https://myhost:6443/clusters/root:a:b $ kubectl ws b Current workspace is "root:a:b". $ kubectl ws .. Current workspace is "root:a". $ kubectl ws - Current workspace is "root:a:b". $ kubectl ws root Current workspace is "root". $ kubectl get workspaces NAME TYPE PHASE URL a organization Ready https://myhost:6443/clusters/root:a
kubeconfig now contains two additional contexts, one which represents the current workspace, and the other to keep
track of our most recently used workspace. This highlights that the
kubectl ws plugin is primarily a convenience
wrapper for managing a
kubeconfig that can be used for working within a workspace.
Understand workspace types
As we can see above, workspaces can contain sub-workspaces, and workspaces have different types. A workspace type defines which sub-workspace types can be created under such workspaces. So, for example:
- A universal workspace is the base workspace type that most other workspace types inherit from - they may contain other universal workspaces, and they have a "default" namespace
- The root workspace primarily contains organization workspaces
- An organization workspace can contain universal workspaces, or can be further subdivided using team workspaces
The final type of workspace is "home workspaces". These are workspaces that are intended to be used privately by
individual users. They appear under the
root:users workspace (type
homeroot) and they are further organized into a
homebucket workspaces based on a hash of their name.
Workspace types and their behaviors are defined using the
$ kubectl get workspacetypes NAME AGE home 74m homebucket 74m homeroot 74m organization 74m root 74m team 74m universal 74m $ kubectl describe workspacetype/team Name: team ... API Version: tenancy.kcp.io/v1alpha1 Kind: WorkspaceType ... Spec: Default Child Workspace Type: Name: universal Path: root Extend: With: Name: universal Path: root Limit Allowed Parents: Types: Name: organization Path: root Name: team Path: root Status: Conditions: Status: True Type: Ready ...
Q: Why do we have both
Workspaces are intended to be the user-facing resource, whereas
ClusterWorkspaces is a low-level resource for kcp
Workspaces are actually a "projected resource", there is no such resource stored in etcd, instead it is served as a
transformation of the
ClusterWorkspaces contain details like shard assignment, which are
low-level fields that users should not see.
We are working on a change of this system behind the scenes. That will probably promote Workspaces to a normal, non-projected resource, and ClusterWorkspaces will change in its role.
Publish some APIs as a service provider
APIBinding resources which allow a service provider operating in one workspace to offer its
capabilities to service consumers in other workspaces.
First we'll create an organization workspace, and then within that create a service provider workspace.
$ kubectl ws create wildwest --enter Workspace "wildwest" (type root:organization) created. Waiting for it to be ready... Workspace "wildwest" (type root:organization) is ready to use. Current workspace is "root:wildwest". $ kubectl ws create cowboys-service --enter Workspace "cowboys-service" (type root:universal) created. Waiting for it to be ready... Workspace "cowboys-service" (type root:universal) is ready to use. Current workspace is "root:wildwest:cowboys-service".
Then we'll use a CRD to generate an
APIExport and apply these within the service provider
apigentool used below can be found here. Builds for the tool are not currently published as part of the kcp release process.
$ mkdir wildwest-schemas/ $ ./bin/apigen --input-dir test/e2e/customresourcedefinition/ --output-dir wildwest-schemas/ $ ls -1 wildwest-schemas/ apiexport-wildwest.dev.yaml apiresourceschema-cowboys.wildwest.dev.yaml $ kubectl apply -f wildwest-schemas/apiresourceschema-cowboys.wildwest.dev.yaml apiresourceschema.apis.kcp.io/v220920-6039d110.cowboys.wildwest.dev created $ kubectl apply -f wildwest-schemas/apiexport-wildwest.dev.yaml apiexport.apis.kcp.io/wildwest.dev created
You can think of an
APIResourceSchema as being equivalent to a CRD, and an
APIExport makes a set of schemas
available to consumers.
Use those APIs as a service consumer
Now we can adopt the service consumer persona and create a workspace from which we will use this new
$ kubectl ws Current workspace is "root:users:zu:yc:kcp-admin". $ kubectl ws create --enter test-consumer Workspace "test-consumer" (type root:universal) created. Waiting for it to be ready... Workspace "test-consumer" (type root:universal) is ready to use. Current workspace is "root:users:zu:yc:kcp-admin:test-consumer". $ kubectl apply -f - <<EOF apiVersion: apis.kcp.io/v1alpha1 kind: APIBinding metadata: name: cowboys spec: reference: workspace: path: root:wildwest:cowboys-service exportName: wildwest.dev EOF apibinding.apis.kcp.io/cowboys created
Now this resource type is available for use within our workspace, so let's create an instance!
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
There is the possibility to further limit the access claim to single resources.
Dig deeper into APIExports
Switching back to the service provider persona:
$ kubectl ws root:wildwest:cowboys-service Current workspace is "root:wildwest:cowboys-service". $ kubectl get apiexport/wildwest.dev -o yaml apiVersion: apis.kcp.io/v1alpha1 kind: APIExport metadata: name: wildwest.dev ... status: ... identityHash: a6a0cc778bec8c4b844e6326965fbb740b6a9590963578afe07276e6a0d41e20
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.
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
APIExportEndpointSlice is created by a service provider, references a single
APIExport and optionally a
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
Looking at the status populated by the controller
$ 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:
The question is ... can we see the instance created by the consumer?
Q: Why is there a new
APIResourceSchema resource type that appears to be very similar to
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 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.
Q: How should we understand an
A: Unlike with CRDs, a kcp instance might have many
APIResourceSchemas of the same Group/Version/Kind, and users need
some way of securely distinguishing them.
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 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 service endpoint?
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 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 an empty list is returned instead of the error.