## Data Plane Aggregator The data plane aggregator is a reverse proxy similar to `kube-aggregator` that sits first in the request path. It is responsible for reverse proxying any registered data plane (anything that isn't CRUD + List + Watch) routes to the appropriate app handlers. Anything that does not match the registered data planeroutes will be delegated to the next apiserver in the handler chain (e.g. `kube-aggregator`). Currently the aggregator uses the Grafana plugin framework to register and proxy requests to Grafana plugins. The benefit of this approach is that it allows apps to be built, packaged, published, and deployed using the existing Grafana plugin framework. ![Request Flow](./requestflow.png) ### Architecture The data plane aggregator is built on top of the Kubernetes API server framework. It introduces a new `DataPlaneService` Custom Resource Definition (CRD) to dynamically register and proxy requests to Grafana plugins. The main components are: * **GrafanaAggregator**: The main server component, located in `pkg/aggregator/apiserver/apiserver.go`. It's responsible for: * Creating a generic API server. * Registering the `DataPlaneService` resource. * Running the `DataPlaneServiceRegistrationController`. * **DataPlaneServiceRegistrationController**: This controller, located in `pkg/aggregator/apiserver/dataplaneservice_controller.go`, is responsible for: * Watching for `add`, `update`, and `delete` events on `DataPlaneService` resources. * Passing the `DataPlaneService` to the `PluginHandler` to add, update, or remove the corresponding proxy handlers. * **PluginHandler**: This handler, located in `pkg/aggregator/apiserver/plugin/handler.go`, is responsible for: * Translating between the HTTP request/response and Grafana plugin request/response. * Adding and removing HTTP handlers to proxy requests to plugins based on the registered `DataPlaneService` resources. * **DataPlaneService**: Defined in `pkg/aggregator/apis/aggregation/v0alpha1/types.go`, this represents a service provided by a Grafana plugin. The following diagram illustrates the request flow: ```mermaid sequenceDiagram autonumber participant plugin as disk participant exe as plugin
binaries participant wrap as apiserver participant agg as kube-aggregator participant data as dataplane-aggregator participant storage as Unified
Storage Note over plugin,storage: Registration wrap->>plugin: Load Scope app manifest wrap->>agg: Create Scope APIService wrap->>plugin: Load SLO app manifest wrap->>agg: Create SLO APIService wrap->>plugin: Load Dashboard app manifest wrap->>agg: Create Dashboard APIService wrap->>data: Create Dashboard DataPlaneService Note over plugin,storage: Storage Access data->>storage: DataPlaneServices agg->>storage: APIServices wrap->>storage: Custom resources Note over plugin,storage: Create, Update, or Delete Hooks data->>agg: Local delegation to kube-aggregator agg->>wrap: HTTP reverse proxy incoming request wrap->>exe: gRPC admission mutation hook wrap->>exe: gRPC admission validation hook Note over plugin,storage: Get, List, or Watch Hooks data->>agg: Local delegation to kube-aggregator agg->>wrap: HTTP reverse proxy incoming request wrap->>exe: gRPC conversion hook Note over plugin,storage: Subresource & Custom Routes data->>exe: gRPC plugin resource calls data->>exe: gRPC plugin query data data->>exe: gRPC plugin stream data->>exe: gRPC plugin health data->>exe: gRPC plugin metrics Note over plugin,storage: Async (controllers) exe->>data: Watch data->>agg: Local delegation to kube-aggregator agg->>wrap: HTTP reverse proxy incoming request wrap->>storage: Watch ``` ## DataPlaneService A `DataPlaneService` is a non-namespaced resource that represents a service exposed by a Grafana plugin. Here is an example: ```yaml apiVersion: aggregation.grafana.app/v0alpha1 kind: DataPlaneService metadata: name: spec: pluginID: pluginType: # app or datasource group: version: services: - type: # admission, conversion, query, stream, route, datasource-proxy method: path: ``` ### Query Example 1. `custom.ini` changes: ```ini [feature_toggles] kubernetesAggregator = true dataplaneAggregator = true grafanaAPIServerEnsureKubectlAccess = true ``` 2. start grafana: ```bash make run ``` 3. enable aggregation for prometheus data source: ```bash export KUBECONFIG=./data/grafana-apiserver/grafana.kubeconfig kubectl apply -f pkg/aggregator/examples/datasource.yml --validate=false dataplaneservice.aggregation.grafana.app/v0alpha1.prometheus.grafana.app created ``` 4. edit `pkg/aggregator/examples/datasource-query.json` and update the datasource UID to match the UID of a prometheus data source. 5. execute query (replace `example` with the UID of a prometheus data source): ```bash curl 'http://admin:admin@localhost:3000/apis/prometheus.grafana.app/v0alpha1/namespaces/default/connections/example/query' -X POST -d '@pkg/aggregator/examples/datasource-query.json' ``` ### Research/PoC's * [PoC Alternate HTTP backend](https://github.com/grafana/grafana/pull/100697) * [PoC Plugin SDK <--> App SDK Adapter](https://github.com/grafana/grafana-app-sdk/pull/493)