Support installing individual controllers from giant controller-manag…

…er (#442)

* standalone-pod-autoscaler

Co-authored-by: nwangfw <113055713+nwangfw@users.noreply.github.com>

* Move files to right overlays

* Refactor the crd folder structure for better reuse

* Support running controllers as standalone component

---------

Co-authored-by: Jiaxin Shan <seedjeffwan@gmail.com>

build/container/Dockerfile

@@ -14,11 +14,7 @@ RUN go mod download
 # Copy the go source
 COPY cmd/ cmd/
 COPY api/ api/
-COPY pkg/controller/ pkg/controller/
-COPY pkg/utils/ pkg/utils/
-COPY pkg/cache/ pkg/cache/
-COPY pkg/client/ pkg/client/
-COPY pkg/metrics/ pkg/metrics/
+COPY pkg/ pkg/
 
 # Build
 # the GOARCH has not a default value to allow the binary be built according to the host where the command

build/container/Dockerfile.gateway

@@ -14,11 +14,7 @@ RUN go mod download
 # Copy the go source
 COPY cmd/ cmd/
 COPY api/ api/
-COPY pkg/plugins/ pkg/plugins/
-COPY pkg/utils/ pkg/utils/
-COPY pkg/cache/ pkg/cache/
-COPY pkg/client/ pkg/client/
-COPY pkg/metrics/ pkg/metrics/
+COPY pkg/ pkg/
 
 # Build
 # the GOARCH has not a default value to allow the binary be built according to the host where the command

cmd/controllers/main.go

@@ -23,13 +23,17 @@ import (
 	"os"
 	"time"
 
+	clientgoscheme "k8s.io/client-go/kubernetes/scheme"
+
+	autoscalingv1alpha1 "github.com/aibrix/aibrix/api/autoscaling/v1alpha1"
+	modelv1alpha1 "github.com/aibrix/aibrix/api/model/v1alpha1"
+	orchestrationv1alpha1 "github.com/aibrix/aibrix/api/orchestration/v1alpha1"
+	"github.com/aibrix/aibrix/pkg/features"
 	rayclusterv1 "github.com/ray-project/kuberay/ray-operator/apis/ray/v1"
-	// Import all Kubernetes client auth plugins (e.g. Azure, GCP, OIDC, etc.)
-	// to ensure that exec-entrypoint and run can make use of them.
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+
 	"k8s.io/apimachinery/pkg/runtime"
 	utilruntime "k8s.io/apimachinery/pkg/util/runtime"
-	clientgoscheme "k8s.io/client-go/kubernetes/scheme"
 	_ "k8s.io/client-go/plugin/pkg/client/auth"
 	"k8s.io/client-go/rest"
 	"k8s.io/client-go/tools/clientcmd"
@@ -41,9 +45,6 @@ import (
 	metricsserver "sigs.k8s.io/controller-runtime/pkg/metrics/server"
 	"sigs.k8s.io/controller-runtime/pkg/webhook"
 
-	autoscalingv1alpha1 "github.com/aibrix/aibrix/api/autoscaling/v1alpha1"
-	modelv1alpha1 "github.com/aibrix/aibrix/api/model/v1alpha1"
-	orchestrationv1alpha1 "github.com/aibrix/aibrix/api/orchestration/v1alpha1"
 	"github.com/aibrix/aibrix/pkg/cache"
 	"github.com/aibrix/aibrix/pkg/controller"
 	//+kubebuilder:scaffold:imports
@@ -64,15 +65,32 @@ var (
 )
 
 func init() {
+	// Only register the base kubernetes scheme here
 	utilruntime.Must(clientgoscheme.AddToScheme(scheme))
 
-	utilruntime.Must(autoscalingv1alpha1.AddToScheme(scheme))
-	utilruntime.Must(modelv1alpha1.AddToScheme(scheme))
+	scheme.AddUnversionedTypes(metav1.SchemeGroupVersion, &metav1.UpdateOptions{}, &metav1.DeleteOptions{}, &metav1.CreateOptions{})
+	//+kubebuilder:scaffold:scheme
+}
+
+func RegisterSchemas(scheme *runtime.Scheme) error {
+
+	if features.IsControllerEnabled(features.PodAutoscalerController) {
+		utilruntime.Must(autoscalingv1alpha1.AddToScheme(scheme))
+	}
+
+	if features.IsControllerEnabled(features.ModelAdapterController) {
+		utilruntime.Must(modelv1alpha1.AddToScheme(scheme))
+	}
+
+	if features.IsControllerEnabled(features.DistributedInferenceController) {
+		utilruntime.Must(orchestrationv1alpha1.AddToScheme(scheme))
+		utilruntime.Must(rayclusterv1.AddToScheme(scheme))
+	}
 
 	scheme.AddUnversionedTypes(metav1.SchemeGroupVersion, &metav1.UpdateOptions{}, &metav1.DeleteOptions{}, &metav1.CreateOptions{})
-	utilruntime.Must(orchestrationv1alpha1.AddToScheme(scheme))
-	utilruntime.Must(rayclusterv1.AddToScheme(scheme))
 	//+kubebuilder:scaffold:scheme
+
+	return nil
 }
 
 func main() {
@@ -86,6 +104,7 @@ func main() {
 	var renewDeadLine time.Duration
 	var leaderElectionResourceLock string
 	var leaderElectionId string
+	var controllers string
 	flag.StringVar(&metricsAddr, "metrics-bind-address", ":8080", "The address the metric endpoint binds to.")
 	flag.StringVar(&probeAddr, "health-probe-bind-address", ":8081", "The address the probe endpoint binds to.")
 	flag.BoolVar(&enableLeaderElection, "leader-elect", false,
@@ -106,6 +125,7 @@ func main() {
 		"leader-election-resource-lock determines which resource lock to use for leader election, defaults to \"leases\".")
 	flag.StringVar(&leaderElectionId, "leader-election-id", "aibrix-controller-manager",
 		"leader-election-id determines the name of the resource that leader election will use for holding the leader lock, Default is aibrix-controller-manager.")
+	flag.StringVar(&controllers, "controllers", "*", "Comma-separated list of controllers to enable or disable, default value is * which indicates all controllers should be started.")
 
 	// Initialize the klog
 	klog.InitFlags(flag.CommandLine)
@@ -115,6 +135,19 @@ func main() {
 	// TODO: we will switch to textlogger or zap later
 	ctrl.SetLogger(klogr.New()) // nolint:staticcheck
 
+	// initialize the controllers
+	if err := features.ValidateControllers(controllers); err != nil {
+		setupLog.Error(err, "unable to validate the controllers, please type the right controller names through --controllers")
+		os.Exit(1)
+	}
+
+	features.InitControllers(controllers)
+
+	if err := RegisterSchemas(scheme); err != nil {
+		setupLog.Error(err, "unable to register schemas")
+		os.Exit(1)
+	}
+
 	// if the enable-http2 flag is false (the default), http/2 should be disabled
 	// due to its vulnerabilities. More specifically, disabling http/2 will
 	// prevent from being vulnerable to the HTTP/2 Stream Cancellation and
@@ -186,7 +219,13 @@ func main() {
 		panic(err)
 	}
 
-	cache.NewCache(config, stopCh, nil)
+	if features.IsControllerEnabled(features.ModelAdapterController) {
+		// cache is enabled for model adapter scheduling.
+		cache.NewCache(config, stopCh, nil)
+	}
+
+	// Initialize controllers
+	controller.Initialize()
 
 	// Kind controller registration is encapsulated inside the pkg/controller/controller.go
 	// So here we can use more clean registration flow and there's no need to change logics in future.

config/crd/autoscaling/kustomization.yaml

@@ -0,0 +1,2 @@
+resources:
+- autoscaling.aibrix.ai_podautoscalers.yaml

config/crd/kustomization.yaml

@@ -2,10 +2,9 @@
 # since it depends on service name and namespace that are out of this kustomize package.
 # It should be run by config/default
 resources:
-- bases/autoscaling.aibrix.ai_podautoscalers.yaml
-- bases/model.aibrix.ai_modeladapters.yaml
-- bases/orchestration.aibrix.ai_rayclusterreplicasets.yaml
-- bases/orchestration.aibrix.ai_rayclusterfleets.yaml
+- autoscaling
+- model
+- orchestration
 #+kubebuilder:scaffold:crdkustomizeresource
 
 patches:

config/crd/model/kustomization.yaml

@@ -0,0 +1,2 @@
+resources:
+- model.aibrix.ai_modeladapters.yaml

config/crd/orchestration/kustomization.yaml

@@ -0,0 +1,3 @@
+resources:
+- orchestration.aibrix.ai_rayclusterreplicasets.yaml
+- orchestration.aibrix.ai_rayclusterfleets.yaml

config/default/kustomization.yaml

@@ -15,6 +15,7 @@ namePrefix: aibrix-
 #    someName: someValue
 
 resources:
+- namespace.yaml
 - ../crd
 - ../rbac
 - ../manager

config/default/namespace.yaml

@@ -0,0 +1,8 @@
+apiVersion: v1
+kind: Namespace
+metadata:
+  labels:
+    control-plane: controller-manager
+    app.kubernetes.io/name: aibrix
+    app.kubernetes.io/managed-by: kustomize
+  name: system

config/manager/manager.yaml

@@ -1,11 +1,3 @@
-apiVersion: v1
-kind: Namespace
-metadata:
-  labels:
-    control-plane: controller-manager
-    app.kubernetes.io/name: aibrix
-    app.kubernetes.io/managed-by: kustomize
-  name: system
 ---
 apiVersion: apps/v1
 kind: Deployment
@@ -66,3 +58,21 @@ spec:
             memory: 64Mi
       serviceAccountName: controller-manager
       terminationGracePeriodSeconds: 10
+---
+apiVersion: v1
+kind: Service
+metadata:
+  labels:
+    control-plane: controller-manager
+    app.kubernetes.io/name: aibrix
+    app.kubernetes.io/managed-by: kustomize
+  name: controller-manager-metrics-service
+  namespace: system
+spec:
+  ports:
+    - name: http
+      port: 8080
+      protocol: TCP
+      targetPort: 8080
+  selector:
+    control-plane: controller-manager

config/rbac/autoscaling/kustomization.yaml

@@ -0,0 +1,3 @@
+resources:
+- autoscaling_podautoscaler_editor_role.yaml
+- autoscaling_podautoscaler_viewer_role.yaml

config/rbac/controller-manager/kustomization.yaml

@@ -0,0 +1,8 @@
+# controller manager are common sa, role, rolebindings that could be reused
+
+resources:
+  - leader_election_role.yaml
+  - leader_election_role_binding.yaml
+  - role.yaml
+  - role_binding.yaml
+  - service_account_controller_manager.yaml

config/rbac/role_binding.yaml → config/rbac/controller-manager/role_binding.yaml

@@ -13,19 +13,3 @@ subjects:
 - kind: ServiceAccount
   name: controller-manager
   namespace: system
----
-apiVersion: rbac.authorization.k8s.io/v1
-kind: ClusterRoleBinding
-metadata:
-  labels:
-    app.kubernetes.io/name: aibrix
-    app.kubernetes.io/managed-by: kustomize
-  name: gateway-plugin-rolebinding
-roleRef:
-  apiGroup: rbac.authorization.k8s.io
-  kind: ClusterRole
-  name: gateway-plugin-role
-subjects:
-- kind: ServiceAccount
-  name: gateway-plugin
-  namespace: system

config/rbac/service_account.yaml → config/rbac/controller-manager/service_account_controller_manager.yaml

@@ -6,12 +6,3 @@ metadata:
     app.kubernetes.io/managed-by: kustomize
   name: controller-manager
   namespace: system
----
-apiVersion: v1
-kind: ServiceAccount
-metadata:
-  labels:
-    app.kubernetes.io/name: aibrix
-    app.kubernetes.io/managed-by: kustomize
-  name: gateway-plugin
-  namespace: system

config/rbac/gateway/kustomization.yaml

@@ -0,0 +1,5 @@
+resources:
+# Unlike others files are generated. Gateway plugin role, role binding and serving account files are maintained by our own.
+- service_account_gateway.yaml
+- role_gateway.yaml
+- role_binding_gateway.yaml

config/rbac/gateway/role_binding_gateway.yaml

@@ -0,0 +1,15 @@
+apiVersion: rbac.authorization.k8s.io/v1
+kind: ClusterRoleBinding
+metadata:
+  labels:
+    app.kubernetes.io/name: aibrix
+    app.kubernetes.io/managed-by: kustomize
+  name: gateway-plugin-rolebinding
+roleRef:
+  apiGroup: rbac.authorization.k8s.io
+  kind: ClusterRole
+  name: gateway-plugin-role
+subjects:
+- kind: ServiceAccount
+  name: gateway-plugin
+  namespace: system

config/rbac/gateway/service_account_gateway.yaml

@@ -0,0 +1,8 @@
+apiVersion: v1
+kind: ServiceAccount
+metadata:
+  labels:
+    app.kubernetes.io/name: aibrix
+    app.kubernetes.io/managed-by: kustomize
+  name: gateway-plugin
+  namespace: system

config/rbac/kustomization.yaml

@@ -4,23 +4,18 @@ resources:
 # if your manager will use a service account that exists at
 # runtime. Be sure to update RoleBinding and ClusterRoleBinding
 # subjects if changing service account names.
-- service_account.yaml
-- role.yaml
-- role_gateway.yaml
-- role_binding.yaml
-- leader_election_role.yaml
-- leader_election_role_binding.yaml
-- metrics_service.yaml
+- controller-manager
 # For each CRD, "Editor" and "Viewer" roles are scaffolded by
 # default, aiding admins in cluster management. Those roles are
 # not used by the Project itself. You can comment the following lines
 # if you do not want those helpers be installed with your Project.
-- orchestration_rayclusterfleet_editor_role.yaml
-- orchestration_rayclusterfleet_viewer_role.yaml
-- orchestration_rayclusterreplicaset_editor_role.yaml
-- orchestration_rayclusterreplicaset_viewer_role.yaml
-- model_modeladapter_editor_role.yaml
-- model_modeladapter_viewer_role.yaml
-- autoscaling_podautoscaler_editor_role.yaml
-- autoscaling_podautoscaler_viewer_role.yaml
-
+- orchestration/orchestration_rayclusterfleet_editor_role.yaml
+- orchestration/orchestration_rayclusterfleet_viewer_role.yaml
+- orchestration/orchestration_rayclusterreplicaset_editor_role.yaml
+- orchestration/orchestration_rayclusterreplicaset_viewer_role.yaml
+- model/model_modeladapter_editor_role.yaml
+- model/model_modeladapter_viewer_role.yaml
+- autoscaling/autoscaling_podautoscaler_editor_role.yaml
+- autoscaling/autoscaling_podautoscaler_viewer_role.yaml
+# other components
+- gateway

config/rbac/model/kustomization.yaml

@@ -0,0 +1,3 @@
+resources:
+- model_modeladapter_editor_role.yaml
+- model_modeladapter_viewer_role.yaml

config/rbac/orchestration/kustomization.yaml

@@ -0,0 +1,5 @@
+resources:
+- orchestration_rayclusterfleet_editor_role.yaml
+- orchestration_rayclusterfleet_viewer_role.yaml
+- orchestration_rayclusterreplicaset_editor_role.yaml
+- orchestration_rayclusterreplicaset_viewer_role.yaml

config/standalone/autoscaler-controller/kustomization.yaml

@@ -0,0 +1,31 @@
+apiVersion: kustomize.config.k8s.io/v1beta1
+kind: Kustomization
+
+resources:
+  - ../../crd/autoscaling
+  - ../../rbac/autoscaling
+  - ../../rbac/controller-manager
+  - ../../manager
+
+# Adds namespace to all resources.
+namespace: aibrix-system
+
+# Value of this field is prepended to the
+# names of all resources, e.g. a deployment named
+# "wordpress" becomes "alices-wordpress".
+# Note that it should also match with the prefix (text before '-') of the namespace
+# field above.
+namePrefix: aibrix-autoscaling-
+
+images:
+  - name: controller
+    newName: aibrix/controller-manager
+    newTag: nightly
+
+patches:
+  - path: patch.yaml
+    target:
+      group: apps
+      version: v1
+      kind: Deployment
+      name: controller-manager