feat: implement decision-based routing with plugin architecture

Signed-off-by: bitliu <[email protected]>

Author: Xunzhuo

deploy/helm/semantic-router/README.md

@@ -6,6 +6,40 @@ A Helm chart for deploying Semantic Router - an intelligent routing system for L
 
 **Homepage:** <https://github.com/vllm-project/semantic-router>
 
+## CRD Management
+
+This Helm chart includes Custom Resource Definitions (CRDs) in the `crds/` directory:
+- `vllm.ai_intelligentpools.yaml` - IntelligentPool CRD
+- `vllm.ai_intelligentroutes.yaml` - IntelligentRoute CRD
+
+### Generating CRDs
+
+CRDs are automatically generated from Go type definitions using `controller-gen`. To regenerate CRDs:
+
+```bash
+# From the repository root
+make generate-crd
+```
+
+This command will:
+1. Generate CRDs from `src/semantic-router/pkg/apis/vllm.ai/v1alpha1` types
+2. Output to `deploy/kubernetes/crds/`
+3. Copy to `deploy/helm/semantic-router/crds/` for Helm chart
+
+### CRD Installation
+
+CRDs in the `crds/` directory are automatically installed by Helm:
+- Installed **before** other resources during `helm install`
+- **Not managed** by Helm (no Helm labels/annotations)
+- **Not updated** during `helm upgrade` (must be updated manually)
+- **Not deleted** during `helm uninstall` (protects custom resources)
+
+To manually update CRDs:
+
+```bash
+kubectl apply -f deploy/helm/semantic-router/crds/
+```
+
 ## Maintainers
 
 | Name | Email | Url |

deploy/kubernetes/crds/examples/intelligentpool-example.yaml

@@ -33,13 +33,25 @@ var _ ext_proc.ExternalProcessorServer = (*OpenAIRouter)(nil)
 
 // NewOpenAIRouter creates a new OpenAI API router instance
 func NewOpenAIRouter(configPath string) (*OpenAIRouter, error) {
-	// Always parse fresh config for router construction (supports live reload)
-	cfg, err := config.Parse(configPath)
-	if err != nil {
-		return nil, fmt.Errorf("failed to load config: %w", err)
+	var cfg *config.RouterConfig
+	var err error
+
+	// Check if we should use the global config (Kubernetes mode) or parse from file
+	globalCfg := config.Get()
+	if globalCfg != nil && globalCfg.ConfigSource == config.ConfigSourceKubernetes {
+		// Use the global config that's managed by the Kubernetes controller
+		cfg = globalCfg
+		logging.Infof("Using Kubernetes-managed configuration")
+	} else {
+		// Parse fresh config from file for file-based configuration (supports live reload)
+		cfg, err = config.Parse(configPath)
+		if err != nil {
+			return nil, fmt.Errorf("failed to load config: %w", err)
+		}
+		// Update global config reference for packages that rely on config.GetConfig()
+		config.Replace(cfg)
+		logging.Debugf("Parsed configuration from file: %s", configPath)
 	}
-	// Update global config reference for packages that rely on config.GetConfig()
-	config.Replace(cfg)
 
 	// Load category mapping if classifier is enabled
 	var categoryMapping *classification.CategoryMapping

deploy/kubernetes/crds/examples/intelligentroute-example.yaml

@@ -46,6 +46,11 @@ install-controller-gen: ## Install controller-gen for code generation
 generate-crd: install-controller-gen ## Generate CRD manifests using controller-gen
 	@echo "Generating CRD manifests..."
 	@cd src/semantic-router && controller-gen crd:crdVersions=v1,allowDangerousTypes=true paths=./pkg/apis/vllm.ai/v1alpha1 output:crd:artifacts:config=../../deploy/kubernetes/crds
+	@echo "Copying CRDs to Helm chart..."
+	@mkdir -p deploy/helm/semantic-router/crds
+	@cp deploy/kubernetes/crds/vllm.ai_intelligentpools.yaml deploy/helm/semantic-router/crds/
+	@cp deploy/kubernetes/crds/vllm.ai_intelligentroutes.yaml deploy/helm/semantic-router/crds/
+	@echo "✅ CRDs generated and copied to Helm chart"
 
 generate-deepcopy: install-controller-gen ## Generate deepcopy methods using controller-gen
 	@echo "Generating deepcopy methods..."