fix(istio): deploy vLLM backend for common E2E tests

Add Demo LLM (vLLM backend) deployment to Istio profile to fix
common E2E test failures.

Root cause: Istio profile was missing the vLLM backend that serves
LLM requests. This caused 404 errors on /v1/chat/completions and
0% accuracy on all common tests (domain-classify, pii-detection,
jailbreak-detection, etc.).

Changes:
- Add Step 4/6: Deploy Demo LLM backend (base-model.yaml)
- Add deployDemoLLM() function to deploy vLLM backend
- Add kubectl helper functions (kubectlApply, kubectlDelete, runKubectl)
- Update cleanup functions to remove Demo LLM on teardown
- Add timeoutDemoLLMDeploy constant for deployment timeout

This fix brings the Istio profile in line with AIBrix and AI-Gateway
profiles, which both deploy the vLLM backend for common tests.

Expected result: All 17 tests (4 Istio-specific + 13 common) should
now pass in CI.

Signed-off-by: Asaad Balum <[email protected]>

Author: asaadbalum

e2e/profiles/istio/profile.go

@@ -26,9 +26,13 @@ const (
 	semanticRouterDeployment = "semantic-router"
 	semanticRouterService    = "semantic-router"
 
+	// Demo LLM Configuration
+	demoLLMDeployment = "vllm-llama3-8b-instruct" // Demo LLM deployment name
+
 	// Timeouts
 	timeoutIstioInstall         = 5 * time.Minute
 	timeoutSemanticRouterDeploy = 10 * time.Minute
+	timeoutDemoLLMDeploy        = 10 * time.Minute
 	timeoutSidecarInjection     = 2 * time.Minute
 	timeoutGatewayReady         = 5 * time.Minute
 	timeoutStabilization        = 60 * time.Second
@@ -74,14 +78,15 @@ func (p *Profile) Setup(ctx context.Context, opts *framework.SetupOptions) error
 		istioInstalled         bool
 		namespaceConfigured    bool
 		semanticRouterDeployed bool
+		demoLLMDeployed        bool
 		istioResourcesCreated  bool
 	)
 
 	// Ensure cleanup on error
 	defer func() {
 		if r := recover(); r != nil {
 			p.log("Panic during setup, cleaning up...")
-			p.cleanupPartialDeployment(ctx, opts, istioInstalled, namespaceConfigured, semanticRouterDeployed, istioResourcesCreated)
+			p.cleanupPartialDeployment(ctx, opts, istioInstalled, namespaceConfigured, semanticRouterDeployed, demoLLMDeployed, istioResourcesCreated)
 			panic(r) // Re-panic after cleanup
 		}
 	}()
@@ -93,41 +98,49 @@ func (p *Profile) Setup(ctx context.Context, opts *framework.SetupOptions) error
 	}
 
 	// Step 1: Install Istio control plane
-	p.log("Step 1/5: Installing Istio control plane (version: %s)", istioVersion)
+	p.log("Step 1/6: Installing Istio control plane (version: %s)", istioVersion)
 	if err := p.installIstio(ctx, opts); err != nil {
 		return fmt.Errorf("failed to install Istio: %w", err)
 	}
 	istioInstalled = true
 
 	// Step 2: Configure namespace with sidecar injection
-	p.log("Step 2/5: Configuring namespace for sidecar injection")
+	p.log("Step 2/6: Configuring namespace for sidecar injection")
 	if err := p.configureNamespace(ctx, opts); err != nil {
-		p.cleanupPartialDeployment(ctx, opts, istioInstalled, false, false, false)
+		p.cleanupPartialDeployment(ctx, opts, istioInstalled, false, false, false, false)
 		return fmt.Errorf("failed to configure namespace: %w", err)
 	}
 	namespaceConfigured = true
 
 	// Step 3: Deploy Semantic Router with Istio sidecar
-	p.log("Step 3/5: Deploying Semantic Router")
+	p.log("Step 3/6: Deploying Semantic Router")
 	if err := p.deploySemanticRouter(ctx, deployer, opts); err != nil {
-		p.cleanupPartialDeployment(ctx, opts, istioInstalled, namespaceConfigured, false, false)
+		p.cleanupPartialDeployment(ctx, opts, istioInstalled, namespaceConfigured, false, false, false)
 		return fmt.Errorf("failed to deploy semantic router: %w", err)
 	}
 	semanticRouterDeployed = true
 
-	// Step 4: Create Istio Gateway and VirtualService
-	p.log("Step 4/5: Creating Istio Gateway and VirtualService")
+	// Step 4: Deploy Demo LLM (vLLM backend)
+	p.log("Step 4/6: Deploying Demo LLM backend")
+	if err := p.deployDemoLLM(ctx, opts); err != nil {
+		p.cleanupPartialDeployment(ctx, opts, istioInstalled, namespaceConfigured, semanticRouterDeployed, false, false)
+		return fmt.Errorf("failed to deploy demo LLM: %w", err)
+	}
+	demoLLMDeployed = true
+
+	// Step 5: Create Istio Gateway and VirtualService
+	p.log("Step 5/6: Creating Istio Gateway and VirtualService")
 	if err := p.createIstioResources(ctx, opts); err != nil {
-		p.cleanupPartialDeployment(ctx, opts, istioInstalled, namespaceConfigured, semanticRouterDeployed, false)
+		p.cleanupPartialDeployment(ctx, opts, istioInstalled, namespaceConfigured, semanticRouterDeployed, demoLLMDeployed, false)
 		return fmt.Errorf("failed to create Istio resources: %w", err)
 	}
 	istioResourcesCreated = true
 
-	// Step 5: Verify environment is ready
-	p.log("Step 5/5: Verifying environment")
+	// Step 6: Verify environment is ready
+	p.log("Step 6/6: Verifying environment")
 	if err := p.verifyEnvironment(ctx, opts); err != nil {
 		p.log("ERROR: Environment verification failed: %v", err)
-		p.cleanupPartialDeployment(ctx, opts, istioInstalled, namespaceConfigured, semanticRouterDeployed, istioResourcesCreated)
+		p.cleanupPartialDeployment(ctx, opts, istioInstalled, namespaceConfigured, semanticRouterDeployed, demoLLMDeployed, istioResourcesCreated)
 		return fmt.Errorf("failed to verify environment: %w", err)
 	}
 
@@ -146,6 +159,9 @@ func (p *Profile) Teardown(ctx context.Context, opts *framework.TeardownOptions)
 	p.log("Cleaning up Istio resources")
 	p.cleanupIstioResources(ctx, opts)
 
+	p.log("Cleaning up Demo LLM")
+	p.kubectlDelete(ctx, opts.KubeConfig, "deploy/kubernetes/aibrix/aigw-resources/base-model.yaml")
+
 	p.log("Uninstalling Semantic Router")
 	deployer.Uninstall(ctx, semanticRouterDeployment, semanticRouterNamespace)
 
@@ -529,7 +545,7 @@ func (p *Profile) verifyDeployment(ctx context.Context, opts *framework.SetupOpt
 	return nil
 }
 
-func (p *Profile) cleanupPartialDeployment(ctx context.Context, opts *framework.SetupOptions, istioInstalled, namespaceConfigured, semanticRouterDeployed, istioResourcesCreated bool) {
+func (p *Profile) cleanupPartialDeployment(ctx context.Context, opts *framework.SetupOptions, istioInstalled, namespaceConfigured, semanticRouterDeployed, demoLLMDeployed, istioResourcesCreated bool) {
 	p.log("Cleaning up partial deployment...")
 
 	teardownOpts := &framework.TeardownOptions{
@@ -543,6 +559,11 @@ func (p *Profile) cleanupPartialDeployment(ctx context.Context, opts *framework.
 		p.cleanupIstioResources(ctx, teardownOpts)
 	}
 
+	if demoLLMDeployed {
+		p.log("Cleaning up Demo LLM")
+		p.kubectlDelete(ctx, opts.KubeConfig, "deploy/kubernetes/aibrix/aigw-resources/base-model.yaml")
+	}
+
 	if semanticRouterDeployed {
 		deployer := helm.NewDeployer(opts.KubeConfig, opts.Verbose)
 		deployer.Uninstall(ctx, semanticRouterDeployment, semanticRouterNamespace)
@@ -631,3 +652,40 @@ func containsMiddle(s, substr string) bool {
 	}
 	return false
 }
+
+// deployDemoLLM deploys the demo LLM (vLLM backend) for testing
+func (p *Profile) deployDemoLLM(ctx context.Context, opts *framework.SetupOptions) error {
+	// Apply base model (Demo LLM)
+	if err := p.kubectlApply(ctx, opts.KubeConfig, "deploy/kubernetes/aibrix/aigw-resources/base-model.yaml"); err != nil {
+		return fmt.Errorf("failed to apply base model: %w", err)
+	}
+
+	// Wait for Demo LLM deployment
+	p.log("Waiting for Demo LLM deployment to be ready...")
+	if err := p.waitForDeployment(ctx, opts, semanticRouterNamespace, demoLLMDeployment, timeoutDemoLLMDeploy); err != nil {
+		return fmt.Errorf("demo LLM deployment not ready: %w", err)
+	}
+
+	return nil
+}
+
+// kubectlApply applies a Kubernetes manifest
+func (p *Profile) kubectlApply(ctx context.Context, kubeConfig, manifest string) error {
+	return p.runKubectl(ctx, kubeConfig, "apply", "--server-side", "-f", manifest)
+}
+
+// kubectlDelete deletes a Kubernetes manifest
+func (p *Profile) kubectlDelete(ctx context.Context, kubeConfig, manifest string) error {
+	return p.runKubectl(ctx, kubeConfig, "delete", "-f", manifest)
+}
+
+// runKubectl runs a kubectl command
+func (p *Profile) runKubectl(ctx context.Context, kubeConfig string, args ...string) error {
+	args = append([]string{"--kubeconfig", kubeConfig}, args...)
+	cmd := exec.CommandContext(ctx, "kubectl", args...)
+	if p.verbose {
+		cmd.Stdout = os.Stdout
+		cmd.Stderr = os.Stderr
+	}
+	return cmd.Run()
+}