add skip env

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

Author: samzong

e2e/profiles/llm-d/profile.go

@@ -7,6 +7,7 @@ import (
 	"os/exec"
 	"path/filepath"
 	"runtime"
+	"strings"
 	"time"
 
 	"k8s.io/client-go/kubernetes"
@@ -30,11 +31,22 @@ const (
 )
 
 type Profile struct {
-	verbose bool
+	verbose     bool
+	useExisting bool
+	skipSetup   bool
+	versions    struct {
+		istio     string
+		gateway   string
+		inference string
+	}
 }
 
 func NewProfile() *Profile {
-	return &Profile{}
+	p := &Profile{}
+	p.versions.istio = istioVersion
+	p.versions.gateway = gatewayCRDURL
+	p.versions.inference = inferenceCRDURL
+	return p
 }
 
 func (p *Profile) Name() string {
@@ -47,47 +59,121 @@ func (p *Profile) Description() string {
 
 func (p *Profile) Setup(ctx context.Context, opts *framework.SetupOptions) error {
 	p.verbose = opts.Verbose
+	p.useExisting = strings.EqualFold(os.Getenv("LLMD_USE_EXISTING"), "true") || os.Getenv("LLMD_USE_EXISTING") == "1"
+	p.skipSetup = strings.EqualFold(os.Getenv("LLMD_SKIP_SETUP"), "true") || os.Getenv("LLMD_SKIP_SETUP") == "1"
+
+	fmt.Printf("[Profile] llm-d setup start (istio=%s, gatewayCRD=%s, inferenceCRD=%s, useExisting=%v, skipSetup=%v)\n",
+		p.versions.istio, p.versions.gateway, p.versions.inference, p.useExisting, p.skipSetup)
+
+	if p.skipSetup {
+		fmt.Println("[Profile] LLMD_SKIP_SETUP set; skipping deploy steps, running verification only")
+		return p.verifyEnvironment(ctx, opts)
+	}
+
+	rollback := []func(){}
+	rollbackAll := func() {
+		for i := len(rollback) - 1; i >= 0; i-- {
+			rollback[i]()
+		}
+	}
+
 	istioctlPath, err := p.ensureIstioctl(ctx)
 	if err != nil {
 		return err
 	}
+	if p.verbose {
+		fmt.Printf("[Profile] istioctl ready at %s\n", istioctlPath)
+	}
 
 	if err := p.kubectlApply(ctx, gatewayCRDURL); err != nil {
 		return fmt.Errorf("gateway CRDs: %w", err)
 	}
+	rollback = append(rollback, func() { _ = p.kubectlDelete(ctx, gatewayCRDURL) })
+	if p.verbose {
+		fmt.Println("[Profile] applied gateway CRDs")
+	}
 	if err := p.kubectlApply(ctx, inferenceCRDURL); err != nil {
+		rollbackAll()
 		return fmt.Errorf("inference CRDs: %w", err)
 	}
+	rollback = append(rollback, func() { _ = p.kubectlDelete(ctx, inferenceCRDURL) })
+	if p.verbose {
+		fmt.Println("[Profile] applied inference CRDs")
+	}
 
 	if err := p.installIstio(ctx, istioctlPath); err != nil {
+		rollbackAll()
 		return fmt.Errorf("install istio: %w", err)
 	}
+	rollback = append(rollback, func() { _ = p.uninstallIstio(ctx) })
+	if p.verbose {
+		fmt.Println("[Profile] istio installed")
+	}
 
 	if err := p.deploySemanticRouter(ctx, opts); err != nil {
+		rollbackAll()
 		return fmt.Errorf("deploy semantic router: %w", err)
 	}
+	rollback = append(rollback, func() {
+		deployer := helm.NewDeployer(opts.KubeConfig, opts.Verbose)
+		_ = deployer.Uninstall(ctx, "semantic-router", semanticNamespace)
+	})
+	if p.verbose {
+		fmt.Println("[Profile] semantic-router deployed")
+	}
 
 	if err := p.deployInferenceSim(ctx, opts); err != nil {
+		rollbackAll()
 		return fmt.Errorf("deploy inference sim: %w", err)
 	}
+	rollback = append(rollback, func() { _ = p.kubectlDelete(ctx, "e2e/profiles/llm-d/manifests/inference-sim.yaml") })
+	if p.verbose {
+		fmt.Println("[Profile] inference simulators deployed")
+	}
 
 	if err := p.deployLLMD(ctx); err != nil {
+		rollbackAll()
 		return fmt.Errorf("deploy llm-d resources: %w", err)
 	}
+	rollback = append(rollback, func() {
+		_ = p.kubectlDelete(ctx, "e2e/profiles/llm-d/manifests/rbac.yaml")
+		_ = p.kubectlDelete(ctx, "deploy/kubernetes/llmd-base/dest-rule-epp-llama.yaml")
+		_ = p.kubectlDelete(ctx, "deploy/kubernetes/llmd-base/dest-rule-epp-phi4.yaml")
+		_ = p.kubectlDelete(ctx, "deploy/kubernetes/llmd-base/inferencepool-llama.yaml")
+		_ = p.kubectlDelete(ctx, "deploy/kubernetes/llmd-base/inferencepool-phi4.yaml")
+	})
+	if p.verbose {
+		fmt.Println("[Profile] llm-d schedulers and pools deployed")
+	}
 
 	if err := p.deployGatewayRoutes(ctx); err != nil {
+		rollbackAll()
 		return fmt.Errorf("deploy gateway routes: %w", err)
 	}
+	rollback = append(rollback, func() {
+		_ = p.kubectlDelete(ctx, "deploy/kubernetes/istio/envoyfilter.yaml")
+		_ = p.kubectlDelete(ctx, "deploy/kubernetes/istio/destinationrule.yaml")
+		_ = p.kubectlDelete(ctx, "e2e/profiles/llm-d/manifests/httproute-services.yaml")
+		_ = p.kubectlDelete(ctx, "deploy/kubernetes/istio/gateway.yaml")
+	})
+	if p.verbose {
+		fmt.Println("[Profile] gateway routes deployed")
+	}
 
 	if err := p.verifyEnvironment(ctx, opts); err != nil {
+		rollbackAll()
 		return fmt.Errorf("verify environment: %w", err)
 	}
 
+	if p.verbose {
+		fmt.Println("[Profile] llm-d setup complete")
+	}
 	return nil
 }
 
 func (p *Profile) Teardown(ctx context.Context, opts *framework.TeardownOptions) error {
 	p.verbose = opts.Verbose
+	fmt.Println("[Profile] llm-d teardown start")
 	_ = p.kubectlDelete(ctx, "e2e/profiles/llm-d/manifests/httproute-services.yaml")
 	_ = p.kubectlDelete(ctx, "deploy/kubernetes/llmd-base/dest-rule-epp-llama.yaml")
 	_ = p.kubectlDelete(ctx, "deploy/kubernetes/llmd-base/dest-rule-epp-phi4.yaml")
@@ -103,18 +189,35 @@ func (p *Profile) Teardown(ctx context.Context, opts *framework.TeardownOptions)
 	deployer.Uninstall(ctx, "semantic-router", semanticNamespace)
 
 	_ = p.uninstallIstio(ctx)
+	_ = p.kubectlDelete(ctx, gatewayCRDURL)
+	_ = p.kubectlDelete(ctx, inferenceCRDURL)
+	fmt.Println("[Profile] llm-d teardown complete")
 
 	return nil
 }
 
 func (p *Profile) GetTestCases() []string {
-	return []string{
+	tests := []string{
 		"llmd-health-check",
 		"llmd-distributed-inference",
 		"llmd-auto-routing",
 		"llmd-failover-recovery",
 		"llmd-performance-baseline",
 	}
+	if strings.EqualFold(os.Getenv("LLMD_PERF_SKIP"), "true") || os.Getenv("LLMD_PERF_SKIP") == "1" {
+		var filtered []string
+		for _, t := range tests {
+			if t == "llmd-performance-baseline" {
+				continue
+			}
+			filtered = append(filtered, t)
+		}
+		if p.verbose {
+			fmt.Println("[Profile] LLMD_PERF_SKIP set; skipping llmd-performance-baseline test")
+		}
+		return filtered
+	}
+	return tests
 }
 
 func (p *Profile) GetServiceConfig() framework.ServiceConfig {