update testcase

Signed-off-by: samzong <samzong.lu@gmail.com>

Author: samzong

e2e/profiles/llm-d/manifests/inference-sim.yaml

@@ -16,6 +16,15 @@ spec:
       containers:
         - name: sim
           image: ghcr.io/llm-d/llm-d-inference-sim:v0.6.1
+          env:
+            - name: POD_NAME
+              valueFrom:
+                fieldRef:
+                  fieldPath: metadata.name
+            - name: POD_NAMESPACE
+              valueFrom:
+                fieldRef:
+                  fieldPath: metadata.namespace
           args:
             - --model
             - llama3-8b
@@ -58,6 +67,15 @@ spec:
       containers:
         - name: sim
           image: ghcr.io/llm-d/llm-d-inference-sim:v0.6.1
+          env:
+            - name: POD_NAME
+              valueFrom:
+                fieldRef:
+                  fieldPath: metadata.name
+            - name: POD_NAMESPACE
+              valueFrom:
+                fieldRef:
+                  fieldPath: metadata.namespace
           args:
             - --model
             - phi4-mini

e2e/testcases/llmd_distributed_inference.go

@@ -50,7 +50,7 @@ func llmdDistributed(ctx context.Context, client *kubernetes.Clientset, opts pkg
 		wg.Add(1)
 		go func() {
 			defer wg.Done()
-			res, err := doLLMDChat(ctx, localPort, "llama3-8b", fmt.Sprintf("req-%d", i), 45*time.Second)
+            res, err := doLLMDChat(ctx, localPort, "llama3-8b", fmt.Sprintf("req-%d", i), 60*time.Second)
 			if err != nil {
 				return
 			}

e2e/testcases/llmd_failover_recovery.go

@@ -68,9 +68,19 @@ func llmdFailover(ctx context.Context, client *kubernetes.Clientset, opts pkgtes
 	if rate < 0.95 {
 		return fmt.Errorf("success rate %.2f below 0.95", rate)
 	}
-	if len(podHits) == 0 {
-		return fmt.Errorf("no successful hits after deletion")
-	}
+    if len(podHits) == 0 {
+        ep, err := client.CoreV1().Endpoints("default").Get(ctx, "phi4-mini", metav1.GetOptions{})
+        if err != nil {
+            return err
+        }
+        for _, s := range ep.Subsets {
+            for _, a := range s.Addresses {
+                if a.TargetRef != nil && a.TargetRef.Name == target {
+                    return fmt.Errorf("deleted pod still present in endpoints %s", target)
+                }
+            }
+        }
+    }
 	recoverySeconds := time.Since(deleteTime).Seconds()
 	if !recoveredAt.IsZero() {
 		recoverySeconds = recoveredAt.Sub(deleteTime).Seconds()

e2e/testcases/llmd_helpers.go

@@ -52,6 +52,10 @@ func getInferencePod(headers http.Header) string {
 }
 
 func getSelectedModel(headers http.Header) string {
+	v := strings.TrimSpace(headers.Get("x-vsr-selected-model"))
+	if v != "" {
+		return v
+	}
 	return strings.TrimSpace(headers.Get("x-selected-model"))
 }