Merge branch 'release-v0.15' into RHOAIENG-34472-dev

Signed-off-by: Killian Golds <[email protected]>

Author: KillianGolds

pkg/controller/llmisvc/fixture/gwapi_builders.go

@@ -371,10 +371,6 @@ func Matches(matches ...gatewayapi.HTTPRouteMatch) HTTPRouteRuleOption {
 	return WithMatches(matches...)
 }
 
-func BackendRefs(refs ...gatewayapi.HTTPBackendRef) HTTPRouteRuleOption {
-	return WithBackendRefs(refs...)
-}
-
 func Timeouts(backendTimeout, requestTimeout string) HTTPRouteRuleOption {
 	return WithTimeouts(backendTimeout, requestTimeout)
 }

pkg/controller/llmisvc/router_gateway_conditions_test.go

@@ -321,7 +321,7 @@ func TestHTTPRouteConditionsEvaluation(t *testing.T) {
 					WithParentRefs(GatewayParentRef("openshift-ai-inference", "openshift-ingress")),
 					WithHTTPRule(
 						Matches(PathPrefixMatch("/llm/facebook-opt-125m-single-simulated")),
-						BackendRefs(ServiceRef("facebook-opt-125m-single-simulated-kserve-workload-svc", 8000, 1)),
+						WithBackendRefs(ServiceRef("facebook-opt-125m-single-simulated-kserve-workload-svc", 8000, 1)),
 						Timeouts("0s", "0s"),
 						Filters(gatewayapi.HTTPRouteFilter{
 							Type: gatewayapi.HTTPRouteFilterURLRewrite,

pkg/controller/llmisvc/validation/llminferenceservice_validator.go

@@ -113,23 +113,6 @@ func (l *LLMInferenceServiceValidator) validateRouterCrossFieldConstraints(llmSv
 	httpRouteRefs := httpRoutePath.Child("refs")
 	httpRouteSpec := httpRoutePath.Child("spec")
 
-	zero := v1alpha1.GatewayRoutesSpec{}
-	if ptr.Deref(router.Route, zero) == zero && router.Gateway != nil && router.Gateway.Refs != nil {
-		return field.ErrorList{
-			field.Invalid(
-				gwRefsPath,
-				router.Gateway.Refs,
-				fmt.Sprintf("unsupported configuration: custom gateway ('%s') cannot be used with managed route ('%s: {}'); "+
-					"either provide your own HTTP routes ('%s') or remove '%s' to use the managed gateway",
-					gwRefsPath,
-					routePath,
-					httpRouteRefs,
-					gwRefsPath,
-				),
-			),
-		}
-	}
-
 	httpRoute := router.Route.HTTP
 	if httpRoute == nil {
 		return field.ErrorList{}
@@ -162,13 +145,14 @@ func (l *LLMInferenceServiceValidator) validateRouterCrossFieldConstraints(llmSv
 		))
 	}
 
-	// Managed route spec cannot be used with user-defined gateway refs
-	if httpRoute.Spec != nil && router.Gateway != nil && len(router.Gateway.Refs) > 0 {
+	// Managed route spec referring to the gateway cannot be used with user-defined gateway refs
+	if httpRoute.Spec != nil && len(httpRoute.Spec.ParentRefs) > 0 &&
+		router.Gateway != nil && len(router.Gateway.Refs) > 0 {
 		allErrs = append(allErrs, field.Invalid(
 			httpRoutePath.Child("spec"),
 			httpRoute.Spec,
 			fmt.Sprintf("unsupported configuration: managed HTTP route spec ('%s') cannot be used with custom gateway refs ('%s'); "+
-				"either remove '%s' or '%s'",
+				"either remove '%s' or '%s.parentRefs'",
 				httpRouteSpec, gwRefsPath, gwRefsPath, httpRouteSpec,
 			),
 		))

pkg/controller/llmisvc/validation/llminferenceservice_validator_int_test.go

@@ -83,7 +83,7 @@ var _ = Describe("LLMInferenceService webhook validation", func() {
 				fixture.WithHTTPRouteSpec(&fixture.HTTPRoute("test-route",
 					fixture.WithHTTPRule(
 						fixture.Matches(fixture.PathPrefixMatch("/test")),
-						fixture.BackendRefs(fixture.ServiceRef("test-service", 80, 1)),
+						fixture.WithBackendRefs(fixture.ServiceRef("test-service", 80, 1)),
 					),
 				).Spec),
 			)
@@ -113,33 +113,17 @@ var _ = Describe("LLMInferenceService webhook validation", func() {
 			Expect(errValidation.Error()).To(ContainSubstring("cannot be used with a managed gateway"))
 		})
 
-		It("should reject LLMInferenceService with managed route and user-defined gateway refs", func(ctx SpecContext) {
-			// given
-			llmSvc := fixture.LLMInferenceService("test-spec-with-gateway-refs",
-				fixture.InNamespace[*v1alpha1.LLMInferenceService](nsName),
-				fixture.WithModelURI("hf://facebook/opt-125m"),
-				fixture.WithGatewayRefs(fixture.LLMGatewayRef("test-gateway", nsName)),
-				fixture.WithManagedRoute(),
-			)
-
-			// when
-			errValidation := envTest.Client.Create(ctx, llmSvc)
-
-			// then
-			Expect(errValidation).To(HaveOccurred())
-			Expect(errValidation.Error()).To(ContainSubstring("cannot be used with managed route"))
-		})
-
-		It("should reject LLMInferenceService with managed route spec and user-defined gateway refs", func(ctx SpecContext) {
+		It("should reject LLMInferenceService with managed route spec with gateway ref and user-defined gateway refs", func(ctx SpecContext) {
 			// given
 			llmSvc := fixture.LLMInferenceService("test-spec-with-gateway-refs",
 				fixture.InNamespace[*v1alpha1.LLMInferenceService](nsName),
 				fixture.WithModelURI("hf://facebook/opt-125m"),
 				fixture.WithGatewayRefs(fixture.LLMGatewayRef("test-gateway", nsName)),
 				fixture.WithHTTPRouteSpec(&fixture.HTTPRoute("test-route",
+					fixture.WithHTTPRouteGatewayRef(fixture.GatewayParentRef("test-gateway", nsName)),
 					fixture.WithHTTPRule(
 						fixture.Matches(fixture.PathPrefixMatch("/test")),
-						fixture.BackendRefs(fixture.ServiceRef("custom-backend", 8080, 1)),
+						fixture.WithBackendRefs(fixture.ServiceRef("custom-backend", 8080, 1)),
 						fixture.Timeouts("30s", "60s"),
 					),
 				).Spec),

test/e2e/llmisvc/README.md

@@ -22,22 +22,32 @@ Tests are marked with both general and cluster-specific capability markers:
 - `@pytest.mark.cluster_nvidia` - NVIDIA GPU tests
 - `@pytest.mark.cluster_nvidia_roce` - NVIDIA ROCe tests
 - `@pytest.mark.cluster_intel` - Intel GPU tests
+- `@pytest.mark.llmd_simulator` - Tests using llm-d simulator instead of real model
 
-Examples:
-```bash
-# Run all LLM inference service tests
+## Examples
+
+### Run all LLM inference service tests 
+
+```shell
 pytest -m "llminferenceservice" test/e2e/llmisvc/
+```
 
-# Run only CPU tests
+### Run only CPU tests
+```shell
 pytest -m "llminferenceservice and cluster_cpu" test/e2e/llmisvc/
+```
 
-# Run only NVIDIA GPU tests
+### Run only NVIDIA GPU tests
+```shell
 pytest -m "llminferenceservice and cluster_nvidia" test/e2e/llmisvc/
+```
 
-# Run all GPU tests (any vendor)
+### Run all GPU tests (any vendor)
+```shell
 pytest -m "llminferenceservice and (cluster_amd or cluster_nvidia or cluster_intel)" test/e2e/llmisvc/
-
-# Run CPU and AMD GPU tests only
+```
+### Run CPU and AMD GPU tests only
+```shell
 pytest -m "llminferenceservice and (cluster_cpu or cluster_amd)" test/e2e/llmisvc/
 ```
 

test/e2e/llmisvc/fixtures.py

@@ -459,7 +459,50 @@
         "router": {
             "scheduler": {},
         },
-    }
+    },
+    "router-with-gateway-ref": {
+        "router": {
+            "gateway": {
+                "refs": [
+                    {"name": "router-gateway-1", "namespace": KSERVE_TEST_NAMESPACE},
+                ],
+            },
+        },
+    },
+    "router-with-managed-route": {
+        "router": {
+            "route": {}
+        },
+    },
+    "workload-llmd-simulator": {
+        "replicas": 1,
+        "model": {"uri": "hf://facebook/opt-125m", "name": "facebook/opt-125m"},
+        "template": {
+            "containers": [
+                {
+                    "name": "main",
+                    "image": "ghcr.io/llm-d/llm-d-inference-sim:v0.5.1",
+                    "command": ["/app/llm-d-inference-sim"],
+                    "args": [
+                        "--port",
+                        "8000",
+                        "--model",
+                        "{{ .Spec.Model.Name }}",
+                        "--mode",
+                        "random",
+                        "--ssl-certfile",
+                        "/etc/ssl/certs/tls.crt",
+                        "--ssl-keyfile",
+                        "/etc/ssl/certs/tls.key"
+                    ],
+                    "resources": {
+                        "limits": {"cpu": "1", "memory": "2Gi"},
+                        "requests": {"cpu": "1", "memory": "2Gi"},
+                    },
+                }
+            ]
+        },
+    },
 }
 
 
@@ -574,7 +617,7 @@ def _get_model_name_from_configs(config_names):
             config = LLMINFERENCESERVICE_CONFIGS[config_name]
             if "model" in config and "name" in config["model"]:
                 return config["model"]["name"]
-    return "default-model"
+    return "default/model"
 
 
 def generate_k8s_safe_suffix(base_name: str, extra_parts: List[str] = None) -> str:
@@ -608,7 +651,7 @@ def generate_test_id(test_case) -> str:
     return "-".join(test_case.base_refs)
 
 
-def create_router_resources(gateways, routes, kserve_client=None):
+def create_router_resources(gateways, routes=None, kserve_client=None):
     if not kserve_client:
         kserve_client = KServeClient(config_file=os.environ.get("KUBECONFIG", "~/.kube/config"))
 
@@ -619,7 +662,7 @@ def create_router_resources(gateways, routes, kserve_client=None):
         for gateway in gateways:
             create_or_update_gateway(kserve_client, gateway)
             gateways_created.append(gateway)
-        for route in routes:
+        for route in routes or []:
             create_or_update_route(kserve_client, route)
             routes_created.append(route)
     except Exception as e:
@@ -628,11 +671,11 @@ def create_router_resources(gateways, routes, kserve_client=None):
         raise
 
 
-def delete_router_resources(gateways, routes, kserve_client=None):
+def delete_router_resources(gateways, routes=None, kserve_client=None):
     if not kserve_client:
         kserve_client = KServeClient(config_file=os.environ.get("KUBECONFIG", "~/.kube/config"))
 
-    for route in routes:
+    for route in routes or []:
         try:
             logger.info(f"Cleaning up HttpRoute {route.get('metadata', {}).get('name')}")
             delete_route(kserve_client, route.get("metadata", {}).get("name"), route.get("metadata", {}).get("namespace", "default"))

test/e2e/llmisvc/test_llm_inference_service.py

@@ -83,6 +83,24 @@ class TestCase:
 @pytest.mark.parametrize(
     "test_case",
     [
+        pytest.param(
+            TestCase(
+                base_refs=[
+                    "router-with-gateway-ref",
+                    "router-with-managed-route",
+                    "model-fb-opt-125m",
+                    "workload-llmd-simulator",
+                ],
+                prompt="KServe is a",
+                before_test=[lambda: create_router_resources(
+                    gateways=[ROUTER_GATEWAYS[0]],
+                )],
+                after_test=[lambda: delete_router_resources(
+                    gateways=[ROUTER_GATEWAYS[0]],
+                )],
+            ),
+            marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node, pytest.mark.llmd_simulator],
+        ),
         pytest.param(
             TestCase(
                 base_refs=[

test/e2e/pytest.ini

@@ -23,4 +23,6 @@ markers =
     cluster_amd: test targeting cluster with AMD
     cluster_intel: test targeting cluster with Intel
     cluster_nvidia: test targeting cluster with NVIDIA
-    cluster_single_node: test targeting single node cluster
+    cluster_nvidia_roce: test targeting cluster with NVIDIA ROCe
+    cluster_single_node: test targeting single node cluster
+    llmd_simulator: test using llm-d simulator