CI: Exploit parallel Router End to End testing in CI

.github/workflows/ci-pr-checks.yaml

@@ -1,4 +1,4 @@
-# Matrix jobs (e2e-image-common, e2e-image-router, e2e-router) are each
+# Matrix jobs (e2e-image-common, e2e-image-router) are each
 # followed by a *-status job that aggregates their result, since GitHub
 # required status checks cannot reference individual matrix legs.
 name: CI - Test
@@ -348,29 +348,6 @@ jobs:
     if: ${{ needs.check-changes.outputs.src == 'true' }}
     runs-on: ubuntu-latest
     timeout-minutes: 45
-    strategy:
-      fail-fast: false
-      matrix:
-        suite:
-          - name: pd
-            label-filter: "!Disruptive && !Extended && !SharedStorage && !Metrics"
-            needs-renderer: "false"
-          - name: pd-shared-storage-deprecated
-            label-filter: "SharedStorage && DeprecatedPD"
-            needs-renderer: "false"
-          - name: pd-shared-storage-disagg
-            label-filter: "SharedStorage && Disagg"
-            needs-renderer: "false"
-          - name: pd-metrics
-            label-filter: "Metrics"
-            needs-renderer: "false"
-          - name: extended
-            label-filter: "Extended"
-            needs-renderer: "true"
-          - name: disruption
-            label-filter: "Disruptive"
-            needs-renderer: "false"
-    name: e2e-router (${{ matrix.suite.name }})
     steps:
       - name: Checkout source
         uses: actions/checkout@v7
@@ -404,7 +381,6 @@ jobs:
           path: ${{ runner.temp }}/e2e-images
 
       - name: Download renderer image
-        if: ${{ matrix.suite.needs-renderer == 'true' }}
         uses: actions/download-artifact@v8
         with:
           name: e2e-image-renderer
@@ -417,26 +393,11 @@ jobs:
             docker load --input "$image"
           done
 
-      - name: Run e2e-router (${{ matrix.suite.name }})
+      - name: Run e2e-router
         shell: bash
         env:
           GO_MOD_CACHE_VOL: ${{ steps.e2e-setup.outputs.go-mod-cache }}
           GO_BUILD_CACHE_VOL: ${{ steps.e2e-setup.outputs.go-build-cache }}
-          E2E_LABEL_FILTER: ${{ matrix.suite.label-filter }}
-          LOAD_VLLM_RENDER_IMAGE: ${{ matrix.suite.needs-renderer }}
           PULL_VLLM_RENDER_IMAGE: "false"
           HF_TOKEN: ${{ secrets.HF_TOKEN }}
         run: make test-e2e-run
-
-  e2e-router-status:
-    needs: e2e-router
-    if: always()
-    runs-on: ubuntu-latest
-    steps:
-      - name: Check e2e-router result
-        run: |
-          result="${{ needs.e2e-router.result }}"
-          if [[ "$result" != "success" && "$result" != "skipped" ]]; then
-            echo "::error::e2e-router failed for at least one matrix suite"
-            exit 1
-          fi

Makefile

@@ -126,7 +126,7 @@ E2E_NUM_PROCS ?= 5
 # Should we pass ALL env vars here?
 E2E_ENV_VARS = EPP_IMAGE VLLM_IMAGE SIDECAR_IMAGE VLLM_RENDER_IMAGE \
                E2E_KEEP_CLUSTER_ON_FAILURE E2E_PORT E2E_METRICS_PORT K8S_CONTEXT READY_TIMEOUT \
-               E2E_NUM_PROCS E2E_LABEL_FILTER LOAD_VLLM_RENDER_IMAGE HF_TOKEN
+               E2E_NUM_PROCS LOAD_VLLM_RENDER_IMAGE HF_TOKEN
 BUILDER_E2E_ENV_FLAGS = $(foreach v,$(E2E_ENV_VARS),$(if $($(v)),-e '$(v)=$($(v))'))
 ifneq ($(filter command line environment,$(origin NAMESPACE)),)
 BUILDER_E2E_ENV_FLAGS += -e NAMESPACE=$(NAMESPACE)

test/e2e/disruption_test.go

@@ -96,7 +96,7 @@ func completionRoutedToNamespace(nsName string) error {
 	return nil
 }
 
-var _ = ginkgo.Describe("Disruption tests", ginkgo.Label(disruptiveTestLabel), func() {
+var _ = ginkgo.Describe("Disruption tests", func() {
 	ginkgo.When("A decode pod is killed mid-request", ginkgo.Ordered, testWrapper(func() {
 		ginkgo.It("should recover and route to surviving pods", func() {
 			nsName := getNamespace()

test/e2e/e2e_suite_test.go

@@ -60,14 +60,6 @@ const (
 	// defaultPort is the envoy gateway's NodePort.
 	defaultPort        = 30080
 	defaultMetricsPort = 32090
-
-	// CI shards scheduler e2e specs with label filters.
-	extendedTestLabel      = "Extended"
-	disruptiveTestLabel    = "Disruptive"
-	sharedStorageTestLabel = "SharedStorage"
-	metricsTestLabel       = "Metrics"
-	deprecatedPDTestLabel  = "DeprecatedPD"
-	disaggTestLabel        = "Disagg"
 )
 
 var (

test/e2e/e2e_test.go

@@ -161,7 +161,7 @@ var _ = ginkgo.Describe("Run end to end tests", func() {
 		})
 	}))
 
-	ginkgo.When("Running a PD configuration with nixlv2 connector(deprecated pd-profile-handler)", ginkgo.Label(metricsTestLabel, deprecatedPDTestLabel), ginkgo.Ordered, testWrapper(func() {
+	ginkgo.When("Running a PD configuration with nixlv2 connector(deprecated pd-profile-handler)", ginkgo.Ordered, testWrapper(func() {
 		ginkgo.It("should run successfully", func() {
 			infPoolObjects := createInferencePool(1)
 
@@ -233,14 +233,12 @@ var _ = ginkgo.Describe("Run end to end tests", func() {
 	for _, tc := range []struct {
 		name   string
 		config string
-		label  string
 	}{
-		{"deprecated pd-profile-handler", deprecatedPdConfig, deprecatedPDTestLabel},
-		{"disagg-profile-handler", pdConfig, disaggTestLabel},
+		{"deprecated pd-profile-handler", deprecatedPdConfig},
+		{"disagg-profile-handler", pdConfig},
 	} {
 		config := tc.config // capture for closure
-		label := tc.label
-		ginkgo.When("Running a PD configuration with shared-storage connector using "+tc.name, ginkgo.Label(sharedStorageTestLabel, label), ginkgo.Ordered, testWrapper(func() {
+		ginkgo.When("Running a PD configuration with shared-storage connector using "+tc.name, ginkgo.Ordered, testWrapper(func() {
 			ginkgo.It("should run regular (non-streaming) requests successfully", func() {
 				infPoolObjects := createInferencePool(1)
 
@@ -475,7 +473,7 @@ var _ = ginkgo.Describe("Run end to end tests", func() {
 		})
 	}))
 
-	ginkgo.When("Running a PD configuration with disagg-profile-handler and metrics validation", ginkgo.Label(metricsTestLabel, disaggTestLabel), ginkgo.Ordered, testWrapper(func() {
+	ginkgo.When("Running a PD configuration with disagg-profile-handler and metrics validation", ginkgo.Ordered, testWrapper(func() {
 
 		ginkgo.It("should run successfully", func() {
 			infPoolObjects := createInferencePool(1)
@@ -572,7 +570,7 @@ var _ = ginkgo.Describe("Run end to end tests", func() {
 		})
 	}))
 
-	ginkgo.When("Running an E/PD (Encode/Prefill-Decode) configuration", ginkgo.Label(extendedTestLabel), ginkgo.Ordered, testWrapper(func() {
+	ginkgo.When("Running an E/PD (Encode/Prefill-Decode) configuration", ginkgo.Ordered, testWrapper(func() {
 		ginkgo.It("should route multimodal requests through encode and decode pods", func() {
 			infPoolObjects := createInferencePool(1)
 
@@ -647,7 +645,7 @@ var _ = ginkgo.Describe("Run end to end tests", func() {
 		})
 	}))
 
-	ginkgo.When("Running an E/P/D (encode/prefill/decode) configuration", ginkgo.Label(extendedTestLabel), ginkgo.Ordered, testWrapper(func() {
+	ginkgo.When("Running an E/P/D (encode/prefill/decode) configuration", ginkgo.Ordered, testWrapper(func() {
 		ginkgo.It("should route multimodal requests through encode, prefill, and decode pods", func() {
 			infPoolObjects := createInferencePool(1)
 
@@ -732,7 +730,7 @@ var _ = ginkgo.Describe("Run end to end tests", func() {
 		})
 	}))
 
-	ginkgo.When("Running an EPD (no disaggregation) configuration", ginkgo.Label(extendedTestLabel), ginkgo.Ordered, testWrapper(func() {
+	ginkgo.When("Running an EPD (no disaggregation) configuration", ginkgo.Ordered, testWrapper(func() {
 		ginkgo.It("should route text and multimodal requests to the single deployment", func() {
 			infPoolObjects := createInferencePool(1)
 
@@ -800,7 +798,7 @@ var _ = ginkgo.Describe("Run end to end tests", func() {
 		})
 	}))
 
-	ginkgo.When("Running simple non-PD KV enabled configuration", ginkgo.Label(extendedTestLabel), ginkgo.Ordered, testWrapper(func() {
+	ginkgo.When("Running simple non-PD KV enabled configuration", ginkgo.Ordered, testWrapper(func() {
 		ginkgo.It("should run successfully", func() {
 			infPoolObjects := createInferencePool(1)
 
@@ -824,7 +822,7 @@ var _ = ginkgo.Describe("Run end to end tests", func() {
 		})
 	}))
 
-	ginkgo.When("Running KV configuration with external tokenizer DataProducer plugin", ginkgo.Label(extendedTestLabel), ginkgo.Ordered, testWrapper(func() {
+	ginkgo.When("Running KV configuration with external tokenizer DataProducer plugin", ginkgo.Ordered, testWrapper(func() {
 		ginkgo.It("should run successfully", func() {
 			infPoolObjects := createInferencePool(1)
 
@@ -859,7 +857,7 @@ var _ = ginkgo.Describe("Run end to end tests", func() {
 		})
 	}))
 
-	ginkgo.When("Scaling up and down the model servers", ginkgo.Label(extendedTestLabel), ginkgo.Ordered, testWrapper(func() {
+	ginkgo.When("Scaling up and down the model servers", ginkgo.Ordered, testWrapper(func() {
 		ginkgo.It("should distribute inference requests across all model servers", func() {
 			infPoolObjects := createInferencePool(1)
 
@@ -917,7 +915,7 @@ var _ = ginkgo.Describe("Run end to end tests", func() {
 		})
 	}))
 
-	ginkgo.When("Running a vLLM Data Parallel configuration", ginkgo.Label(extendedTestLabel), ginkgo.Ordered, testWrapper(func() {
+	ginkgo.When("Running a vLLM Data Parallel configuration", ginkgo.Ordered, testWrapper(func() {
 		ginkgo.It("should schedule inference on all ranks", func() {
 			infPoolObjects := createInferencePool(2)
 

test/e2e/generate_endpoint_test.go

@@ -43,7 +43,7 @@ var twoImages = []imageSpec{
 	{Hash: "e2e-image-hash-1", Offset: 4, Length: 5},
 }
 
-var _ = ginkgo.Describe("Direct gateway /inference/v1/generate encode against encode-only", ginkgo.Label(extendedTestLabel), ginkgo.Ordered, testWrapper(func() {
+var _ = ginkgo.Describe("Direct gateway /inference/v1/generate encode against encode-only", ginkgo.Ordered, testWrapper(func() {
 	// Uses single-profile-handler (generateEncodeConfig) so the EPP routes
 	// directly to encode pods without requiring a decode stage first.
 	ginkgo.It("returns ec_transfer_params for encode bodies", func() {
@@ -77,7 +77,7 @@ var _ = ginkgo.Describe("Direct gateway /inference/v1/generate encode against en
 	})
 }))
 
-var _ = ginkgo.Describe("Direct gateway /inference/v1/generate prefill against prefill-only", ginkgo.Label(extendedTestLabel), ginkgo.Ordered, testWrapper(func() {
+var _ = ginkgo.Describe("Direct gateway /inference/v1/generate prefill against prefill-only", ginkgo.Ordered, testWrapper(func() {
 	// Uses single-profile-handler (generatePrefillConfig) so the EPP routes
 	// directly to prefill pods without requiring a decode stage first.
 	ginkgo.It("returns kv_transfer_params for prefill bodies", func() {
@@ -237,7 +237,7 @@ func expectKVTransferParams(parsed map[string]any, raw []byte) {
 		"kv_transfer_params is empty: %s", string(raw))
 }
 
-var _ = ginkgo.Describe("P/D gateway /inference/v1/generate disaggregates via sidecar", ginkgo.Label(sharedStorageTestLabel, disaggTestLabel), ginkgo.Ordered, testWrapper(func() {
+var _ = ginkgo.Describe("P/D gateway /inference/v1/generate disaggregates via sidecar", ginkgo.Ordered, testWrapper(func() {
 	// Regression test for https://github.com/llm-d/llm-d-router/issues/1461:
 	// the pd-sidecar previously had no route for /inference/v1/generate, so
 	// token-in P/D requests silently fell through to decode-only. This test

test/scripts/e2e-common.sh

@@ -29,15 +29,8 @@ e2e_handle_interrupt() {
   exit 130 # SIGINT (Ctrl+C)
 }
 
-# run_ginkgo_suite runs the Ginkgo e2e suite in the given package directory,
-# applying E2E_LABEL_FILTER when set.
+# run_ginkgo_suite runs the Ginkgo e2e suite in the given package directory.
 run_ginkgo_suite() {
   local pkg="$1"
-  if [ -n "${E2E_LABEL_FILTER:-}" ]; then
-    echo "Label filter: ${E2E_LABEL_FILTER}"
-    # Tests not being run in parallel using Ginkgo
-    E2E_NUM_PROCS=1 go test -v -timeout 45m "${pkg}" -ginkgo.v -ginkgo.fail-fast "-ginkgo.label-filter=${E2E_LABEL_FILTER}"
-  else
-    ginkgo run --procs="${E2E_NUM_PROCS}" --timeout 45m -v --fail-fast "${pkg}"
-  fi
+  ginkgo run --procs="${E2E_NUM_PROCS}" --timeout 45m -v --fail-fast "${pkg}"
 }