Merge remote-tracking branch 'upstream/main' into rhoai-3.5

Author: m-rafeeq

.github/CODEOWNERS

@@ -3,4 +3,4 @@
 # Code owners are automatically requested for review when someone opens a pull request that modifies code that they own.
 
 # Default owners for everything in the repo
-* @lioraron @yizhaodev @j-mok-dev @acardace
+* @lioraron @yizhaodev @j-mok-dev @vishbhat

.github/workflows/ci-dco-signoff.yml

@@ -13,7 +13,7 @@ jobs:
     permissions:
       contents: read
     steps:
-      - uses: actions/checkout@v6
+      - uses: actions/checkout@v7
         with:
           fetch-depth: 0
 

.github/workflows/ci-integration-tests.yml

@@ -19,7 +19,7 @@ permissions:
 
 jobs:
   integration-tests:
-    name: Integration Tests (${{ matrix.file_client_type }}, ${{ matrix.db_client_type }}, ${{ matrix.exchange_client_type }}${{ matrix.enable_gie == 'true' && ', gie' || '' }})
+    name: Integration Tests (${{ matrix.file_client_type }}, ${{ matrix.db_client_type }}, ${{ matrix.exchange_client_type }}${{ matrix.enable_gie == 'true' && ', gie' || '' }}${{ matrix.enable_dispatcher == 'true' && ', dispatcher' || '' }})
     runs-on: ubuntu-latest
     timeout-minutes: 30
     strategy:
@@ -45,8 +45,12 @@ jobs:
             db_client_type: postgresql
             exchange_client_type: redis
             enable_gie: "true"
+          - file_client_type: s3
+            db_client_type: postgresql
+            exchange_client_type: redis
+            enable_dispatcher: "true"
     steps:
-      - uses: actions/checkout@v6
+      - uses: actions/checkout@v7
 
       - uses: actions/setup-go@v6
         with:
@@ -65,6 +69,7 @@ jobs:
           DB_CLIENT_TYPE: ${{ matrix.db_client_type }}
           EXCHANGE_CLIENT_TYPE: ${{ matrix.exchange_client_type }}
           ENABLE_GIE: ${{ matrix.enable_gie || 'false' }}
+          ENABLE_DISPATCHER: ${{ matrix.enable_dispatcher || 'false' }}
 
       - name: Run integration tests
         run: make test-e2e

.github/workflows/ci-release.yaml

@@ -22,7 +22,7 @@ jobs:
 
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v6
+        uses: actions/checkout@v7
 
       - name: Set up QEMU
         uses: docker/setup-qemu-action@v4

.github/workflows/create-release.yml

@@ -19,7 +19,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout
-        uses: actions/checkout@v6
+        uses: actions/checkout@v7
         with:
           fetch-depth: 0
 
@@ -65,7 +65,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout
-        uses: actions/checkout@v6
+        uses: actions/checkout@v7
 
       - name: Set up Go
         uses: actions/setup-go@v6
@@ -89,7 +89,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout
-        uses: actions/checkout@v6
+        uses: actions/checkout@v7
         with:
           fetch-depth: 0
 

.github/workflows/pre-commit.yml

@@ -20,7 +20,7 @@ jobs:
       FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: true
 
     steps:
-      - uses: actions/checkout@v6
+      - uses: actions/checkout@v7
 
       - name: Set up Go
         uses: actions/setup-go@v6
@@ -42,5 +42,12 @@ jobs:
       - name: Run pre-commit hooks
         uses: pre-commit/action@v3.0.1
 
+      - name: Run integration tests
+        run: make test-integration
+
       - name: Lint Helm chart
-        run: helm lint charts/batch-gateway
+        run: |
+          helm lint charts/batch-gateway
+          for f in charts/batch-gateway/ci/values-*.yaml; do
+            helm lint charts/batch-gateway -f "$f"
+          done

.gitignore

@@ -1,5 +1,13 @@
 bin/
+.build/
 .vscode/
 .claude/
 .cursor/
 .codegraph/
+.venv/
+__pycache__/
+*.pyc
+benchmarks/results/*
+!benchmarks/results/.gitkeep
+.dispatcher-port-forward.pid
+.dispatcher-sim-port-forward.pid

CLAUDE.md

@@ -33,5 +33,6 @@ OpenAI-compatible batch API gateway for llm-d. Three binaries: apiserver, batch-
 ## Local Testing
 
 - Unit tests: `make test`
-- Integration tests: `make test-integration` (or `make test-all` for unit + integration)
+- Regression tests: `make test-regression` (API schema compatibility + past-bug guards)
+- Integration tests: `make test-integration` (or `make test-all` for unit + regression + integration)
 - E2E: `make dev-deploy` to deploy to a local Kind cluster, then `make test-e2e`

CONTRIBUTING.md

@@ -80,13 +80,14 @@ For changes that fix broken code or add small changes within a component:
 
 ## Testing Requirements
 
-We use three tiers of testing:
+We use four tiers of testing:
 
-1. **Unit tests**: Fast verification of code parts, testing different arguments
-2. **Integration tests**: Testing protocols between components and built artifacts
-3. **End-to-end (e2e) tests**: Whole system testing including benchmarking
+1. **Unit tests**: Fast verification of code parts, testing different arguments (`make test`)
+2. **Regression tests**: API schema compatibility (golden-file round-trips) and serialization guards (`make test-regression`)
+3. **Integration tests**: Feature-level validation through the real HTTP stack with in-memory backends, plus component integration with external services — tests that require external services skip gracefully when unavailable (`make test-integration`)
+4. **End-to-end (e2e) tests**: Whole system testing including benchmarking against a live deployment (`make test-e2e`)
 
-Strong e2e coverage is required for deployed systems to prevent performance regression. Appropriate test coverage is an important part of code review.
+Integration tests run in CI on every PR alongside unit tests. Tests that require external infrastructure (Docker, S3) skip gracefully when those services are unavailable. Strong e2e coverage is required for deployed systems to prevent performance regression. Appropriate test coverage is an important part of code review.
 
 ## Security
 

Makefile

@@ -1,4 +1,4 @@
-.PHONY: help build build-apiserver build-processor build-gc run-apiserver run-processor run-gc run-apiserver-dev run-processor-dev run-gc-dev build-release package-release publish-helm-chart generate-release test test-coverage test-coverage-func clean lint fmt vet tidy install-tools deps-get deps-verify bench check check-container-tool ci image-build image-build-apiserver image-build-processor image-build-gc test-integration test-all test-e2e test-helm dev-deploy dev-clean dev-rm-cluster pre-commit
+.PHONY: help build build-apiserver build-processor build-gc run-apiserver run-processor run-gc run-apiserver-dev run-processor-dev run-gc-dev build-release package-release publish-helm-chart generate-release test test-coverage test-coverage-func clean lint fmt vet tidy install-tools deps-get deps-verify bench check check-container-tool ci image-build image-build-apiserver image-build-processor image-build-gc test-regression test-integration test-all test-e2e test-helm dev-deploy dev-clean dev-rm-cluster pre-commit benchmark-local benchmark-local-teardown
 
 SHELL := /usr/bin/env bash
 
@@ -218,6 +218,12 @@ tidy:
 	@echo "Tidying go modules..."
 	$(GO) mod tidy
 
+## update-deps: Upgrade all dependencies to latest and tidy
+update-deps:
+	@echo "Upgrading all dependencies..."
+	$(GO) get -u ./...
+	$(GO) mod tidy
+
 ## clean: Remove build artifacts and coverage files
 clean:
 	@echo "Cleaning..."
@@ -310,15 +316,22 @@ deps-verify:
 	@echo "Verifying dependencies..."
 	$(GO) mod verify
 
-## test-integration: Run integration tests (each test spawns its own mock server)
+## test-regression: Run regression tests (API schema compat, past-bug guards)
+test-regression:
+	@echo "Running regression tests..."
+	@$(GO) test $(TEST_FLAGS) -v -count=1 ./test/regression/... || \
+		(echo "\n❌ Regression tests failed" && exit 1)
+	@echo "\n✅ Regression tests passed!"
+
+## test-integration: Run integration tests (in-process server with mock backends and external service integration, no cluster needed)
 test-integration:
 	@echo "Running integration tests..."
 	@$(GO) test -v -tags=integration ./... || \
 		(echo "\n❌ Integration tests failed" && exit 1)
 	@echo "\n✅ Integration tests passed!"
 
-## test-all: Run all tests (unit + integration)
-test-all: test test-integration
+## test-all: Run all tests (unit + regression + integration)
+test-all: test test-regression test-integration
 
 KIND_CLUSTER_NAME ?= batch-gateway-dev
 
@@ -340,12 +353,45 @@ dev-rm-cluster:
 	@kind delete cluster --name batch-gateway-dev || echo "Cluster not found or already deleted"
 	@echo "✅ Kind cluster deleted"
 
+BENCHMARK_CONTEXT ?= kind-$(KIND_CLUSTER_NAME)
+BENCHMARK_SCENARIO ?= 3
+BENCHMARK_RESULTS_DIR ?= benchmarks/results/local-run
+
+## benchmark-local: Run benchmark e2e on local Kind cluster with inference-sim (no GPU required)
+##                   Set BENCHMARK_KEEP_CLUSTER=1 to skip teardown for inspection.
+benchmark-local:
+	@kind get clusters 2>/dev/null | grep -q $(KIND_CLUSTER_NAME) || \
+		{ echo "ERROR: Kind cluster '$(KIND_CLUSTER_NAME)' not found. Run 'make dev-deploy' first."; exit 1; }
+	@$(CONTAINER_TOOL) exec $(KIND_CLUSTER_NAME)-control-plane crictl images 2>/dev/null | grep -q batch-gateway-apiserver || \
+		{ echo "ERROR: batch-gateway images not loaded in Kind. Run 'make dev-deploy' first."; exit 1; }
+	@echo "=== Benchmark local e2e (MODE=sim, scenario $(BENCHMARK_SCENARIO)) ==="
+	@echo "Step 1/4: Setting up infrastructure..."
+	@MODE=sim KUBE_CONTEXT=$(BENCHMARK_CONTEXT) SCENARIO=$(BENCHMARK_SCENARIO) BG_IMAGE_TAG=0.0.1 BG_PULL_POLICY=IfNotPresent bash benchmarks/setup.sh
+	@echo "Step 2/4: Generating prompts..."
+	@python3 benchmarks/generate_prompts.py --num-requests 50 --isl-distribution fixed --isl-mean 256 --model sim-model --multi-job --output-dir $(BENCHMARK_RESULTS_DIR)
+	@echo "Step 3/4: Running benchmark..."
+	@python3 benchmarks/benchmark.py --context $(BENCHMARK_CONTEXT) --scenarios $(BENCHMARK_SCENARIO) --model sim-model --batch-size 50 --num-jobs 1 --results-dir $(BENCHMARK_RESULTS_DIR)
+	@echo "Step 4/4: Done!"
+	@echo "Report: $(BENCHMARK_RESULTS_DIR)/report.html"
+	@echo "Metadata: $(BENCHMARK_RESULTS_DIR)/run-metadata.json"
+	@if [ -z "$(BENCHMARK_KEEP_CLUSTER)" ]; then \
+		echo "Tearing down benchmark namespace..."; \
+		KUBE_CONTEXT=$(BENCHMARK_CONTEXT) SCENARIO=$(BENCHMARK_SCENARIO) bash benchmarks/teardown.sh; \
+	else \
+		echo "Keeping cluster (BENCHMARK_KEEP_CLUSTER set). Teardown with: make benchmark-local-teardown"; \
+	fi
+
+## benchmark-local-teardown: Teardown local benchmark environment
+benchmark-local-teardown:
+	@KUBE_CONTEXT=$(BENCHMARK_CONTEXT) SCENARIO=$(BENCHMARK_SCENARIO) bash benchmarks/teardown.sh
+
 ## test-e2e: Run E2E tests against a live API server (requires TEST_BASE_URL or dev-deploy NodePort services)
 ##           Use TEST_RUN to filter tests, e.g.: make test-e2e TEST_RUN=TestE2E/Batches/Cancel/InProgress
 test-e2e:
 	@echo "Running E2E tests..."
 	@OUT=$$(mktemp); \
-	cd test/e2e && $(GO) test -v -count=1 $(if $(TEST_RUN),-run $(TEST_RUN)) ./... 2>&1 | tee $$OUT; \
+	echo "Processor observability endpoint: auto-resolved by the e2e test helpers"; \
+	cd test/e2e && $(GO) test -v -count=1 -timeout=20m $(if $(TEST_RUN),-run $(TEST_RUN)) ./... 2>&1 | tee $$OUT; \
 	TEST_EXIT=$${PIPESTATUS[0]}; \
 	PASS_COUNT=$$(grep -- '--- PASS:' $$OUT 2>/dev/null | wc -l | tr -d ' '); \
 	FAIL_COUNT=$$(grep -- '--- FAIL:' $$OUT 2>/dev/null | wc -l | tr -d ' '); \