chore: removes envoy installation requirement for running extproc tests

.envoy-version

@@ -0,0 +1 @@
+1.36.2

.github/workflows/build_and_test.yaml

@@ -136,20 +136,13 @@ jobs:
   test_extproc:
     needs: changes
     if: ${{ needs.changes.outputs.code == 'true' }}
-    name: External Processor Test (Envoy v${{ matrix.version }} on ${{ matrix.os }})
+    name: External Processor Test (${{ matrix.os }})
     strategy:
       fail-fast: false
       matrix:
-        # Note: we cannot run the latest Envoy version on macOS due to https://github.com/tetratelabs/archive-envoy/issues/12.
-        # Once it's supported, the following "binary installation" steps below can be just removed and
-        # we can simply exec.Cmd with "go tool -modfile=tools/go.mod func-e run" with the envoy version configured via ENVOY_VERSION env var.
-        include:
-          - version: 1.35.0 # NOTE: when updating this, also update the comment in the CONTRIBUTING.md file.
-            os: ubuntu-latest
-          - version: 1.35.0 # NOTE: when updating this, also update the comment in the CONTRIBUTING.md file.
-            os: macos-latest
-          - version: latest
-            os: ubuntu-latest
+        os:
+          - ubuntu-latest
+          - macos-latest
     runs-on: ${{ matrix.os }}
     steps:
       - uses: actions/checkout@v4
@@ -164,19 +157,6 @@ jobs:
             ~/go/pkg/mod
             ~/go/bin
           key: extproc-tests-${{ hashFiles('**/go.mod', '**/go.sum', '**/Makefile') }}
-      - name: Install stable Envoy via func-e
-        if: matrix.version != 'latest'
-        run: |
-          go tool -modfile=tools/go.mod func-e use ${{ matrix.version }}
-          echo $HOME/.func-e/versions/${{ matrix.version }}/bin >> $GITHUB_PATH
-      - name: Install latest Envoy
-        if: matrix.version == 'latest'
-        run: |
-          export ENVOY_BIN_DIR=$HOME/envoy/bin
-          mkdir -p $ENVOY_BIN_DIR
-          docker run -v $ENVOY_BIN_DIR:/tmp/ci -w /tmp/ci \
-          --entrypoint /bin/cp envoyproxy/envoy-dev:latest /usr/local/bin/envoy .
-          echo $ENVOY_BIN_DIR >> $GITHUB_PATH
       - env:
           TEST_AWS_ACCESS_KEY_ID: ${{ secrets.AWS_BEDROCK_USER_AWS_ACCESS_KEY_ID }}
           TEST_AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_BEDROCK_USER_AWS_SECRET_ACCESS_KEY }}

CONTRIBUTING.md

@@ -43,12 +43,6 @@ For example,
 
 - The latest `kubectl` binary for running `make test-e2e`.
   - See: https://kubernetes.io/docs/tasks/tools/
-- The latest `envoy` binary for running `make test-extproc`. The current required version is v1.35 or later.
-  - On Linux, you can download the latest Envoy binary as described in https://www.envoyproxy.io/docs/envoy/latest/start/install.
-    Alternatively, you can use `func-e` on Linux as well like on macOS below.
-  - On macOS, since `brew envoy` tends to behind the latest version, it is recommended use `func-e` to run the latest Envoy. See https://func-e.io/.
-  - `alias envoy='func-e run'` is a convenient way to run the latest Envoy binary via `func-e` on both macOS and Linux.
-    For example, `func-e use 1.34` can be used to switch to a specific version of Envoy to be run with `func-e run`.
 
 Other than that, everything will be automatically managed and installed via `make` targets,
 and you should not need to worry about the dependencies (tell us if you do).

Makefile

@@ -155,6 +155,8 @@ test-crdcel: apigen ## Run the integration tests of CEL validation in CRD defini
 .PHONY: test-extproc # This requires the extproc binary to be built.
 test-extproc: build.extproc ## Run the integration tests for extproc without controller or k8s at all.
 	@$(MAKE) build.testupstream CMD_PATH_PREFIX=tests/internal/testupstreamlib
+	@echo "Ensure func-e is built and Envoy is installed"
+	@@$(GO_TOOL) func-e run --version >/dev/null 2>&1
 	@echo "Run ExtProc test"
 	@EXTPROC_BIN=$(OUTPUT_DIR)/extproc-$(shell go env GOOS)-$(shell go env GOARCH) go test ./tests/extproc/... $(GO_TEST_E2E_ARGS)
 

tests/extproc/extproc_test.go

@@ -59,9 +59,7 @@ var (
 		Region:      "gcp-region",
 		ProjectName: "gcp-project-name",
 	}}}
-	// This always failing backend is configured to have AWS Bedrock schema so that
-	// we can test that the extproc can fallback to the different schema. E.g. Primary AWS and then OpenAI.
-	alwaysFailingBackend = filterapi.Backend{Name: "always-failing-backend", Schema: awsBedrockSchema}
+	alwaysFailingBackend = filterapi.Backend{Name: "always-failing-backend", Schema: openAISchema}
 
 	// envoyConfig is the embedded Envoy configuration template.
 	//

tests/internal/testenvironment/test_environment.go

@@ -13,6 +13,7 @@ import (
 	"net"
 	"os"
 	"os/exec"
+	"path/filepath"
 	"runtime"
 	"strconv"
 	"strings"
@@ -24,6 +25,7 @@ import (
 
 	"github.com/envoyproxy/ai-gateway/cmd/extproc/mainlib"
 	internaltesting "github.com/envoyproxy/ai-gateway/internal/testing"
+	testsinternal "github.com/envoyproxy/ai-gateway/tests/internal"
 )
 
 // TestEnvironment holds all the services needed for tests.
@@ -257,14 +259,42 @@ func requireEnvoy(t testing.TB,
 	envoyYamlPath := t.TempDir() + "/envoy.yaml"
 	require.NoError(t, os.WriteFile(envoyYamlPath, []byte(processedConfig), 0o600))
 
-	cmd := exec.CommandContext(t.Context(), "envoy",
+	// Note: do not pass t.Context() to CommandContext, as it's canceled
+	// *before* t.Cleanup functions are called.
+	//
+	// > Context returns a context that is canceled just before
+	// > Cleanup-registered functions are called.
+	//
+	// That means the subprocess gets killed before we can send it an interrupt
+	// signal for graceful shutdown, which results in orphaned subprocesses.
+	ctx, cancel := context.WithCancel(context.Background())
+	cmd := testsinternal.GoToolCmdContext(ctx, "func-e", "run",
 		"-c", envoyYamlPath,
 		"--concurrency", strconv.Itoa(max(runtime.NumCPU(), 2)),
 		// This allows multiple Envoy instances to run in parallel.
 		"--base-id", strconv.Itoa(time.Now().Nanosecond()),
 		// Add debug logging for http.
 		"--component-log-level", "http:debug",
 	)
+	// func-e will use the version specified in the project root's .envoy-version file.
+	cmd.Dir = internaltesting.FindProjectRoot()
+	version, err := os.ReadFile(filepath.Join(cmd.Dir, ".envoy-version"))
+	require.NoError(t, err)
+	t.Logf("Starting Envoy version %s", strings.TrimSpace(string(version)))
+	cmd.WaitDelay = 3 * time.Second // auto-kill after 3 seconds.
+	t.Cleanup(func() {
+		defer cancel()
+		// Graceful shutdown, should kill the Envoy subprocess, too.
+		if err := cmd.Process.Signal(os.Interrupt); err != nil {
+			t.Logf("Failed to send interrupt to aigw process: %v", err)
+		}
+		// Wait for the process to exit gracefully, in worst case this is
+		// killed in 3 seconds by WaitDelay above. In that case, you may
+		// have a zombie Envoy process left behind!
+		if _, err := cmd.Process.Wait(); err != nil {
+			t.Logf("Failed to wait for aigw process to exit: %v", err)
+		}
+	})
 
 	// wait for the ready message or exit.
 	StartAndAwaitReady(t, cmd, stdout, stderr, "starting main dispatch loop")