fix(ci): harden smoke diagnostics

Author: mahmut-Abi

scripts/ci/ci_assets_test.go

@@ -240,6 +240,18 @@ func TestSecurityScanWorkflowUsesExistingHelper(t *testing.T) {
 	}
 }
 
+func TestCollectKindDiagnosticsCapturesFrontendTopology(t *testing.T) {
+	contents := string(readFile(t, repoPath("scripts", "ci", "collect-kind-diagnostics.sh")))
+
+	for _, want := range []string{
+		`run_capture service-topology kubectl get svc,endpoints,endpointslice -A -o yaml`,
+	} {
+		if !strings.Contains(contents, want) {
+			t.Fatalf("kind diagnostics helper missing %q", want)
+		}
+	}
+}
+
 func TestSmokeScriptForwardsToProgrammedGatewayListener(t *testing.T) {
 	contents := string(readFile(t, repoPath("test", "e2e", "smoke", "run.sh")))
 

scripts/ci/collect-kind-diagnostics.sh

@@ -17,6 +17,7 @@ run_capture pods-all kubectl get pods -A -o wide
 run_capture events-all kubectl get events -A --sort-by=.lastTimestamp
 run_capture nantian-pods kubectl describe pods -n nantian-gw
 run_capture nantian-logs kubectl logs -n nantian-gw -l app.kubernetes.io/part-of=nantian-gw --all-containers=true --tail=300 --prefix=true
+run_capture service-topology kubectl get svc,endpoints,endpointslice -A -o yaml
 run_capture gateway-resources kubectl get gateway,httproute -A -o yaml
 run_capture nantian-images kubectl get pods -n nantian-gw -o json
 

scripts/ci/smoke_script_test.go

@@ -0,0 +1,67 @@
+package ci
+
+import (
+	"os"
+	"os/exec"
+	"path/filepath"
+	"strings"
+	"testing"
+)
+
+func TestSmokeScriptMarksEarlyBootstrapFailureAsFailed(t *testing.T) {
+	stubDir := filepath.Join(t.TempDir(), "bin")
+	if err := os.MkdirAll(stubDir, 0o755); err != nil {
+		t.Fatalf("create stub dir: %v", err)
+	}
+
+	kubectlPath := filepath.Join(stubDir, "kubectl")
+	kubectlStub := `#!/usr/bin/env bash
+set -euo pipefail
+
+if [[ "${1:-}" == "create" ]]; then
+  printf 'apiVersion: v1\nkind: Namespace\nmetadata:\n  name: nantian-e2e\n'
+  exit 0
+fi
+
+if [[ "${1:-}" == "apply" ]]; then
+  cat >/dev/null
+  exit 0
+fi
+
+if [[ "${1:-}" == "wait" ]]; then
+  echo "error: timed out waiting for the condition on pods/echo-test" >&2
+  exit 1
+fi
+
+echo "unexpected kubectl invocation: $*" >&2
+exit 1
+`
+	if err := os.WriteFile(kubectlPath, []byte(kubectlStub), 0o755); err != nil {
+		t.Fatalf("write kubectl stub: %v", err)
+	}
+
+	scriptPath, err := filepath.Abs(repoPath("test", "e2e", "smoke", "run.sh"))
+	if err != nil {
+		t.Fatalf("resolve smoke script path: %v", err)
+	}
+
+	cmd := exec.Command("bash", scriptPath, "--no-cleanup", "--skip-bootstrap")
+	cmd.Env = append(
+		os.Environ(),
+		"PATH="+stubDir+string(os.PathListSeparator)+os.Getenv("PATH"),
+		"TIMEOUT=1",
+	)
+
+	output, runErr := cmd.CombinedOutput()
+	if runErr == nil {
+		t.Fatalf("expected smoke script to fail, output:\n%s", output)
+	}
+
+	text := string(output)
+	if !strings.Contains(text, "Smoke test FAILED") {
+		t.Fatalf("expected FAILED summary, output:\n%s", text)
+	}
+	if strings.Contains(text, "Smoke test PASSED") {
+		t.Fatalf("unexpected PASSED summary on failure, output:\n%s", text)
+	}
+}

test/e2e/smoke/run.sh

@@ -2,7 +2,7 @@
 # e2e smoke test — deploy, route, request, verify, cleanup.
 # Usage: ./run.sh              # full cycle (deploy → test → cleanup)
 #        ./run.sh --no-cleanup # keep cluster running for debugging
-set -euo pipefail
+set -Eeuo pipefail
 
 SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
 GATEWAY_ROOT="$(cd "$SCRIPT_DIR/../../.." && pwd)"
@@ -252,6 +252,7 @@ send_request() {
 
 # ── Main ──
 main() {
+    trap 'FAILED=true' ERR
     trap 'cleanup_cluster; if $FAILED; then red "✗ Smoke test FAILED"; else green "✓ Smoke test PASSED"; fi' EXIT
 
     if [[ "$BOOTSTRAP" == "true" ]]; then