fix(ci): keep infrastructure failures visible in smoke verdicts

.github/workflows/core-smoke.yml

@@ -88,7 +88,9 @@ jobs:
               if path.exists():
                   data = json.loads(path.read_text())
                   out.write(f"Core smoke: {'PASS' if data['passed'] else 'FAIL'}; "
-                            f"{data['elapsed_seconds']:.2f}/60 seconds; {data['requests']} requests.\n")
+                            f"{data.get('elapsed_seconds', 0):.2f}/60 seconds; {data.get('requests', 0)} requests.\n")
+                  if data.get('infrastructure_error'):
+                      out.write('Preparation or teardown failed; inspect service logs.\n')
                   out.write('Service/dependency preparation is outside the 60-second execution budget.\n')
               else:
                   out.write('Core smoke did not produce a verdict; inspect preparation failure.\n')

tools/regression_ci.py

@@ -216,7 +216,17 @@ def main():
     parser.add_argument('--lbctl', type=Path, required=True)
     parser.add_argument('--plugins', type=Path, required=True)
     args = parser.parse_args()
-    result = run(args)
+    if args.output.exists():
+        parser.error('--output must be a new directory')
+    try:
+        result = run(args)
+    except Exception as exc:
+        verdict = args.output / 'gate' / 'gate.json'
+        result = json.loads(verdict.read_text()) if verdict.exists() else {}
+        result.update(passed=False, infrastructure_error=f'{type(exc).__name__}: {exc}')
+        verdict.parent.mkdir(parents=True, exist_ok=True)
+        verdict.write_text(json.dumps(result, indent=2) + '\n')
+        raise
     print(json.dumps(result, indent=2))
     raise SystemExit(0 if result['passed'] else 1)
 

tools/regression_gate.py

@@ -10,7 +10,7 @@ import subprocess
 import sys
 import time
 
-from regression_suite import validate
+from regression_suite import Suite, validate
 from regression_traffic import read_process_snapshot
 
 
@@ -84,6 +84,10 @@ def main():
     validate(profile)
     if 'smoke' not in profile:
         parser.error('a smoke profile is required')
+    suite = Suite(profile, args.profile.resolve(), args.output / 'suite')
+    pids = dict(profile.get('observe', {}).get('pids', {}))
+    for name, path in profile.get('observe', {}).get('pid_files', {}).items():
+        pids[name] = int(Path(suite.path(path)).read_text().strip())
     result = supervise(
         [
             sys.executable,
@@ -96,7 +100,7 @@ def main():
         ],
         args.output,
         60,
-        profile.get('observe', {}).get('pids', {}),
+        pids,
     )
     print(json.dumps(result, indent=2))
     raise SystemExit(0 if result['passed'] else 1)

tools/test_regression_gate.py

@@ -1,10 +1,12 @@
 """A gate must fail closed on timeout, missing verdict, and inner failure."""
 
 import os
+import json
 from pathlib import Path
 import sys
 import tempfile
 import unittest
+from unittest.mock import patch
 
 from regression_gate import supervise
 
@@ -30,6 +32,30 @@ class GateTest(unittest.TestCase):
             self.assertTrue(result['timed_out'])
             self.assertLess(result['elapsed_seconds'], 2)
 
+    def test_teardown_failure_overrides_successful_smoke_verdict(self):
+        from regression_ci import main
+
+        with tempfile.TemporaryDirectory() as directory:
+            output = Path(directory) / 'ci'
+
+            def fail_teardown(args):
+                verdict = args.output / 'gate' / 'gate.json'
+                verdict.parent.mkdir(parents=True)
+                verdict.write_text(json.dumps({'passed': True}))
+                raise RuntimeError('sandbox teardown failed')
+
+            with (
+                patch('regression_ci.run', side_effect=fail_teardown),
+                patch.object(
+                    sys, 'argv', ['regression_ci', '--output', str(output), '--lbctl', 'unused', '--plugins', 'unused']
+                ),
+                self.assertRaises(RuntimeError),
+            ):
+                main()
+            result = json.loads((output / 'gate/gate.json').read_text())
+            self.assertFalse(result['passed'])
+            self.assertIn('sandbox teardown failed', result['infrastructure_error'])
+
     def test_zero_exit_without_business_verdict_is_failed(self):
         with tempfile.TemporaryDirectory() as directory:
             result = supervise([sys.executable, '-c', 'pass'], Path(directory) / 'gate', 5, {})