fix: preserve verified LXC state before a failed start

AppImage/scripts/flask_server.py

@@ -2076,7 +2076,14 @@ def _refresh_started_guest(vmid: int, vm_type: str) -> None:
 
             lxc_apps.invalidate_suggestion_probes(vmid)
 
+            try:
                 managed_installs.refresh_lxc(vmid)
+            except Exception as exc:
+                # Package inventory is one stage of a started-guest rebuild.
+                # A slow/failed `pct list` must not skip app, cache, and UI
+                # refreshes that can still complete for this guest.
+                print(f'[ProxMenux] lifecycle refresh lxc {vmid}: '
+                      f'package inventory: {exc}', flush=True)
 
             sidecar = lxc_apps.check_all(vmid, force=True)
             sidecar = sidecar or {'vmid': vmid, 'apps': []}
@@ -14551,10 +14558,22 @@ def _finalize_lxc_update(
             import managed_installs
             import lxc_update_observation
             observation = lxc_update_observation.read(vmid, safe_run_id)
+            started = lxc_update_observation.read_started(vmid, safe_run_id)
+            marker_write_failed = str(reason or '').endswith('code 8')
+            # A failed start, foreign-node rejection, or pre-start cancel has
+            # no trusted start marker and must not erase the last observed
+            # package state. A rejected marker is unknown and is handled as a
+            # touched run; runner exit 8 also means a start occurred but its
+            # marker could not be recorded.
+            if observation is None and started is False and not marker_write_failed:
+                verified = True
+                untouched = True
+            else:
                 verified = managed_installs.store_lxc_update_observation(vmid, observation)
+                untouched = False
             print(f'[ProxMenux] stopped LXC verification: vmid={vmid} '
                   f'run_id={safe_run_id} handoff={observation is not None} '
-                  f'verified={verified}', flush=True)
+                  f'started={started} untouched={untouched} verified={verified}', flush=True)
             if not verified:
                 verification_errors.append('post-update OS verification unavailable')
             apps_persisted = True

AppImage/scripts/lxc_update_observation.py

@@ -29,6 +29,11 @@ def _path(vmid: int, run_id: str) -> Path:
     return OBSERVATION_DIR / f"{int(vmid)}-{run_id}.json"
 
 
+def _started_path(vmid: int, run_id: str) -> Path:
+    _path(vmid, run_id)
+    return OBSERVATION_DIR / f"{int(vmid)}-{run_id}.started"
+
+
 def _private_directory(*, create=True) -> None:
     if create:
         OBSERVATION_DIR.mkdir(mode=0o700, parents=True, exist_ok=True)
@@ -38,6 +43,65 @@ def _private_directory(*, create=True) -> None:
         raise OSError("unsafe LXC observation directory")
 
 
+def mark_started(vmid: int, run_id: str) -> None:
+    """Record that this run started a stopped CT before it can change state."""
+    path = _started_path(vmid, run_id)
+    _private_directory()
+    payload = {
+        "vmid": int(vmid),
+        "run_id": run_id,
+        "started_at": datetime.datetime.now(datetime.timezone.utc).isoformat(),
+    }
+    fd, temporary = tempfile.mkstemp(prefix=".started-", dir=OBSERVATION_DIR)
+    try:
+        os.fchmod(fd, 0o600)
+        with os.fdopen(fd, "w", encoding="utf-8") as handle:
+            json.dump(payload, handle)
+        os.replace(temporary, path)
+        print(f"[ProxMenux] LXC update started: vmid={int(vmid)} run_id={run_id}", flush=True)
+    finally:
+        if os.path.exists(temporary):
+            os.unlink(temporary)
+
+
+def read_started(vmid: int, run_id: str) -> bool | None:
+    """Return True for a trusted start marker, False when absent, else None."""
+    path = _started_path(vmid, run_id)
+
+    def reject(reason: str) -> None:
+        print(f"[ProxMenux] LXC update start marker rejected: vmid={int(vmid)} "
+              f"run_id={run_id} reason={reason}", flush=True)
+        return None
+
+    try:
+        _private_directory(create=False)
+    except FileNotFoundError:
+        return False
+    except OSError as exc:
+        return reject(type(exc).__name__)
+    try:
+        fd = os.open(path, os.O_RDONLY | os.O_NOFOLLOW | os.O_NONBLOCK)
+    except FileNotFoundError:
+        return False
+    except OSError as exc:
+        return reject(type(exc).__name__)
+    try:
+        with os.fdopen(fd, "r", encoding="utf-8") as handle:
+            info = os.fstat(handle.fileno())
+            if (not stat.S_ISREG(info.st_mode) or info.st_uid != os.geteuid()
+                    or info.st_mode & 0o077 or info.st_size > MAX_BYTES):
+                return reject('unsafe-file')
+            marker = json.load(handle)
+        started = datetime.datetime.fromisoformat(marker["started_at"])
+        age = (datetime.datetime.now(datetime.timezone.utc) - started).total_seconds()
+        if (not isinstance(marker, dict) or marker.get("vmid") != int(vmid)
+                or marker.get("run_id") != run_id or not 0 <= age <= MAX_AGE):
+            return reject('identity-age-or-shape')
+        return True
+    except (OSError, ValueError, TypeError, KeyError, OverflowError) as exc:
+        return reject(type(exc).__name__)
+
+
 def capture(vmid: int, run_id: str, *, refresh_indexes: bool = False,
             include_apps: bool = False) -> dict:
     path = _path(vmid, run_id)
@@ -46,7 +110,7 @@ def capture(vmid: int, run_id: str, *, refresh_indexes: bool = False,
     now = datetime.datetime.now(datetime.timezone.utc)
     for old in OBSERVATION_DIR.iterdir():
         info = old.lstat()
-        if (re.fullmatch(r"[1-9][0-9]*-[A-Za-z0-9._:-]{1,128}\.json", old.name)
+        if (re.fullmatch(r"[1-9][0-9]*-[A-Za-z0-9._:-]{1,128}(?:\.json|\.started)", old.name)
                 and stat.S_ISREG(info.st_mode) and info.st_uid == os.geteuid()
                 and now.timestamp() - info.st_mtime > MAX_AGE):
             old.unlink()
@@ -156,7 +220,11 @@ def read(vmid: int, run_id: str) -> dict | None:
 def discard(vmid: int, run_id: str) -> None:
     try:
         _private_directory(create=False)
-        _path(vmid, run_id).unlink()
+        for path in (_path(vmid, run_id), _started_path(vmid, run_id)):
+            try:
+                path.unlink()
+            except FileNotFoundError:
+                pass
     except FileNotFoundError:
         pass
 
@@ -165,10 +233,15 @@ def main(argv=None) -> int:
     parser = argparse.ArgumentParser(description=__doc__)
     parser.add_argument("--vmid", required=True, type=int)
     parser.add_argument("--run-id", required=True)
+    parser.add_argument('--mark-started', action='store_true',
+                        help='Record that this run started a previously stopped CT')
     parser.add_argument('--include-apps', action='store_true',
                         help='Capture registered installed versions after an update')
     args = parser.parse_args(argv)
     try:
+        if args.mark_started:
+            mark_started(args.vmid, args.run_id)
+            return 0
         capture(args.vmid, args.run_id, include_apps=args.include_apps)
         return 0
     except Exception as exc:

AppImage/scripts/managed_installs.py

@@ -540,6 +540,10 @@ def _list_pve_lxcs(*, strict: bool = False) -> list[dict]:
             raise RuntimeError('LXC inventory could not be read')
         return []
     lines = r.stdout.splitlines()
+    # `pct list` returns successful empty stdout when the node has no CTs.
+    # That is a complete inventory, so strict callers may safely retire rows.
+    if not any(line.strip() for line in lines):
+        return []
     if strict and (not lines or not lines[0].strip().startswith('VMID') or 'Status' not in lines[0]):
         raise RuntimeError('LXC inventory is incomplete')
 
@@ -1892,7 +1896,7 @@ def store_lxc_update_observation(vmid: int, observation: Optional[dict]) -> bool
         reg = _read_registry()
         items = reg.setdefault("items", [])
         target = next((it for it in items if it.get("id") == f"lxc:{vmid}"
-                       and it.get("type") == "lxc" and not it.get("removed_at")), None)
+                       and it.get("type") == "lxc"), None)
         if target is None:
             if observation is None:
                 return False

scripts/lxc/apply_updates.sh

@@ -44,6 +44,7 @@
 #      nor explicitly-enabled verified helper) available in the CT
 #   6  post-update restart failed
 #   7  another ProxMenux update is already running for this CT
+#   8  started CT could not record its verification handoff identity
 #
 # The frontend surfaces exit code + duration in a follow-up POST to
 # /api/lxc-updates/<vmid>/applied so the notification event fires with
@@ -183,6 +184,15 @@ if [[ "$STATE" != "running" ]]; then
     exit 2
   fi
   STARTED_BY_PROXMENUX=1
+  # The finalizer must distinguish a failed/pre-start run from a run that
+  # actually started this previously stopped CT but later lacks a result.
+  if [[ -n "${RUN_ID:-}" ]]; then
+    if ! python3 /usr/local/share/proxmenux/monitor-app/usr/bin/lxc_update_observation.py \
+          --vmid "$VMID" --run-id "$RUN_ID" --mark-started; then
+      echo "ERROR: started CT verification identity could not be recorded." >&2
+      exit 8
+    fi
+  fi
   # give the CT a moment for services to come up
   sleep 3
 fi

tests/lxc_updates/test_stopped_registry_retention.py

@@ -128,6 +128,24 @@ class StoppedRegistryTests(unittest.TestCase):
         self.assertEqual(row['_os_family'], 'alpine')
         self.assert_no_guest_probe()
 
+    def test_targeted_observation_reactivates_removed_row_without_a_duplicate(self):
+        registry = managed._read_registry()
+        registry['items'][0]['removed_at'] = self.now
+        managed._write_registry(registry)
+        observation = {
+            'entry': {**self.row, '_os_family': 'debian'},
+            'result': {'available': True, 'last_check': self.now, '_count': 2},
+        }
+
+        self.assertTrue(managed.store_lxc_update_observation(210, observation))
+
+        registry = managed.get_registry()
+        rows = [row for row in registry['items'] if row['id'] == 'lxc:210']
+        self.assertEqual(len(rows), 1)
+        self.assertNotIn('removed_at', rows[0])
+        self.assertEqual(rows[0]['update_check']['_count'], 2)
+        self.assert_no_guest_probe()
+
     def test_retained_result_still_expires_at_existing_48_hour_boundary(self):
         managed.detect_and_register()
         current = managed.get_active_items()[0]['update_check']
@@ -145,7 +163,7 @@ class StoppedRegistryTests(unittest.TestCase):
 
 class InventoryContractTests(unittest.TestCase):
     def test_default_failure_contract_and_strict_detector_contract(self):
-        for result in [subprocess.CompletedProcess([], 1, '', ''), subprocess.CompletedProcess([], 0, '', '')]:
+        result = subprocess.CompletedProcess([], 1, '', '')
         with patch.object(managed.subprocess, 'run', return_value=result):
             self.assertEqual(managed._list_pve_lxcs(), [])
             with self.assertRaises(RuntimeError):
@@ -162,7 +180,11 @@ class InventoryContractTests(unittest.TestCase):
                 managed._list_pve_lxcs(strict=True)
 
     def test_confirmed_empty_inventory_is_not_a_failed_read(self):
-        with patch.object(managed.subprocess, 'run', return_value=subprocess.CompletedProcess([], 0, 'VMID Status Lock Name\n', '')):
+        for stdout in ['', '\n', 'VMID Status Lock Name\n']:
+            with self.subTest(stdout=repr(stdout)), patch.object(
+                managed.subprocess, 'run',
+                return_value=subprocess.CompletedProcess([], 0, stdout, ''),
+            ):
                 self.assertEqual(managed._list_pve_lxcs(strict=True), [])
 
 

tests/lxc_updates/test_stopped_update_observation.py

@@ -59,9 +59,13 @@ root = pathlib.Path(os.environ['FIXTURE_DIR'])
 args = sys.argv[1:]
 with (root/'commands').open('a') as log: log.write(json.dumps(args)+'\\n')
 state = root/'state'
-if args[0] == 'list': print('VMID Status Name\\n100 '+state.read_text()+' fixture')
+if args[0] == 'list':
+    if os.environ.get('FOREIGN') == '1': print('VMID Status Name\\n200 stopped foreign')
+    else: print('VMID Status Name\\n100 '+state.read_text()+' fixture')
 elif args[0] == 'status': print('status: '+state.read_text())
-elif args[0] == 'start': state.write_text('running')
+elif args[0] == 'start':
+    if os.environ.get('START_FAIL') == '1': sys.exit(1)
+    state.write_text('running')
 elif args[0] == 'shutdown':
     if os.environ.get('SHUTDOWN_FAIL') == '1': sys.exit(1)
     state.write_text('stopped')
@@ -102,8 +106,10 @@ m._PCT_BIN = str(root/'bin/pct')
 h.OBSERVATION_DIR = root/'observations'
 apps._APPS_DIR = str(root/'apps')
 apps._PCT_BIN = str(root/'bin/pct')
-with (root/'commands').open('a') as log: log.write('CAPTURE\\n')
+marker = '--mark-started' in os.sys.argv
-if os.environ.get('CAPTURE_FAIL') == '1': raise SystemExit(1)
+with (root/'commands').open('a') as log: log.write('MARK_STARTED\\n' if marker else 'CAPTURE\\n')
+if os.environ.get('MARKER_FAIL') == '1' and marker: raise SystemExit(1)
+if os.environ.get('CAPTURE_FAIL') == '1' and not marker: raise SystemExit(1)
 m._detect_lxc_containers = lambda **_: ([{'id':'lxc:100','type':'lxc','_vmid':'100',
     '_is_oci':os.environ.get('OCI') == '1'}] if (root/'state').read_text() == 'running' else [])
 raise SystemExit(h.main())
@@ -169,11 +175,12 @@ raise SystemExit(h.main())
         self.assertIsNone(row['error'])
         self.assertEqual(row['os_family'], 'debian')
         log = self.log.read_text()
+        self.assertLess(log.index('MARK_STARTED'), log.index('CAPTURE'))
         self.assertLess(log.index('CAPTURE'), log.index('shutdown'))
         self.assertLess(log.index('apt list --upgradable'), log.index('shutdown'))
         self.assertNotIn('apt-get update', log)
         self.assertEqual(log.count('dist-upgrade'), 1)
-        self.assertFalse(list(self.observations.glob('*.json')))
+        self.assertFalse(list(self.observations.iterdir()))
         self.assertEqual(managed.get_registry()['items'][1]['update_check']['_count'], 7)
         self.assertEqual(self.fixture.commands, [], 'GET and stopped finalization run no guest probes')
 
@@ -198,6 +205,33 @@ raise SystemExit(h.main())
     def test_missing_capture_does_not_present_old_count_or_fresh_zero(self):
         self.assert_unverified('CAPTURE_FAIL')
 
+    def test_pre_start_failure_or_foreign_target_preserves_last_verified_result(self):
+        for condition in ('START_FAIL', 'FOREIGN'):
+            with self.subTest(condition=condition):
+                registry = managed.get_registry()
+                registry['items'][0]['update_check'] = {
+                    'available': True, 'last_check': '2026-10-14T00:00:00Z',
+                    '_count': 20, 'error': None,
+                }
+                managed._write_registry(registry)
+                result = self.run_manual(**{condition: '1'})
+                self.assertNotEqual(result.returncode, 0)
+                self.assertEqual(self.state.read_text(), 'stopped')
+                stored = managed.get_registry()['items'][0]['update_check']
+                self.assertEqual(stored['_count'], 20)
+                self.assertIsNone(stored['error'])
+                log = self.log.read_text()
+                self.assertNotIn('MARK_STARTED', log)
+                self.assertNotIn('CAPTURE', log)
+
+    def test_started_run_with_unwritable_marker_is_reported_as_unverified(self):
+        result = self.run_manual(MARKER_FAIL='1', CAPTURE_FAIL='1')
+        self.assertEqual(result.returncode, 8)
+        self.assertEqual(self.state.read_text(), 'stopped')
+        stored = managed.get_registry()['items'][0]['update_check']
+        self.assertTrue(stored['error'])
+        self.assertNotIn('_count', stored)
+
     def test_web_runner_without_identity_reports_unverified_instead_of_silent_skip(self):
         env = {**self.env, 'EXECUTION_MODE': 'web'}
         env.pop('RUN_ID')
@@ -210,6 +244,7 @@ raise SystemExit(h.main())
 
     def test_rejected_handoff_is_retained_with_safe_diagnostic_and_no_false_zero(self):
         self.observations.mkdir(mode=0o700)
+        handoff.mark_started(100, 'manual-fixture')
         path = self.observations / '100-manual-fixture.json'
         path.write_text(json.dumps({'private_payload': 'must-not-be-logged'}))
         path.chmod(0o600)
@@ -219,7 +254,7 @@ raise SystemExit(h.main())
                 status='success', source='manual', target='os')
         self.assertTrue(path.exists())
         self.assertIn('reason=KeyError', output.getvalue())
-        self.assertIn('handoff=False verified=False', output.getvalue())
+        self.assertIn('handoff=False started=True untouched=False verified=False', output.getvalue())
         self.assertNotIn('must-not-be-logged', output.getvalue())
         self.assertTrue(self.fixture.row()['update_check']['error'])
         self.assertNotIn('_count', managed.get_registry()['items'][0]['update_check'])

tests/lxc_updates/test_update_contract.py

@@ -1,6 +1,9 @@
+import ast
 import sys
 import subprocess
 import tempfile
+import threading
+import time
 import unittest
 from types import SimpleNamespace
 from pathlib import Path
@@ -23,6 +26,7 @@ class UpdateStrategyValidationTests(unittest.TestCase):
         self.assertTrue(ok, config)
         self.assertEqual(config["update_strategy"], "custom_override")
 
+
     def test_legacy_helper_then_custom_is_normalised_to_override(self):
         ok, config = lxc_apps.validate_config({
             "name": "Jellyfin",
@@ -42,6 +46,48 @@ class UpdateStrategyValidationTests(unittest.TestCase):
         self.assertEqual(config["update_strategy"], "custom_override")
 
 
+class LifecycleInventoryResilienceTests(unittest.TestCase):
+    def test_failed_targeted_inventory_refresh_does_not_abort_started_lxc_rebuild(self):
+        source = ast.parse((REPO_ROOT / 'AppImage/scripts/flask_server.py').read_text())
+        function = next(node for node in source.body if isinstance(node, ast.FunctionDef)
+                        and node.name == '_refresh_started_guest')
+        warmed = []
+        published = MagicMock()
+        lxc = SimpleNamespace(
+            invalidate_suggestion_probes=MagicMock(),
+            check_all=MagicMock(return_value={'vmid': 100, 'apps': []}),
+            _pct_exec=MagicMock(return_value=(1, '', 'not installed')),
+            ensure_oci_registration=MagicMock(),
+            get_suggestions=MagicMock(return_value={'suggestions': []}),
+        )
+        managed = SimpleNamespace(
+            refresh_lxc=MagicMock(side_effect=RuntimeError('pct list timed out')),
+        )
+        scope = {
+            'time': time,
+            '_wait_for_started_guest': lambda *_: True,
+            '_vm_cache_invalidate': MagicMock(),
+            '_vm_disk_cache': {},
+            '_warm_started_guest_handler': lambda vmid, route, _: warmed.append((vmid, route)),
+            'get_vm_config': object(), 'api_vm_backups': object(),
+            'api_vm_apps_schedule': object(), 'api_lxc_mount_points': object(),
+            '_publish_guest_modal_cache_revision': published,
+            '_vm_cache_put': MagicMock(), '_vm_app_suggestions_cache': {},
+            '_guest_lifecycle_lock': threading.RLock(),
+            '_guest_refresh_inflight': {100}, '_guest_modal_cache_revision': {},
+            '_guest_modal_cache_epoch': 0,
+        }
+        with patch.dict(sys.modules, {'lxc_apps': lxc, 'managed_installs': managed}):
+            exec(compile(ast.Module(body=[function], type_ignores=[]), 'lifecycle-refresh', 'exec'), scope)
+            scope['_refresh_started_guest'](100, 'lxc')
+
+        managed.refresh_lxc.assert_called_once_with(100)
+        lxc.check_all.assert_called_once_with(100, force=True)
+        self.assertEqual(len(warmed), 4)
+        self.assertGreaterEqual(published.call_count, 1)
+        self.assertNotIn(100, scope['_guest_refresh_inflight'])
+
+
 class HelperEvidenceTests(unittest.TestCase):
     def test_legacy_literal_wrapper_slug_is_supported(self):
         wrapper = (