fix: clear obsolete LXC check failure markers

AppImage/scripts/lxc_check_job.py

@@ -38,6 +38,53 @@ def acquire(vmid):
         raise
 
 
+def _invalid_marker(vmid):
+    return LOCK_DIR / f'proxmenux-lxc-check-invalid-{int(vmid)}'
+
+
+def _acquire_invalid_marker_guard(vmid):
+    path = LOCK_DIR / f'proxmenux-lxc-check-invalid-{int(vmid)}.guard'
+    handle = os.fdopen(os.open(path, os.O_RDWR | os.O_CREAT | os.O_NOFOLLOW, 0o600), 'r+')
+    try:
+        fcntl.flock(handle, fcntl.LOCK_EX)
+        return handle
+    except BaseException:
+        handle.close()
+        raise
+
+
+def _mark_check_invalid(vmid):
+    handle = _acquire_invalid_marker_guard(vmid)
+    try:
+        _invalid_marker(vmid).touch(mode=0o600)
+    finally:
+        handle.close()
+
+
+def clear_invalid_marker(vmid, *, check_started_ns=None):
+    """Clear an obsolete failed-check marker without hiding a newer run."""
+    try:
+        handle = _acquire_invalid_marker_guard(vmid)
+    except OSError:
+        return False
+    try:
+        marker = _invalid_marker(vmid)
+        try:
+            if (check_started_ns is not None
+                    and marker.stat().st_mtime_ns > check_started_ns):
+                return False
+            marker.unlink()
+            return True
+        except FileNotFoundError:
+            # A concurrent successful check may already have published the
+            # valid state; there is no stale marker left to clear.
+            return True
+        except OSError:
+            return False
+    finally:
+        handle.close()
+
+
 def guard(vmid):
     if not managed_installs._lxc_updates_detection_enabled():
         raise RuntimeError('LXC update detection is disabled')
@@ -101,9 +148,8 @@ def checked_os(vmid):
 def run(vmid, job_id, reservation):
     error = None
     check_started = False
-    invalid = LOCK_DIR / f'proxmenux-lxc-check-invalid-{int(vmid)}'
     try:
-        invalid.touch(mode=0o600)
+        _mark_check_invalid(vmid)
         guard(vmid)
         if status(vmid) != 'running':
             raise RuntimeError('Container is stopped; start it first')
@@ -124,9 +170,7 @@ def run(vmid, job_id, reservation):
                 error += '; package result could not be persisted'
     finally:
         if not error:
-            try:
+            if not clear_invalid_marker(vmid):
-                invalid.unlink()
-            except OSError:
                 error = 'Successful check could not be published for installation'
         reservation.close()
         with _lock:

AppImage/scripts/managed_installs.py

@@ -1796,6 +1796,22 @@ def _store_update_result(item: dict, result: dict) -> None:
             item["update_check"][extra_key] = result[extra_key]
 
 
+def _clear_lxc_check_invalid_marker(vmid: int, check_started_ns: int) -> bool:
+    """Publish a persisted newer LXC check to the installation gate.
+
+    The marker is created by the explicit refresh worker.  Its own lock also
+    protects this cleanup against a later refresh starting while a scheduled
+    or lifecycle check is still completing.
+    """
+    try:
+        import lxc_check_job
+        return lxc_check_job.clear_invalid_marker(
+            vmid, check_started_ns=check_started_ns)
+    except Exception as exc:
+        print(f'[ProxMenux] could not clear stale LXC check marker for {vmid}: {exc}')
+        return False
+
+
 def refresh_oci_app(app_id: str) -> Optional[dict]:
     """Read one OCI-managed application again and store what it reports.
 
@@ -1834,6 +1850,7 @@ def check_for_updates(force: bool = False) -> list[dict]:
         _nvidia_cache["fetched_at"] = 0
 
     updates_available: list[dict] = []
+    successful_lxc_checks: list[tuple[int, int]] = []
     with _lock:
         reg = _read_registry()
         items = reg.get("items", [])
@@ -1854,6 +1871,7 @@ def check_for_updates(force: bool = False) -> list[dict]:
             checker = _CHECKERS.get(it.get("type"))
             if not checker:
                 continue
+            check_started_ns = time.time_ns() if it.get('type') == 'lxc' else None
             try:
                 result = checker(it)
             except Exception as e:
@@ -1863,13 +1881,20 @@ def check_for_updates(force: bool = False) -> list[dict]:
                           "last_check": _now_iso(), "error": str(e)}
 
             _store_update_result(it, result)
+            if check_started_ns is not None and result.get('error') is None:
+                try:
+                    successful_lxc_checks.append((int(it.get('_vmid')), check_started_ns))
+                except (TypeError, ValueError):
+                    pass
 
             if it["update_check"]["available"]:
                 updates_available.append(it)
 
         reg["items"] = items
         reg["last_check_run"] = _now_iso()
-        _write_registry(reg)
+        if _write_registry(reg):
+            for vmid, check_started_ns in successful_lxc_checks:
+                _clear_lxc_check_invalid_marker(vmid, check_started_ns)
 
     return updates_available
 
@@ -1945,6 +1970,7 @@ def refresh_lxc(vmid: int, *, refresh_package_cache: bool = True) -> Optional[di
         else:
             _merge_detected_entry(target, entry, now)
 
+        check_started_ns = time.time_ns()
         try:
             result = (_check_lxc_updates(target) if refresh_package_cache else
                       _check_lxc_updates(target, refresh_package_cache=False))
@@ -1961,4 +1987,6 @@ def refresh_lxc(vmid: int, *, refresh_package_cache: bool = True) -> Optional[di
         reg["last_targeted_refresh"] = now
         if not _write_registry(reg):
             raise OSError("could not persist LXC update check")
+        if result.get('error') is None:
+            _clear_lxc_check_invalid_marker(target_vmid, check_started_ns)
         return dict(target)

tests/lxc_updates/test_check_only_job.py

@@ -1,6 +1,7 @@
 """Real job, Flask routes and shared flock; guest transport is isolated."""
 import ast
 import datetime
+import os
 import subprocess
 import sys
 import tempfile
@@ -90,6 +91,26 @@ class CheckOnlyTests(unittest.TestCase):
         self.assertEqual(result['state'], 'complete')
         self.assertFalse((self.folder / 'proxmenux-lxc-check-invalid-100').exists())
 
+    def test_marker_cleanup_failure_reports_failure_and_releases_the_operation_lock(self):
+        self.state = 'running'
+        with patch.object(job, 'clear_invalid_marker', return_value=False):
+            result, _, _ = self.run_check()
+        self.assertEqual(result['state'], 'failed')
+        self.assertIn('could not be published', result['error'])
+        reservation = job.acquire(100)
+        reservation.close()
+
+    def test_marker_created_after_a_check_started_is_not_cleared(self):
+        marker = self.folder / 'proxmenux-lxc-check-invalid-100'
+        marker.touch()
+        os.utime(marker, ns=(2_000, 2_000))
+
+        self.assertFalse(job.clear_invalid_marker(100, check_started_ns=1_000))
+        self.assertTrue(marker.exists())
+
+    def test_already_cleared_marker_is_a_successful_publish(self):
+        self.assertTrue(job.clear_invalid_marker(100))
+
     def test_index_read_timeout_and_persistence_errors_keep_running_guest_running(self):
         self.state = 'running'
         for error in [RuntimeError('index failed'), OSError('read failed'),

tests/lxc_updates/test_observed_state.py

@@ -365,6 +365,35 @@ class ObservedStateRoutesTests(unittest.TestCase):
         self.assertEqual(managed.get_registry()["items"][0]["_os_family"], "debian")
         self.assertEqual(self.row()["update_check"]["count"], 0)
 
+    def test_daily_successful_check_clears_an_older_failed_check_marker(self):
+        marker = Path(self.temp.name) / 'proxmenux-lxc-check-invalid-100'
+        marker.touch()
+
+        with patch.object(managed, '_refresh_lxc_pkg_cache_if_stale', return_value={}):
+            managed.check_for_updates()
+
+        self.assertFalse(marker.exists())
+        self.assertIsNone(managed.get_registry()['items'][0]['update_check']['error'])
+
+    def test_daily_check_keeps_a_marker_created_after_it_started(self):
+        marker = Path(self.temp.name) / 'proxmenux-lxc-check-invalid-100'
+        marker.touch()
+        os.utime(marker, ns=(2_000, 2_000))
+
+        with patch.object(managed.time, 'time_ns', return_value=1_000):
+            managed.check_for_updates()
+
+        self.assertTrue(marker.exists())
+
+    def test_daily_check_does_not_clear_marker_when_result_cannot_be_persisted(self):
+        marker = Path(self.temp.name) / 'proxmenux-lxc-check-invalid-100'
+        marker.touch()
+
+        with patch.object(managed, '_write_registry', return_value=False):
+            managed.check_for_updates()
+
+        self.assertTrue(marker.exists())
+
     def test_existing_global_refresh_route_keeps_its_collector_contract(self):
         with patch.object(managed, "_DETECTORS", [managed._detect_lxc_containers]), \
                 patch.object(managed, "_refresh_lxc_pkg_cache_if_stale", return_value={}) as refresh:
@@ -374,10 +403,13 @@ class ObservedStateRoutesTests(unittest.TestCase):
         self.assertEqual(managed.get_registry()["items"][0]["_os_family"], "debian")
 
     def test_existing_lifecycle_refresh_keeps_index_refresh_default(self):
+        marker = Path(self.temp.name) / 'proxmenux-lxc-check-invalid-100'
+        marker.touch()
         with patch.object(managed, "_refresh_lxc_pkg_cache_if_stale", return_value={}) as refresh:
             item = managed.refresh_lxc(100)
         refresh.assert_called_once_with("100", "debian")
         self.assertEqual(item["update_check"]["_count"], 0)
+        self.assertFalse(marker.exists())
 
     def test_other_lxc_has_independent_ip_cache_and_manual_rate_limit(self):
         conf200 = Path(self.temp.name) / "200.conf"