fix(monitor): write Proxmox's notification settings only while /etc/pve is mounted

- Nothing is written under /etc/pve unless it is the mounted cluster
  filesystem; a Monitor that starts while pve-cluster is being restarted
  waits for it and registers its webhook afterwards.
- The Monitor service is ordered after pve-cluster.service.
- The settings are written only when they have to change. The copies
  taken before a change are kept in
  /usr/local/share/proxmenux/backups/pve-notifications, and the ones left
  in /etc/pve by earlier versions are moved there.
- Uninstalling the Monitor removes its target and matcher from Proxmox.

Reported in #427.

AppImage/scripts/flask_notification_routes.py

@@ -895,6 +895,20 @@ def _pve_webhook_url() -> str:
 _PVE_WEBHOOK_URL = _pve_webhook_url()
 
 
+_PVE_NOT_MOUNTED = 'The Proxmox cluster filesystem (/etc/pve) is not mounted'
+
+
+def _pve_cluster_fs_mounted() -> bool:
+    """Whether /etc/pve is the Proxmox cluster filesystem.
+
+    While its service is stopped or restarting, /etc/pve is an empty
+    directory of the root disk. A file written there is not a configuration
+    file: it keeps the service from mounting the filesystem again.
+    """
+    import os
+    return os.path.ismount(os.path.dirname(_PVE_NOTIFICATIONS_CFG))
+
+
 def _pve_read_file(path):
     """Read file, return (content, error). Content is '' if missing."""
     try:
@@ -909,26 +923,47 @@ def _pve_read_file(path):
 
 
 _PVE_BACKUPS_KEPT = 3
+# Copies of Proxmox's notification settings, taken before changing them. They
+# are kept with ProxMenux, not next to the files: /etc/pve belongs to Proxmox
+# and is replicated to every node of a cluster.
+_PVE_BACKUP_DIR = '/usr/local/share/proxmenux/backups/pve-notifications'
+_PVE_BACKUP_MARK = '.proxmenux_backup_'
+
+
+def _pve_backup_prefix(path):
+    """Name the copies of a file are kept under. The private file carries
+    its own name, as both are called notifications.cfg."""
+    import os
+    name = os.path.basename(path)
+    return ('priv-' if path == _PVE_PRIV_CFG else '') + name + _PVE_BACKUP_MARK
+
+
+def _pve_backups(path):
+    """The copies kept of a file, newest first."""
+    import os
+    prefix = _pve_backup_prefix(path)
+    try:
+        names = os.listdir(_PVE_BACKUP_DIR)
+    except OSError:
+        return []
+    found = [os.path.join(_PVE_BACKUP_DIR, name) for name in names if name.startswith(prefix)]
+    return sorted(found, key=os.path.getmtime, reverse=True)
 
 
 def _pve_backup_file(path):
-    """Create timestamped backup if file exists. Never fails fatally.
+    """Keep a timestamped copy of the file if it exists. Never fails fatally.
 
-    The backups live next to the file, inside the cluster filesystem, whose
+    A copy identical to the newest one is not written again and only the
-    size is limited: a copy identical to the newest one is not written again
+    newest few are kept. Restore uses the newest.
-    and only the newest few are kept. Restore uses the newest.
     """
     import os, shutil
     from datetime import datetime
     try:
         if not os.path.exists(path):
             return
-        directory, name = os.path.split(path)
+        os.makedirs(_PVE_BACKUP_DIR, mode=0o700, exist_ok=True)
-        prefix = f"{name}.proxmenux_backup_"
+        os.chmod(_PVE_BACKUP_DIR, 0o700)
-        existing = sorted(
+        existing = _pve_backups(path)
-            (os.path.join(directory, f) for f in os.listdir(directory) if f.startswith(prefix)),
-            key=os.path.getmtime, reverse=True,
-        )
         with open(path, 'rb') as f:
             current = f.read()
         newest_matches = False
@@ -940,16 +975,48 @@ def _pve_backup_file(path):
                 pass
         if not newest_matches:
             ts = datetime.now().strftime('%Y%m%d_%H%M%S')
-            shutil.copy2(path, os.path.join(directory, prefix + ts))
+            destination = os.path.join(_PVE_BACKUP_DIR, _pve_backup_prefix(path) + ts)
-            existing = sorted(
+            shutil.copy2(path, destination)
-                (os.path.join(directory, f) for f in os.listdir(directory) if f.startswith(prefix)),
+            # The private file holds the secrets of every notification target.
-                key=os.path.getmtime, reverse=True,
+            os.chmod(destination, 0o600)
-            )
+        _pve_prune_backups(path)
-        for stale in existing[_PVE_BACKUPS_KEPT:]:
+    except Exception:
-            try:
+        pass
-                os.remove(stale)
+
-            except OSError:
+
-                pass
+def _pve_prune_backups(path):
+    """Keep only the newest copies of a file."""
+    import os
+    for stale in _pve_backups(path)[_PVE_BACKUPS_KEPT:]:
+        try:
+            os.remove(stale)
+        except OSError:
+            pass
+
+
+def _pve_collect_legacy_backups(path):
+    """Earlier versions left the copies next to the file, inside /etc/pve.
+    The newest ones are moved with the rest and none is left there."""
+    import os, shutil
+    try:
+        directory, name = os.path.split(path)
+        legacy = sorted(
+            (os.path.join(directory, f) for f in os.listdir(directory) if f.startswith(name + _PVE_BACKUP_MARK)),
+            key=os.path.getmtime, reverse=True,
+        )
+        if not legacy:
+            return
+        os.makedirs(_PVE_BACKUP_DIR, mode=0o700, exist_ok=True)
+        os.chmod(_PVE_BACKUP_DIR, 0o700)
+        for position, source in enumerate(legacy):
+            if position < _PVE_BACKUPS_KEPT:
+                stamp = os.path.basename(source)[len(name + _PVE_BACKUP_MARK):]
+                destination = os.path.join(_PVE_BACKUP_DIR, _pve_backup_prefix(path) + stamp)
+                if not os.path.exists(destination):
+                    shutil.copy2(source, destination)
+                    os.chmod(destination, 0o600)
+            os.remove(source)
+        _pve_prune_backups(path)
     except Exception:
         pass
 
@@ -1091,6 +1158,11 @@ def setup_pve_webhook_core() -> dict:
     }
     
     try:
+        if not _pve_cluster_fs_mounted():
+            result['error'] = _PVE_NOT_MOUNTED
+            result['waiting'] = True
+            return result
+
         # ── Step 1: Ensure webhook secret exists (for our own internal use) ──
         secret = notification_manager.get_webhook_secret()
         if not secret:
@@ -1109,11 +1181,10 @@ def setup_pve_webhook_core() -> dict:
         if err:
             priv_text = None
         
-        # ── Step 4: Create backups before ANY modification ──
+        # ── Step 4: Backups are taken below, only of a file about to change ──
-        _pve_backup_file(_PVE_NOTIFICATIONS_CFG)
+        _pve_collect_legacy_backups(_PVE_NOTIFICATIONS_CFG)
-        if priv_text is not None:
+        _pve_collect_legacy_backups(_PVE_PRIV_CFG)
-            _pve_backup_file(_PVE_PRIV_CFG)
+
-        
         # ── Step 5: Remove any previous proxmenux blocks from BOTH files ──
         cleaned_cfg = _pve_remove_our_blocks(cfg_text, _PVE_OUR_HEADERS)
         
@@ -1161,10 +1232,12 @@ def setup_pve_webhook_core() -> dict:
         
         new_cfg = cleaned_cfg + endpoint_block + '\n' + matcher_block
         
-        # ── Step 8: Write main config ──
+        # ── Step 8: Write main config, when it is not already as wanted ──
         try:
-            with open(_PVE_NOTIFICATIONS_CFG, 'w') as f:
+            if new_cfg != cfg_text:
-                f.write(new_cfg)
+                _pve_backup_file(_PVE_NOTIFICATIONS_CFG)
+                with open(_PVE_NOTIFICATIONS_CFG, 'w') as f:
+                    f.write(new_cfg)
         except PermissionError:
             result['error'] = f'Permission denied writing {_PVE_NOTIFICATIONS_CFG}'
             result['fallback_commands'] = _build_webhook_fallback()
@@ -1214,8 +1287,11 @@ def setup_pve_webhook_core() -> dict:
             new_priv = priv_block
         
         try:
-            with open(_PVE_PRIV_CFG, 'w') as f:
+            if new_priv != priv_text:
-                f.write(new_priv)
+                if priv_text is not None:
+                    _pve_backup_file(_PVE_PRIV_CFG)
+                with open(_PVE_PRIV_CFG, 'w') as f:
+                    f.write(new_priv)
         except PermissionError:
             result['error'] = f'Permission denied writing {_PVE_PRIV_CFG}'
             result['fallback_commands'] = _build_webhook_fallback()
@@ -1238,6 +1314,17 @@ def setup_pve_webhook_core() -> dict:
         return result
 
 
+def setup_pve_webhook_when_mounted(timeout=900, interval=5) -> dict:
+    """Configure the webhook once /etc/pve is mounted. For the Monitor that
+    starts while the cluster service is still starting or being restarted."""
+    deadline = time.monotonic() + timeout
+    while not _pve_cluster_fs_mounted():
+        if time.monotonic() >= deadline:
+            return {'configured': False, 'error': _PVE_NOT_MOUNTED, 'waiting': True}
+        time.sleep(interval)
+    return setup_pve_webhook_core()
+
+
 @notification_bp.route('/api/notifications/proxmox/setup-webhook', methods=['POST'])
 @require_auth
 def setup_proxmox_webhook():
@@ -1254,6 +1341,10 @@ def cleanup_pve_webhook_core() -> dict:
     result = {'cleaned': False, 'error': None}
     
     try:
+        if not _pve_cluster_fs_mounted():
+            result['error'] = _PVE_NOT_MOUNTED
+            return result
+
         # Read both files
         cfg_text, err = _pve_read_file(_PVE_NOTIFICATIONS_CFG)
         if err:
@@ -1336,10 +1427,6 @@ def read_pve_notification_cfg():
         'priv_cfg': '/etc/pve/priv/notifications.cfg',
     }
     
-    # Also look for any backups we created
-    backup_dir = '/etc/pve'
-    priv_backup_dir = '/etc/pve/priv'
-    
     result = {}
     for key, path in files.items():
         try:
@@ -1357,24 +1444,18 @@ def read_pve_notification_cfg():
         except Exception as e:
             result[key] = {'path': path, 'content': None, 'size': 0, 'error': str(e)}
     
-    # Find backups
+    # The copies kept of both files
     backups = []
-    for d in [backup_dir, priv_backup_dir]:
+    for fpath in _pve_backups(_PVE_NOTIFICATIONS_CFG) + _pve_backups(_PVE_PRIV_CFG):
         try:
-            for fname in sorted(os.listdir(d)):
+            with open(fpath, 'r') as f:
-                if 'proxmenux_backup' in fname:
+                backups.append({
-                    fpath = os.path.join(d, fname)
+                    'path': fpath,
-                    try:
+                    'content': f.read(),
-                        with open(fpath, 'r') as f:
+                    'size': os.path.getsize(fpath),
-                            backups.append({
+                })
-                                'path': fpath,
-                                'content': f.read(),
-                                'size': os.path.getsize(fpath),
-                            })
-                    except Exception:
-                        backups.append({'path': fpath, 'content': None, 'error': 'read_failed'})
         except Exception:
-            pass
+            backups.append({'path': fpath, 'content': None, 'error': 'read_failed'})
     
     result['backups'] = backups
     return jsonify(result), 200
@@ -1391,35 +1472,26 @@ def restore_pve_notification_cfg():
     import os
     import shutil
     
-    files_to_restore = {
-        '/etc/pve': '/etc/pve/notifications.cfg',
-        '/etc/pve/priv': '/etc/pve/priv/notifications.cfg',
-    }
-    
     restored = []
     errors = []
-    
-    for search_dir, target_path in files_to_restore.items():
-        try:
-            # Pick the most recent backup by mtime, not lexicographic name.
-            # An attacker (or accidental rename) with a write primitive
-            # could craft `notifications.cfg.proxmenux_backup_99999999_999999`
-            # and have it sort first, hijacking the restore. mtime tracks
-            # the actual file age so renamed/touched files don't fool us.
-            # Audit Tier 3.1 — restore-cfg sort lexicográfico.
-            candidates = [
-                f for f in os.listdir(search_dir)
-                if 'proxmenux_backup' in f and f.startswith('notifications.cfg')
-            ]
 
+    if not _pve_cluster_fs_mounted():
+        return jsonify({'restored': [], 'errors': [{'target': '/etc/pve', 'error': _PVE_NOT_MOUNTED}],
+                        'success': False}), 200
+
+    for target_path in (_PVE_NOTIFICATIONS_CFG, _PVE_PRIV_CFG):
+        try:
+            _pve_collect_legacy_backups(target_path)
+            # The newest by modification time: a name can be crafted to sort
+            # first, a file's age cannot.
+            candidates = _pve_backups(target_path)
             if candidates:
-                candidates.sort(
+                # Content only: the cluster filesystem sets its own permissions.
-                    key=lambda f: os.path.getmtime(os.path.join(search_dir, f)),
+                with open(candidates[0], 'rb') as source:
-                    reverse=True,
+                    content = source.read()
-                )
+                with open(target_path, 'wb') as target:
-                backup_path = os.path.join(search_dir, candidates[0])
+                    target.write(content)
-                shutil.copy2(backup_path, target_path)
+                restored.append({'target': target_path, 'from_backup': candidates[0]})
-                restored.append({'target': target_path, 'from_backup': backup_path})
             else:
                 errors.append({'target': target_path, 'error': 'no_backup_found'})
         except Exception as e:

AppImage/scripts/notification_manager.py

@@ -1141,6 +1141,11 @@ class NotificationManager:
             wh_result = setup_pve_webhook_core()
             if wh_result.get('configured'):
                 print("[NotificationManager] PVE webhook configured OK.")
+            elif wh_result.get('waiting'):
+                # /etc/pve is not mounted yet: nothing is written until it is.
+                print("[NotificationManager] PVE webhook: waiting for /etc/pve to be mounted.")
+                threading.Thread(target=self._configure_pve_webhook_when_mounted, daemon=True,
+                                 name='pve-webhook-setup').start()
             elif wh_result.get('error'):
                 print(f"[NotificationManager] PVE webhook warning: {wh_result['error']}")
         except ImportError:
@@ -2285,6 +2290,18 @@ class NotificationManager:
         'auth_fail',
     )
 
+    def _configure_pve_webhook_when_mounted(self):
+        """Configure the PVE webhook once the cluster filesystem is back."""
+        try:
+            from flask_notification_routes import setup_pve_webhook_when_mounted
+            result = setup_pve_webhook_when_mounted()
+            if result.get('configured'):
+                print("[NotificationManager] PVE webhook configured OK.")
+            elif result.get('error'):
+                print(f"[NotificationManager] PVE webhook warning: {result['error']}")
+        except Exception as e:
+            print(f"[NotificationManager] PVE webhook setup error: {e}")
+
     def _reset_cooldowns_on_start(self):
         """Clear DB rows in notification_last_sent for the curated set of
         event types listed in `_EVENT_TYPES_RESET_ON_START`.

AppImage/scripts/tests/test_pve_notifications_cfg_writes.py

@@ -0,0 +1,170 @@
+"""Proxmox's notification files are only written while /etc/pve is the
+mounted cluster filesystem, and only when their content has to change."""
+import ast
+import os
+import tempfile
+import unittest
+from pathlib import Path
+from unittest import mock
+
+SCRIPTS = Path(__file__).resolve().parents[1]
+SOURCE = SCRIPTS / 'flask_notification_routes.py'
+NAMES = {'_pve_cluster_fs_mounted', '_pve_read_file', '_pve_backup_file', '_pve_prune_backups', '_pve_backup_prefix',
+         '_pve_backups', '_pve_collect_legacy_backups', 'restore_pve_notification_cfg', '_pve_remove_our_blocks', '_pve_webhook_url',
+         '_build_webhook_fallback', 'setup_pve_webhook_core', 'cleanup_pve_webhook_core', 'setup_pve_webhook_when_mounted'}
+ASSIGNED = {'_PVE_ENDPOINT_ID', '_PVE_MATCHER_ID', '_PVE_OUR_HEADERS', '_PVE_NOT_MOUNTED', '_PVE_BACKUPS_KEPT',
+            '_PVE_HEADER_RE', 'WEBHOOK_LOOPBACK_PORT', '_PVE_BACKUP_MARK'}
+
+
+class FakeManager:
+    def __init__(self):
+        self.secret = 'c2VjcmV0LXRva2Vu'
+
+    def get_webhook_secret(self):
+        return self.secret
+
+    def _save_setting(self, _key, value):
+        self.secret = value
+
+
+def load(directory):
+    """The functions under test, pointed at a directory standing for /etc/pve."""
+    body = []
+    for node in ast.parse(SOURCE.read_text()).body:
+        if isinstance(node, ast.FunctionDef) and node.name in NAMES:
+            node.decorator_list = []
+            body.append(node)
+        elif isinstance(node, ast.Assign) and any(getattr(t, 'id', None) in ASSIGNED for t in node.targets):
+            body.append(node)
+        elif isinstance(node, ast.Import) and any(a.asname == '_re_pve_cfg' for a in node.names):
+            body.append(node)
+    import time
+    scope = {'notification_manager': FakeManager(), 'time': time, 'jsonify': lambda value: value,
+             '_PVE_NOTIFICATIONS_CFG': str(directory / 'notifications.cfg'),
+             '_PVE_PRIV_CFG': str(directory / 'priv' / 'notifications.cfg'),
+             '_PVE_BACKUP_DIR': str(directory.parent / 'proxmenux' / 'backups' / 'pve-notifications')}
+    exec(compile(ast.fix_missing_locations(ast.Module(body=body, type_ignores=[])), str(SOURCE), 'exec'), scope)
+    return scope
+
+
+class PveNotificationWriteTests(unittest.TestCase):
+    def setUp(self):
+        tmp = tempfile.TemporaryDirectory()
+        self.addCleanup(tmp.cleanup)
+        self.pve = Path(tmp.name) / 'pve'
+        self.pve.mkdir()
+        self.scope = load(self.pve)
+        self.kept = Path(self.scope['_PVE_BACKUP_DIR'])
+        self.mounted = True
+        patcher = mock.patch('os.path.ismount', side_effect=lambda path: self.mounted and path == str(self.pve))
+        patcher.start()
+        self.addCleanup(patcher.stop)
+
+    def files(self):
+        return sorted(str(p.relative_to(self.pve)) for p in self.pve.rglob('*') if p.is_file())
+
+    def test_nothing_is_written_while_the_cluster_filesystem_is_down(self):
+        # The bare directory of the root disk: empty, and it has to stay so.
+        self.mounted = False
+        result = self.scope['setup_pve_webhook_core']()
+        self.assertFalse(result['configured'])
+        self.assertTrue(result['waiting'])
+        self.assertEqual(self.scope['cleanup_pve_webhook_core']()['error'], self.scope['_PVE_NOT_MOUNTED'])
+        self.assertEqual(list(self.pve.iterdir()), [])
+
+    def test_the_webhook_is_written_once_and_not_again(self):
+        (self.pve / 'priv').mkdir()
+        existing = 'sendmail: mail-to-root\n\tcomment Send mails to root@pam\n\tmailto-user root@pam\n'
+        (self.pve / 'notifications.cfg').write_text(existing)
+        (self.pve / 'priv' / 'notifications.cfg').write_text('')
+        self.assertTrue(self.scope['setup_pve_webhook_core']()['configured'])
+        written = (self.pve / 'notifications.cfg').read_text()
+        self.assertTrue(written.startswith(existing))
+        self.assertIn('webhook: proxmenux-webhook\n', written)
+        self.assertIn('matcher: proxmenux-default\n', written)
+        first = {name: (self.pve / name).stat().st_mtime_ns for name in self.files()}
+        # The copies taken before the change are kept with ProxMenux, not in /etc/pve.
+        self.assertEqual(sorted(first), ['notifications.cfg', 'priv/notifications.cfg'])
+        copies = sorted(p.name.split('.proxmenux_backup_')[0] for p in self.kept.iterdir())
+        self.assertEqual(copies, ['notifications.cfg', 'priv-notifications.cfg'])
+        self.assertEqual({oct(p.stat().st_mode & 0o777) for p in self.kept.iterdir()}, {'0o600'})
+        self.assertEqual(oct(self.kept.stat().st_mode & 0o777), '0o700')
+        # A second start finds both files as wanted: no write, no new backup.
+        os.utime(self.pve / 'notifications.cfg', ns=(1, 1))
+        os.utime(self.pve / 'priv' / 'notifications.cfg', ns=(1, 1))
+        self.assertTrue(self.scope['setup_pve_webhook_core']()['configured'])
+        self.assertEqual(self.files(), sorted(first))
+        self.assertEqual((self.pve / 'notifications.cfg').stat().st_mtime_ns, 1)
+        self.assertEqual((self.pve / 'priv' / 'notifications.cfg').stat().st_mtime_ns, 1)
+
+    def test_copies_left_in_etc_pve_by_an_earlier_version_are_collected_without_a_write(self):
+        (self.pve / 'priv').mkdir()
+        (self.pve / 'priv' / 'notifications.cfg').write_text('')
+        (self.pve / 'notifications.cfg').write_text('')
+        self.scope['setup_pve_webhook_core']()
+        for item in self.kept.iterdir():
+            item.unlink()
+        for index in range(40):
+            for folder in (self.pve, self.pve / 'priv'):
+                stale = folder / f'notifications.cfg.proxmenux_backup_20260101_{index:06d}'
+                stale.write_text(f'old {index}')
+                os.utime(stale, ns=(index * 10 ** 9, index * 10 ** 9))
+        current = (self.pve / 'notifications.cfg').read_text()
+        self.assertTrue(self.scope['setup_pve_webhook_core']()['configured'])
+        self.assertEqual(self.files(), ['notifications.cfg', 'priv/notifications.cfg'])
+        self.assertEqual((self.pve / 'notifications.cfg').read_text(), current)
+        kept = sorted(p.name for p in self.kept.iterdir())
+        self.assertEqual(kept, [f'notifications.cfg.proxmenux_backup_20260101_{i:06d}' for i in (37, 38, 39)]
+                         + [f'priv-notifications.cfg.proxmenux_backup_20260101_{i:06d}' for i in (37, 38, 39)])
+        self.assertEqual((self.kept / kept[2]).read_text(), 'old 39')
+
+    def test_restore_takes_the_newest_copy_of_each_file(self):
+        (self.pve / 'priv').mkdir()
+        (self.pve / 'notifications.cfg').write_text('before\n')
+        (self.pve / 'priv' / 'notifications.cfg').write_text('private before\n')
+        self.scope['setup_pve_webhook_core']()
+        answer, _status = self.scope['restore_pve_notification_cfg']()
+        self.assertTrue(answer['success'])
+        self.assertEqual((self.pve / 'notifications.cfg').read_text(), 'before\n')
+        self.assertEqual((self.pve / 'priv' / 'notifications.cfg').read_text(), 'private before\n')
+        self.mounted = False
+        answer, _status = self.scope['restore_pve_notification_cfg']()
+        self.assertFalse(answer['success'])
+
+    def test_uninstalling_the_monitor_removes_its_target_from_proxmox(self):
+        menu = (SCRIPTS.parents[1] / 'scripts/menus/config_menu.sh').read_text()
+        body = menu[menu.index('remove_monitor_notification_target() {'):menu.index('uninstall_proxmenux_monitor() {')]
+        self.assertIn('mountpoint -q /etc/pve || return 0', body)
+        # The rule first: Proxmox does not delete a target that a rule still uses.
+        self.assertLess(body.index('matchers/proxmenux-default'), body.index('endpoints/webhook/proxmenux-webhook'))
+        self.assertIn('/etc/pve/priv/notifications.cfg.proxmenux_backup_*', body)
+        uninstall = menu[menu.index('uninstall_proxmenux_monitor() {'):menu.index('check_monitor_status() {')]
+        self.assertIn('remove_monitor_notification_target', uninstall)
+
+    def test_a_start_during_a_restart_waits_for_the_mount(self):
+        self.mounted = False
+        (self.pve / 'priv').mkdir()
+
+        def back(_seconds):
+            self.mounted = True
+        with mock.patch('time.sleep', side_effect=back):
+            result = self.scope['setup_pve_webhook_when_mounted'](timeout=60, interval=1)
+        self.assertTrue(result['configured'])
+        self.assertIn('webhook: proxmenux-webhook', (self.pve / 'notifications.cfg').read_text())
+
+    def test_the_wait_ends_without_writing(self):
+        self.mounted = False
+        result = self.scope['setup_pve_webhook_when_mounted'](timeout=0, interval=0)
+        self.assertTrue(result['waiting'])
+        self.assertEqual(list(self.pve.iterdir()), [])
+
+    def test_the_monitor_starts_after_the_cluster_service(self):
+        root = SCRIPTS.parents[1]
+        for name in ('install_proxmenux.sh', 'install_proxmenux_beta.sh', 'scripts/menus/config_menu.sh'):
+            text = (root / name).read_text()
+            self.assertIn('After=network.target pve-cluster.service', text, name)
+            self.assertNotIn('After=network.target\n\n[Service]\nType=simple\nUser=root\nWorkingDirectory=$MONITOR', text, name)
+
+
+if __name__ == '__main__':
+    unittest.main()

install_proxmenux.sh

@@ -623,7 +623,7 @@ _proxmenux_rewrite_monitor_unit_for_apprun() {
         cat > "$MONITOR_SERVICE_FILE" << EOF
 [Unit]
 Description=ProxMenux Monitor - Web Dashboard
-After=network.target
+After=network.target pve-cluster.service
 
 [Service]
 Type=simple
@@ -655,7 +655,7 @@ create_monitor_service() {
         cat > "$MONITOR_SERVICE_FILE" << EOF
 [Unit]
 Description=ProxMenux Monitor - Web Dashboard
-After=network.target
+After=network.target pve-cluster.service
 
 [Service]
 Type=simple

install_proxmenux_beta.sh

@@ -496,7 +496,7 @@ update_monitor_service() {
     cat > "$MONITOR_SERVICE_FILE" << EOF
 [Unit]
 Description=ProxMenux Monitor - Web Dashboard (Beta)
-After=network.target
+After=network.target pve-cluster.service
 Before=shutdown.target reboot.target halt.target
 Conflicts=shutdown.target reboot.target halt.target
 
@@ -535,7 +535,7 @@ create_monitor_service() {
         cat > "$MONITOR_SERVICE_FILE" << EOF
 [Unit]
 Description=ProxMenux Monitor - Web Dashboard (Beta)
-After=network.target
+After=network.target pve-cluster.service
 Before=shutdown.target reboot.target halt.target
 Conflicts=shutdown.target reboot.target halt.target
 

scripts/menus/config_menu.sh

@@ -71,6 +71,17 @@ BACKTITLE="$(translate "$BACKTITLE")"
 
 # ==========================================================
 
+# Proxmox would keep sending every notification to a Monitor that is gone.
+# Its own interface removes the target and the rule that feeds it; the copies
+# of the settings that earlier versions left in /etc/pve are removed too.
+remove_monitor_notification_target() {
+    mountpoint -q /etc/pve || return 0
+    pvesh delete /cluster/notifications/matchers/proxmenux-default > /dev/null 2>&1 || true
+    pvesh delete /cluster/notifications/endpoints/webhook/proxmenux-webhook > /dev/null 2>&1 || true
+    rm -f /etc/pve/notifications.cfg.proxmenux_backup_* /etc/pve/priv/notifications.cfg.proxmenux_backup_*
+    echo " - Notification target removed from Proxmox"
+}
+
 uninstall_proxmenux_monitor() {
 
     # 1. Stop service if it is running
@@ -81,6 +92,8 @@ uninstall_proxmenux_monitor() {
     echo " - Service is not running (ok)"
     fi
 
+    remove_monitor_notification_target
+
     # 2. Disable service if enabled
     if systemctl is-enabled --quiet "${MONITOR_SERVICE}"; then
     echo " - Disabling service..."
@@ -213,7 +226,7 @@ normalize_stable_monitor_service() {
     cat > "$MONITOR_UNIT_FILE" << EOF
 [Unit]
 Description=ProxMenux Monitor - Web Dashboard
-After=network.target
+After=network.target pve-cluster.service
 
 [Service]
 Type=simple

web/messages/en/docs/monitor/notifications.json

@@ -342,7 +342,7 @@
     "practiceBody": "The PVE setup writes the target as <code>http://127.0.0.1:8008</code>, so PVE-emitted notifications always go through the loopback path with rate-limit-only security. The remote-caller path with the shared secret is opt-in for custom integrations — point an external service at <code>https://&lt;monitor-host&gt;:&lt;port&gt;/api/notifications/webhook</code> and supply the <code>X-Webhook-Secret</code> header to use it.",
     "actionsIntro": "The Monitor manages this target through three actions on the Settings tab:",
     "actionsItems": [
-      "<strong>Setup</strong> — runs automatically when you enable Notifications. Creates the entry in <code>/etc/pve/notifications.cfg</code> after backing up the current file.",
+      "<strong>Setup</strong> — runs automatically when you enable Notifications. Creates the entry in <code>/etc/pve/notifications.cfg</code>. The file is written only when its content has to change, and a copy of the previous one is kept in <code>/usr/local/share/proxmenux/backups/pve-notifications/</code> (the three newest). Nothing is written while <code>/etc/pve</code> is not mounted: a Monitor that starts during a restart of <code>pve-cluster</code> registers the entry once it is back.",
       "<strong>Cleanup</strong> — removes the entry. The previous backup of the file is kept.",
       "<strong>Read config</strong> — shows the current targets and matchers as PVE sees them. This is how you confirm the Monitor's entry is the one firing when PVE has multiple notification routes configured."
     ],

web/messages/en/docs/settings/uninstall-proxmenux.json

@@ -19,7 +19,7 @@
     "items": [
       "<strong>Confirmation dialog.</strong> Yes/no — “Are you sure you want to uninstall ProxMenux?”",
       "<strong>Dependency removal checklist.</strong> Pre-selected list depends on install type (see table below). All boxes default to OFF — you opt in to removing each dependency.",
-      "<strong>Removal runs with a progress bar.</strong> Removes the venv (Translation install only), the <code>menu</code> launcher, the <code>/usr/local/share/proxmenux/</code> tree, the ProxMenux Monitor service if present, the dependencies you ticked, and finally restores the system file backups (see below).",
+      "<strong>Removal runs with a progress bar.</strong> Removes the venv (Translation install only), the <code>menu</code> launcher, the <code>/usr/local/share/proxmenux/</code> tree, the ProxMenux Monitor service if present together with the notification target it registered in Proxmox VE (<code>proxmenux-webhook</code> and its matcher), the dependencies you ticked, and finally restores the system file backups (see below).",
       "<strong>Completion dialog.</strong> Lists removed dependencies and exits."
     ]
   },

web/messages/es/docs/monitor/notifications.json

@@ -342,7 +342,7 @@
     "practiceBody": "El setup de PVE escribe el destino como <code>http://127.0.0.1:8008</code>, así que las notificaciones emitidas por PVE siempre pasan por el camino loopback con seguridad solo rate-limit. El camino de llamador remoto con shared secret es opt-in para integraciones personalizadas — apunta un servicio externo a <code>https://&lt;host-monitor&gt;:&lt;port&gt;/api/notifications/webhook</code> y suministra la cabecera <code>X-Webhook-Secret</code> para usarlo.",
     "actionsIntro": "El Monitor gestiona este destino a través de tres acciones en la pestaña Settings:",
     "actionsItems": [
-      "<strong>Setup</strong> — corre automáticamente cuando activas Notifications. Crea la entrada en <code>/etc/pve/notifications.cfg</code> tras hacer backup del archivo actual.",
+      "<strong>Setup</strong> — corre automáticamente cuando activas Notifications. Crea la entrada en <code>/etc/pve/notifications.cfg</code>. El archivo solo se escribe cuando su contenido tiene que cambiar, y se guarda un backup del anterior en <code>/usr/local/share/proxmenux/backups/pve-notifications/</code> (los tres más recientes). No se escribe nada mientras <code>/etc/pve</code> no está montado: un Monitor que arranca durante un reinicio de <code>pve-cluster</code> registra la entrada cuando vuelve.",
       "<strong>Cleanup</strong> — elimina la entrada. El backup previo del archivo se conserva.",
       "<strong>Read config</strong> — muestra los destinos y matchers actuales como PVE los ve. Así es como confirmas que la entrada del Monitor es la que dispara cuando PVE tiene múltiples rutas de notificación configuradas."
     ],

web/messages/es/docs/settings/uninstall-proxmenux.json

@@ -19,7 +19,7 @@
     "items": [
       "<strong>Diálogo de confirmación.</strong> Sí/no — \"¿Seguro que quieres desinstalar ProxMenux?\"",
       "<strong>Checklist de eliminación de dependencias.</strong> La lista preseleccionada depende del tipo de instalación (mira la tabla de abajo). Todas las casillas por defecto en OFF — te apuntas a eliminar cada dependencia.",
-      "<strong>La eliminación corre con una barra de progreso.</strong> Elimina el venv (solo instalación con traducción), el lanzador <code>menu</code>, el árbol <code>/usr/local/share/proxmenux/</code>, el servicio ProxMenux Monitor si está presente, las dependencias que marcaste y, por último, restaura los backups de archivos del sistema (mira abajo).",
+      "<strong>La eliminación corre con una barra de progreso.</strong> Elimina el venv (solo instalación con traducción), el lanzador <code>menu</code>, el árbol <code>/usr/local/share/proxmenux/</code>, el servicio ProxMenux Monitor si está presente junto con el destino de notificaciones que registró en Proxmox VE (<code>proxmenux-webhook</code> y su matcher), las dependencias que marcaste y, por último, restaura los backups de archivos del sistema (mira abajo).",
       "<strong>Diálogo de finalización.</strong> Lista las dependencias eliminadas y sale."
     ]
   },