Skip to content

ServerKit is a lightweight, modern server control panel for managing web applications, databases, and services on your VPS or dedicated server. Built with Python/Flask backend and React frontend.

fix(servers): restore the Observed-mode surface lost in the recovery rebuild

Juan Denis69136ef48ff6
diff --git a/backend/app/models/notification_preferences.py b/backend/app/models/notification_preferences.py
index ddffb91b..3913923a 100644
--- a/backend/app/models/notification_preferences.py
+++ b/backend/app/models/notification_preferences.py
@@ -48,6 +48,14 @@ class NotificationPreferences(TimestampMixin, JsonColumnMixin, db.Model):
digest_cadence = db.Column(db.String(12), default='off')
digest_last_sent_at = db.Column(db.DateTime, nullable=True)
+ # Per-event channel overrides (plan 24 Phase 2):
+ # {"<event_key>": {"<channel>": bool}}. The resolution/catalog surface was
+ # lost in the 2026-07 data loss; the column stays mapped so migration
+ # 059's data survives and the schema matches the migrations
+ # (test_migration_schema_drift) until the feature is rebuilt.
+ events_json = db.Column(db.Text, nullable=True, default='{}',
+ server_default='{}')
+
# Quiet hours (don't send non-critical notifications during these hours)
quiet_hours_enabled = db.Column(db.Boolean, default=False)
quiet_hours_start = db.Column(db.String(5), default='22:00') # HH:MM format
diff --git a/backend/app/models/server.py b/backend/app/models/server.py
index d0219c89..d3bdd544 100644
--- a/backend/app/models/server.py
+++ b/backend/app/models/server.py
@@ -111,6 +111,24 @@ class Server(TimestampMixin, JsonColumnMixin, db.Model):
auto_upgrade = db.Column(db.Boolean, default=False)
upgrade_channel = db.Column(db.String(20), default='stable') # stable, beta
+ # Managed vs observed (plan 27 D1, plan 31 #10). Migrations 065/068 have
+ # always carried these columns; the model fields and MANAGEMENT_MODES were
+ # lost in the 2026-07 data loss, which left the survey management-mode
+ # routes raising AttributeError (restored via plan 82's migration↔model
+ # drift gate). Observed boxes refuse mutating agent commands;
+ # allow_agent_update_observed is the explicit break-glass that still
+ # permits agent:update on an observed box.
+ MANAGEMENT_MODES = ('managed', 'observed')
+ management_mode = db.Column(db.String(16), nullable=False,
+ default='managed', server_default='managed')
+ allow_agent_update_observed = db.Column(db.Boolean, nullable=False,
+ default=False, server_default='0')
+
+ @property
+ def is_managed(self):
+ """False only for observed boxes (plan 27 D1)."""
+ return (self.management_mode or 'managed') != 'observed'
+
# System Info (reported by agent)
os_type = db.Column(db.String(20)) # linux, windows, darwin
os_version = db.Column(db.String(100))
@@ -440,6 +458,8 @@ class Server(TimestampMixin, JsonColumnMixin, db.Model):
'status': self.status,
'last_seen': self.last_seen.isoformat() if self.last_seen else None,
'last_error': self.last_error,
+ 'management_mode': self.management_mode or 'managed',
+ 'allow_agent_update_observed': bool(self.allow_agent_update_observed),
'onboarding_state': self.onboarding_state,
'onboarding_updated_at': self.onboarding_updated_at.isoformat() if self.onboarding_updated_at else None,
'agent_version': self.agent_version,
diff --git a/backend/app/services/agent_registry.py b/backend/app/services/agent_registry.py
index 8b16f141..6d11e0e1 100644
--- a/backend/app/services/agent_registry.py
+++ b/backend/app/services/agent_registry.py
@@ -151,6 +151,12 @@ class AgentRegistry:
# Connected agents by server_id
self._agents: Dict[str, ConnectedAgent] = {}
+ # Commands refused by the Observe guard per server (plan 31 #11's
+ # "blocked N commands" counter). In-memory like all registry state —
+ # single-worker by architecture. The dispatch-level guard itself is
+ # still hollow post-loss (plan 82 §H); until it is restored the
+ # counter only grows where callers record blocks explicitly.
+ self._observed_blocks: Dict[str, int] = {}
# Socket ID to server ID mapping
self._socket_to_server: Dict[str, str] = {}
# Lock for thread safety
@@ -476,6 +482,17 @@ class AgentRegistry:
with self._lock:
return list(self._agents.keys())
+ def record_observed_block(self, server_id: str) -> None:
+ """Count one command refused because the server is observed."""
+ with self._lock:
+ self._observed_blocks[server_id] = (
+ self._observed_blocks.get(server_id, 0) + 1)
+
+ def observed_blocked_count(self, server_id: str) -> int:
+ """Commands the Observe guard refused for this server (plan 31 #11)."""
+ with self._lock:
+ return self._observed_blocks.get(server_id, 0)
+
# ==================== Heartbeat ====================
def update_heartbeat(self, server_id: str, metrics: dict = None, client_timestamp: int = None):
diff --git a/backend/tests/test_survey_api.py b/backend/tests/test_survey_api.py
index 5fcc5dfa..73eb68d4 100644
--- a/backend/tests/test_survey_api.py
+++ b/backend/tests/test_survey_api.py
@@ -7,15 +7,12 @@ gates dispatch.
Reconstructed for plan 42 Phase 1 from the fragmented ``test_survey_api`` pyc +
the surviving ``survey_service`` / ``app/api/survey.py``.
-HOLLOW-FEATURE FINDINGS (reported to the orchestrator):
- 1. The survey REST blueprint (``survey_bp`` in ``app/api/survey.py``) is defined
- but NEVER registered in ``app/__init__.py`` — every ``/api/v1/servers/...``
- survey route 404s. The recovery rebuild dropped the blueprint registration.
- The business logic (``survey_service``) survived, so the honest-degrade /
- happy / catalog / diff paths are exercised at the SERVICE layer below.
- 2. The Observed-mode surface (``Server.management_mode`` / ``is_managed`` /
- ``MANAGEMENT_MODES``) referenced by ``survey.py`` was NOT restored on the
- ``Server`` model, so the management-mode tests are skipped with a reason.
+Post-loss recovery status: the survey blueprint registration came back via
+core_blueprints, and the Observed-mode surface (``Server.management_mode`` /
+``allow_agent_update_observed`` / ``MANAGEMENT_MODES`` / ``is_managed`` /
+``agent_registry.observed_blocked_count``) was restored by plan 82 §D after
+the migration↔model drift gate exposed it — see the section at the bottom.
+The dispatch-level Observe guard remains hollow (plan 82 §H).
"""
import pytest
@@ -114,27 +111,73 @@ def test_diff_maps_reports_removed_service(app):
assert 'nginx' in removed
-# --- HOLLOW: Observed-mode surface lost in the recovery rebuild ------------- #
+# --- Observed-mode surface (restored by plan 82 §D) ------------------------- #
+# The migration↔model drift gate (test_migration_schema_drift) surfaced that
+# migrations 065/068 carried servers.management_mode /
+# allow_agent_update_observed while the model fields, MANAGEMENT_MODES,
+# is_managed and the registry counter were lost in the recovery rebuild —
+# leaving these routes raising AttributeError. The model surface is back;
+# NOTE the dispatch-level Observe guard is still hollow (plan 82 §H), so
+# observed_blocked_count only counts explicitly recorded blocks.
[email protected](reason="plan 42: hollow — Server.is_managed lost in recovery "
- "rebuild (plan 27/31 Observed mode column gone)")
-def test_is_managed_property():
- pass
+def test_is_managed_property(app):
+ from app import db
+ s = _mk_server(db)
+ assert s.management_mode == 'managed'
+ assert s.is_managed is True
+ s.management_mode = 'observed'
+ assert s.is_managed is False
[email protected](reason="plan 42: hollow — Server.management_mode / MANAGEMENT_MODES "
- "lost + survey_bp unregistered (plan 27/31)")
-def test_switch_management_mode():
- pass
-
[email protected](reason="plan 42: hollow — Server.management_mode lost; mode-validation "
- "branch unreachable + survey_bp unregistered (plan 27/31)")
-def test_switch_management_mode_rejects_bad_value():
- pass
-
-
[email protected](reason="plan 42: hollow — Server.to_dict no longer emits management "
- "mode (plan 27/31 column gone)")
-def test_server_to_dict_includes_mode():
- pass
+def test_switch_management_mode(app, client):
+ from app import db
+ from app.models.server import Server
+ from factories import make_user, headers_for
+ with app.app_context():
+ server_id = _mk_server(db).id
+ admin_headers = headers_for(make_user(db, role='admin'))
+
+ res = client.post(f'/api/v1/servers/{server_id}/management-mode',
+ json={'mode': 'observed',
+ 'allow_agent_update_observed': True},
+ headers=admin_headers)
+ assert res.status_code == 200, res.get_json()
+ body = res.get_json()
+ assert body['management_mode'] == 'observed'
+ assert body['allow_agent_update_observed'] is True
+ assert body['observed_blocked_count'] == 0
+
+ # The flip must PERSIST — before the columns were restored, the
+ # assignment landed on a transient Python attribute and evaporated.
+ with app.app_context():
+ db.session.expire_all()
+ stored = Server.query.get(server_id)
+ assert stored.management_mode == 'observed'
+ assert stored.allow_agent_update_observed is True
+
+ res = client.get(f'/api/v1/servers/{server_id}/observed-status',
+ headers=admin_headers)
+ assert res.status_code == 200
+ assert res.get_json()['management_mode'] == 'observed'
+
+
+def test_switch_management_mode_rejects_bad_value(app, client):
+ from app import db
+ from factories import make_user, headers_for
+ with app.app_context():
+ server_id = _mk_server(db).id
+ admin_headers = headers_for(make_user(db, role='admin'))
+
+ res = client.post(f'/api/v1/servers/{server_id}/management-mode',
+ json={'mode': 'yolo'}, headers=admin_headers)
+ assert res.status_code == 400
+ assert 'mode must be one of' in res.get_json()['error']
+
+
+def test_server_to_dict_includes_mode(app):
+ from app import db
+ s = _mk_server(db)
+ d = s.to_dict()
+ assert d['management_mode'] == 'managed'
+ assert d['allow_agent_update_observed'] is False