Add Pydantic models and FastAPI router structure for agent and backend

agent/app/services/runtime_probe.py

@@ -0,0 +1,59 @@
+import socket
+import time
+from datetime import datetime, timezone
+from urllib.request import urlopen
+
+from app.config import settings
+from app.models import CheckResult, HealthResponse, RuntimeResponse
+from app.services.systemd_manager import SystemdManager
+
+START_TIME = time.time()
+HEALTH_TARGETS = {
+    'backend': 'http://127.0.0.1:8000/health',
+    'worker': None,
+    'frontend': 'http://127.0.0.1:5173/',
+}
+
+
+def probe_http(url: str | None) -> CheckResult:
+    if not url:
+        return CheckResult(status='warn', detail='no health endpoint configured')
+    try:
+        started = time.time()
+        with urlopen(url, timeout=2) as resp:
+            latency = round((time.time() - started) * 1000, 2)
+            return CheckResult(status='pass' if resp.status < 400 else 'fail', latency_ms=latency, detail=f'HTTP {resp.status}')
+    except Exception as exc:
+        return CheckResult(status='fail', detail=str(exc))
+
+
+def build_health() -> HealthResponse:
+    return HealthResponse(
+        status='pass',
+        service=settings.service_name,
+        version=settings.version,
+        time=datetime.now(timezone.utc),
+        uptime_s=round(time.time() - START_TIME, 2),
+        checks={'dbus': CheckResult(status='pass', detail='system bus reachable')},
+    )
+
+
+def build_runtime(systemd_manager: SystemdManager) -> RuntimeResponse:
+    services = systemd_manager.list_services(settings.monitored_services)
+    checks = {'dbus': CheckResult(status='pass', detail='system bus reachable')}
+    checks.update({k: probe_http(v) for k, v in HEALTH_TARGETS.items()})
+    overall = 'pass'
+    if any(c.status == 'fail' for c in checks.values()) or any(s.status == 'fail' for s in services):
+        overall = 'fail'
+    elif any(c.status == 'warn' for c in checks.values()) or any(s.status == 'warn' for s in services):
+        overall = 'warn'
+    return RuntimeResponse(
+        status=overall,
+        service=settings.service_name,
+        version=settings.version,
+        time=datetime.now(timezone.utc),
+        hostname=socket.gethostname(),
+        uptime_s=round(time.time() - START_TIME, 2),
+        services=services,
+        checks=checks,
+    )

agent/app/services/systemd_manager.py

@@ -0,0 +1,64 @@
+from datetime import datetime, timezone
+from typing import List
+import dbus
+
+from app.models import ServiceState
+
+
+class SystemdManager:
+    def __init__(self) -> None:
+        self.bus = dbus.SystemBus()
+        systemd = self.bus.get_object('org.freedesktop.systemd1', '/org/freedesktop/systemd1')
+        self.manager = dbus.Interface(systemd, 'org.freedesktop.systemd1.Manager')
+
+    def list_services(self, names: list[str]) -> List[ServiceState]:
+        items: list[ServiceState] = []
+        for name in names:
+            try:
+                unit_path = self.manager.GetUnit(name)
+                unit = self.bus.get_object('org.freedesktop.systemd1', unit_path)
+                props = dbus.Interface(unit, 'org.freedesktop.DBus.Properties')
+                active_state = str(props.Get('org.freedesktop.systemd1.Unit', 'ActiveState'))
+                sub_state = str(props.Get('org.freedesktop.systemd1.Unit', 'SubState'))
+                loaded = str(props.Get('org.freedesktop.systemd1.Unit', 'LoadState'))
+                description = str(props.Get('org.freedesktop.systemd1.Unit', 'Description'))
+                try:
+                    pid = int(props.Get('org.freedesktop.systemd1.Service', 'MainPID'))
+                    if pid == 0:
+                        pid = None
+                except Exception:
+                    pid = None
+                status = 'pass' if active_state == 'active' else 'warn' if active_state in {'activating', 'deactivating'} else 'fail'
+                items.append(ServiceState(
+                    name=name,
+                    description=description,
+                    active_state=active_state,
+                    sub_state=sub_state,
+                    loaded=loaded,
+                    status=status,
+                    pid=pid,
+                    since=datetime.now(timezone.utc),
+                ))
+            except Exception as exc:
+                items.append(ServiceState(
+                    name=name,
+                    active_state='failed',
+                    sub_state='not-found',
+                    loaded='not-found',
+                    status='fail',
+                    notes=str(exc),
+                ))
+        return items
+
+    def _act(self, name: str, method: str) -> None:
+        fn = getattr(self.manager, method)
+        fn(name, 'replace')
+
+    def start(self, name: str) -> None:
+        self._act(name, 'StartUnit')
+
+    def stop(self, name: str) -> None:
+        self._act(name, 'StopUnit')
+
+    def restart(self, name: str) -> None:
+        self._act(name, 'RestartUnit')