feat: Gateway health monitoring (Hermes CLI, API, frontend)

- Hermes CLI wrapper using subprocess.run for gateway status - Gateway API: status list (with filter), events (placeholder), start/stop/restart with audit logging - Frontend GatewayPage: timeline, filter bar, status table with 10s polling, operation modals - usePolling hook for periodic data refresh Co-Authored-By: Claude <noreply@anthropic.com>
2026-07-22 17:05:41 +08:00
parent 76045538bb
commit 6d4a5c955b
6 changed files with 553 additions and 3 deletions
@@ -0,0 +1,120 @@
+"""Gateway API — 状态、事件、启停操作。"""
+import logging
+import subprocess
+
+from fastapi import APIRouter, Depends, HTTPException, Query
+from sqlalchemy.ext.asyncio import AsyncSession
+
+from app.config import settings
+from app.database import get_db
+from app.middleware.auth import require_auth, require_role
+from app.models.audit import AuditLog
+from app.models.user import User
+from app.services import hermes_cli
+
+logger = logging.getLogger(__name__)
+
+router = APIRouter(prefix="/gateway", tags=["gateway"])
+
+# ── 占位事件数据（TODO: 从 hermes 日志或系统日志解析） ──
+
+PLACEHOLDER_EVENTS = [
+    {"time": "2026-07-22T18:45", "profile": "monitor", "type": "heartbeat_timeout",
+     "detail": "连续 3 次心跳无响应"},
+    {"time": "2026-07-22T14:20", "profile": "feishu-gw", "type": "feishu_disconnect",
+     "detail": "WebSocket 断开 8 分钟"},
+    {"time": "2026-07-22T10:30", "profile": "gateway-monitor", "type": "restart",
+     "detail": "Gateway 重启"},
+]
+
+
+@router.get("/status")
+async def get_gateway_status(
+    status: str | None = Query(None, pattern="^(online|offline|warn)$"),
+    _user: User = Depends(require_auth),
+):
+    """返回全部 Profile 的 Gateway 状态列表，支持 ?status= 筛选。"""
+    gateways = hermes_cli.gateway_status_all()
+    for gw in gateways:
+        gw.setdefault("downtime_minutes", 0)
+        gw.setdefault("restarts_today", 0)
+        gw.setdefault("last_event", "")
+        gw.setdefault("profile_type", "worker")
+    if status:
+        gateways = [g for g in gateways if g.get("status") == status]
+    return gateways
+
+
+@router.get("/events")
+async def get_gateway_events(
+    _user: User = Depends(require_auth),
+):
+    """返回最近 24h 异常事件。"""
+    # TODO: 从 hermes CLI 日志或系统日志解析真实事件
+    return PLACEHOLDER_EVENTS
+
+
+async def _operate_gateway(
+    profile: str,
+    action: str,
+    user: User,
+    db: AsyncSession,
+) -> dict:
+    """执行 hermes gateway {action} --profile {profile} 并写入审计日志。"""
+    try:
+        proc = subprocess.run(
+            [settings.hermes_cli_path, "gateway", action, "--profile", profile],
+            capture_output=True,
+            text=True,
+            timeout=30,
+        )
+        success = proc.returncode == 0
+        detail = (proc.stdout or proc.stderr).strip()
+    except Exception as e:
+        success = False
+        detail = str(e)
+
+    audit = AuditLog(
+        user_id=user.id,
+        username=user.username,
+        action=f"gateway:{action}",
+        target=f"profile:{profile}",
+        detail=detail,
+    )
+    db.add(audit)
+    await db.commit()
+
+    if not success:
+        raise HTTPException(status_code=500, detail=f"Gateway {action} 失败: {detail}")
+
+    return {"success": True, "action": action, "profile": profile, "detail": detail}
+
+
+@router.post("/{profile}/start")
+async def start_gateway(
+    profile: str,
+    user: User = Depends(require_role("owner", "operator")),
+    db: AsyncSession = Depends(get_db),
+):
+    """启动 Gateway。"""
+    return await _operate_gateway(profile, "start", user, db)
+
+
+@router.post("/{profile}/stop")
+async def stop_gateway(
+    profile: str,
+    user: User = Depends(require_role("owner", "operator")),
+    db: AsyncSession = Depends(get_db),
+):
+    """停止 Gateway。"""
+    return await _operate_gateway(profile, "stop", user, db)
+
+
+@router.post("/{profile}/restart")
+async def restart_gateway(
+    profile: str,
+    user: User = Depends(require_role("owner", "operator")),
+    db: AsyncSession = Depends(get_db),
+):
+    """重启 Gateway。"""
+    return await _operate_gateway(profile, "restart", user, db)