SenMeshNet/electron/bridges/inspection/alertRecordStore.cjs

"use strict";

/**
 * 告警记录存储
 *
 * 告警记录结构：
 * {
 *   id: "alert_xxx",
 *   rule_id: "rule_cpu_critical",
 *   rule_name: "CPU 使用率过高",
 *   device_id: "host_xxx",
 *   device_name: "核心交换机-01",
 *   metric: "cpu_usage",
 *   metric_value: "85",
 *   threshold: "80",
 *   level: "critical",
 *   message: "核心交换机-01 CPU 使用率 85%，超过阈值 80%",
 *   triggered_at: timestamp,
 *   status: "active",
 *   acknowledged_by: null,
 *   acknowledged_at: null,
 *   resolved_at: null,
 *   resolved_reason: null,
 *   note: null,
 * }
 */

const { STORAGE_KEY_ALERT_RECORDS, STORAGE_KEY_ALERT_COUNTERS } = require("./storageKeys.cjs");
const { ALERT_RECORD_STATUS, ALERT_LEVELS } = require("./types.cjs");

const MAX_ALERTS = 2000;

let _alerts = null;
let _counters = null; // 防抖动计数器 { "deviceId_ruleId": count }

function _ensureLoaded(readFn) {
  if (_alerts === null) {
    const raw = readFn(STORAGE_KEY_ALERT_RECORDS);
    _alerts = raw ? _safeParse(raw, []) : [];
  }
  if (_counters === null) {
    const raw = readFn(STORAGE_KEY_ALERT_COUNTERS);
    _counters = raw ? _safeParse(raw, {}) : {};
  }
}

function _safeParse(text, def) {
  if (text === null || text === undefined) return def;
  if (typeof text === "object") return text;
  try { return JSON.parse(text); }
  catch { return def; }
}

function _saveAlerts(writeFn) {
  if (_alerts.length > MAX_ALERTS) {
    // 只保留最近的
    _alerts = _alerts.sort((a, b) => b.triggered_at - a.triggered_at).slice(0, MAX_ALERTS);
  }
  writeFn(STORAGE_KEY_ALERT_RECORDS, JSON.stringify(_alerts));
}

function _saveCounters(writeFn) {
  writeFn(STORAGE_KEY_ALERT_COUNTERS, JSON.stringify(_counters));
}

// ---------- 告警触发（带防抖动） ----------

/**
 * 处理告警：检查是否满足持续次数，满足则创建/更新告警记录
 * @param {object} device - 设备信息 { device_id, device_name }
 * @param {object} rule - 告警规则
 * @param {number|string} metricValue - 当前指标值
 * @param {boolean} isTriggered - 是否触发（指标值 > 阈值）
 * @returns {object|null} 新创建或恢复的告警记录，无变化则返回 null
 */
function processAlert(readFn, writeFn, device, rule, metricValue, isTriggered) {
  _ensureLoaded(readFn);

  const counterKey = `${device.device_id}_${rule.id}`;
  const activeAlert = _findActiveAlert(device.device_id, rule.id);

  if (isTriggered) {
    // 触发：计数 +1
    _counters[counterKey] = (_counters[counterKey] || 0) + 1;
    _saveCounters(writeFn);

    // 达到持续次数且没有活跃告警 → 创建告警
    if (_counters[counterKey] >= rule.duration && !activeAlert) {
      const alert = _createAlert(device, rule, metricValue);
      _alerts.push(alert);
      _saveAlerts(writeFn);
      return { ...alert, action: "triggered" };
    }

    // 已有活跃告警 → 更新指标值 + 刷新触发时间
    if (activeAlert) {
      // 如果告警之前被手动确认过（acknowledged）但设备继续超限，重置回活跃状态
      if (activeAlert.status === ALERT_RECORD_STATUS.ACKNOWLEDGED) {
        activeAlert.status = ALERT_RECORD_STATUS.ACTIVE;
        activeAlert.acknowledged_by = null;
        activeAlert.acknowledged_at = null;
      }
      activeAlert.metric_value = _stringifyMetric(metricValue);
      activeAlert.threshold = _stringifyMetric(rule.threshold);
      activeAlert.message = _buildMessage(device.device_name, rule, _stringifyMetric(metricValue), _stringifyMetric(rule.threshold));
      activeAlert.triggered_at = Date.now();
      _saveAlerts(writeFn);
      return { ...activeAlert, action: "updated" };
    }

    return null;
  } else {
    // 未触发：重置计数
    if (_counters[counterKey]) {
      delete _counters[counterKey];
      _saveCounters(writeFn);
    }

    // 注意：不再自动恢复告警（auto_recover 已移除）。
    // 告警一旦触发就保持 ACTIVE / ACKNOWLEDGED，必须由值班员手动 resolveAlert() 或 closeAlert() 才会关闭。
    // 原因：闪断/偶发正常会导致告警"闪了一下就没了"，值班员还没来得及处理就消失了。

    return null;
  }
}

function _findActiveAlert(deviceId, ruleId) {
  return _alerts.find(
    (a) =>
      a.device_id === deviceId &&
      a.rule_id === ruleId &&
      (a.status === ALERT_RECORD_STATUS.ACTIVE ||
        a.status === ALERT_RECORD_STATUS.ACKNOWLEDGED)
  );
}

function _stringifyMetric(v) {
  if (v === null || v === undefined) return "";
  if (typeof v === "number" || typeof v === "string" || typeof v === "boolean") return String(v);
  if (Array.isArray(v)) {
    // 数组：数字数组 join 逗号，对象数组 JSON 然后截断
    if (v.length === 0) return "";
    if (v.every(x => typeof x === "number" || typeof x === "string" || typeof x === "boolean")) {
      return v.map(x => String(x)).join(", ");
    }
    try { return JSON.stringify(v).slice(0, 200); } catch { return ""; }
  }
  if (typeof v === "object") {
    // 对象：优先取 value / current / usage 等常见数值字段
    if (typeof v.value === "number" || typeof v.value === "string") return String(v.value);
    if (typeof v.current === "number" || typeof v.current === "string") return String(v.current);
    if (typeof v.usage === "number" || typeof v.usage === "string") return String(v.usage);
    if (typeof v.percent === "number" || typeof v.percent === "string") return String(v.percent);
    try { return JSON.stringify(v).slice(0, 200); } catch { return ""; }
  }
  return String(v);
}

function _createAlert(device, rule, metricValue) {
  const mv = _stringifyMetric(metricValue);
  const tv = _stringifyMetric(rule.threshold);
  return {
    id: `alert_${Date.now()}_${Math.random().toString(36).slice(2, 8)}`,
    rule_id: rule.id,
    rule_name: rule.name,
    device_id: device.device_id,
    device_name: device.device_name || device.device_id,
    metric: rule.metric,
    metric_value: mv,
    threshold: tv,
    level: rule.level,
    message: _buildMessage(device.device_name || device.device_id, rule, mv, tv),
    triggered_at: Date.now(),
    status: ALERT_RECORD_STATUS.ACTIVE,
    acknowledged_by: null,
    acknowledged_at: null,
    resolved_at: null,
    resolved_reason: null,
    note: null,
  };
}

function _buildMessage(deviceName, rule, metricValueStr, thresholdStr) {
  const unit = rule.metric.includes("_usage") || rule.metric.includes("_percent") ? "%" : "";
  return `${deviceName} ${rule.name}，当前值 ${metricValueStr}${unit}，阈值 ${thresholdStr}${unit}`;
}

/**
 * 创建执行错误告警（timeout / failed 等采集器执行异常）
 * 每台设备 + rule_offline 全局唯一：
 *   - 已存在任何非 RESOLVED 状态 → 复用，强制激活，覆盖 message/triggered_at
 *   - 同设备多条残留（历史遗留）→ 自动合并：最新一条保留，其余降级为 RESOLVED
 *   - 全 RESOLVED → 新建一条 ACTIVE
 */
function createExecutionErrorAlert(readFn, writeFn, device, status, errorMessage) {
  _ensureLoaded(readFn);

  const offlineRuleId = "rule_offline";
  const offlineRuleName = "设备离线";

  // 1. 强制去重：同 device + rule_offline 的非 RESOLVED 告警，只保留最新的一条
  const sameDeviceOffline = _alerts.filter(
    (a) => a.device_id === device.device_id && a.rule_id === offlineRuleId
  );
  const activeList = sameDeviceOffline.filter(
    (a) => a.status !== ALERT_RECORD_STATUS.RESOLVED && a.status !== ALERT_RECORD_STATUS.CLOSED
  );
  if (activeList.length > 1) {
    // 多条残留 → triggered_at 最新的保留，其余强制降级为 RESOLVED（清理历史重复）
    activeList.sort((a, b) => b.triggered_at - a.triggered_at);
    for (let i = 1; i < activeList.length; i++) {
      activeList[i].status = ALERT_RECORD_STATUS.RESOLVED;
      activeList[i].resolved_at = Date.now();
      activeList[i].resolved_reason = "dedup_execution_error";
    }
    _saveAlerts(writeFn);
  }

  // 2. 复用唯一剩余的那条非 RESOLVED 告警（ACTIVE 或 ACKNOWLEDGED 都算）
  const existing = activeList[0];
  if (existing && existing.status !== ALERT_RECORD_STATUS.RESOLVED && existing.status !== ALERT_RECORD_STATUS.CLOSED) {
    if (existing.status === ALERT_RECORD_STATUS.ACKNOWLEDGED) {
      existing.status = ALERT_RECORD_STATUS.ACTIVE;
      existing.acknowledged_by = null;
      existing.acknowledged_at = null;
    }
    existing.metric_value = "false";
    existing.message = `${device.device_name || device.device_id} ${offlineRuleName}: ${errorMessage}`;
    existing.triggered_at = Date.now();
    _saveAlerts(writeFn);
    return { ...existing, action: "updated" };
  }

  // 3. 没有 → 新建
  const alert = {
    id: `alert_${Date.now()}_${Math.random().toString(36).slice(2, 8)}`,
    rule_id: offlineRuleId,
    rule_name: offlineRuleName,
    device_id: device.device_id,
    device_name: device.device_name || device.device_id,
    metric: "online",
    metric_value: "false",
    threshold: "true",
    level: "critical",
    message: `${device.device_name || device.device_id} ${offlineRuleName}: ${errorMessage}`,
    triggered_at: Date.now(),
    status: ALERT_RECORD_STATUS.ACTIVE,
    acknowledged_by: null,
    acknowledged_at: null,
    resolved_at: null,
    resolved_reason: null,
    note: null,
  };

  _alerts.push(alert);
  _saveAlerts(writeFn);
  return { ...alert, action: "triggered" };
}

// ---------- 告警查询 ----------

function listAlerts(readFn, options = {}) {
  _ensureLoaded(readFn);
  let result = [..._alerts];

  if (options.status) {
    result = result.filter((a) => a.status === options.status);
  }

  // 读取时去重：同 device_id + rule_offline + (ACTIVE/ACKNOWLEDGED) 只留最新一条
  // 这是对历史遗留重复数据的保护，正常流程已由 createExecutionErrorAlert 保证唯一
  const offlineGroups = new Map();
  result = result.filter((a) => {
    if (a.rule_id === "rule_offline" && (a.status === "active" || a.status === "acknowledged")) {
      const key = `${a.device_id}::offline-active`;
      const prev = offlineGroups.get(key);
      if (prev) {
        // 已存在 → 保留触发时间新的那条
        if (a.triggered_at > prev.triggered_at) {
          offlineGroups.set(key, a);
          return true; // 这条更新，过滤掉 prev（但 prev 已经在 result 里了，下面处理）
        }
        return false; // 这条旧，丢弃
      }
      offlineGroups.set(key, a);
    }
    return true;
  });
  // 二次清理：把每个 group 里最旧的那条也去掉
  const bestOffline = new Set(offlineGroups.values().map((a) => a.id));
  result = result.filter(
    (a) => !(a.rule_id === "rule_offline" && (a.status === "active" || a.status === "acknowledged")) || bestOffline.has(a.id)
  );

  if (options.level) {
    result = result.filter((a) => a.level === options.level);
  }
  if (options.device_id) {
    result = result.filter((a) => a.device_id === options.device_id);
  }

  return result.sort((a, b) => b.triggered_at - a.triggered_at).slice(0, options.limit || MAX_ALERTS);
}

function getActiveAlerts(readFn) {
  return listAlerts(readFn, { status: ALERT_RECORD_STATUS.ACTIVE, limit: MAX_ALERTS });
}

function getAlert(readFn, alertId) {
  _ensureLoaded(readFn);
  return _alerts.find((a) => a.id === alertId) || null;
}

// ---------- 告警处理 ----------

function acknowledgeAlert(readFn, writeFn, alertId, operator = "system") {
  _ensureLoaded(readFn);
  const alert = _alerts.find((a) => a.id === alertId);
  if (!alert) return null;

  alert.status = ALERT_RECORD_STATUS.ACKNOWLEDGED;
  alert.acknowledged_by = operator;
  alert.acknowledged_at = Date.now();
  _saveAlerts(writeFn);
  return { ...alert };
}

function resolveAlert(readFn, writeFn, alertId, reason = "manual", operator = "system") {
  _ensureLoaded(readFn);
  const alert = _alerts.find((a) => a.id === alertId);
  if (!alert) return null;

  alert.status = ALERT_RECORD_STATUS.RESOLVED;
  alert.resolved_at = Date.now();
  alert.resolved_reason = reason;
  alert.acknowledged_by = alert.acknowledged_by || operator;
  alert.acknowledged_at = alert.acknowledged_at || Date.now();
  _saveAlerts(writeFn);
  return { ...alert };
}

function closeAlert(readFn, writeFn, alertId, note = "") {
  _ensureLoaded(readFn);
  const alert = _alerts.find((a) => a.id === alertId);
  if (!alert) return null;

  alert.status = ALERT_RECORD_STATUS.CLOSED;
  alert.note = note;
  if (!alert.resolved_at) {
    alert.resolved_at = Date.now();
    alert.resolved_reason = "closed";
  }
  _saveAlerts(writeFn);
  return { ...alert };
}

function updateNote(readFn, writeFn, alertId, note) {
  _ensureLoaded(readFn);
  const alert = _alerts.find((a) => a.id === alertId);
  if (!alert) return null;

  alert.note = note;
  _saveAlerts(writeFn);
  return { ...alert };
}

// ---------- 统计 ----------

function getAlertStats(readFn) {
  _ensureLoaded(readFn);

  const active = _alerts.filter((a) => a.status === ALERT_RECORD_STATUS.ACTIVE);
  return {
    total: _alerts.length,
    active: active.length,
    active_critical: active.filter((a) => a.level === ALERT_LEVELS.CRITICAL).length,
    active_warning: active.filter((a) => a.level === ALERT_LEVELS.WARNING).length,
    active_info: active.filter((a) => a.level === ALERT_LEVELS.INFO).length,
    acknowledged: _alerts.filter((a) => a.status === ALERT_RECORD_STATUS.ACKNOWLEDGED).length,
    resolved: _alerts.filter((a) => a.status === ALERT_RECORD_STATUS.RESOLVED).length,
    closed: _alerts.filter((a) => a.status === ALERT_RECORD_STATUS.CLOSED).length,
  };
}

// ---------- 重置（测试用） ----------
function _resetForTest() {
  _alerts = null;
  _counters = null;
}

module.exports = {
  // 告警触发
  processAlert,
  createExecutionErrorAlert,
  // 查询
  listAlerts,
  getActiveAlerts,
  getAlert,
  // 处理
  acknowledgeAlert,
  resolveAlert,
  closeAlert,
  updateNote,
  // 统计
  getAlertStats,
  // 内部
  _resetForTest,
};