feat: 接入企微智能机器人——异常报警/数据核验日报/设备健康日报

- flows/shared/wecom.py: WSClient 按需连接(连接→认证→发送→断开),全容错,超长分段 - flows/monitor/alert_failure.py: Prefect 事件触发(Failed/Crashed→run deployment),发群报警 - flows/monitor/daily_report.py: 每日 18:30 近7天 RIIN/SDS 核验(复用 verify)+商品完整性+OCR+失败回顾 - flows/monitor/device_check.py: 每日 09:00 内存/磁盘/GPU(80%阈值)+Docker 容器异常+服务探活 - db.py: 新增 get_goods_integrity/ping; compose worker 挂载 docker.sock+/:/host:ro+GPU - scripts/wecom_capture_chatid.py: 一次性捕获群 chat_id 工具
2026-08-27 09:40:53 +00:00
parent b9e0bc42be
commit 1d0da2a00b
12 changed files with 836 additions and 0 deletions
@@ -0,0 +1,129 @@
+"""企微机器人异常报警:任意 flow run 失败/崩溃时由 Prefect 事件触发器自动调用。
+
+对应 wecom-alert deployment(prefect.yaml 里的 triggers 声明):
+- match: prefect.resource.id = prefect.flow-run.*
+- expect: prefect.flow-run.Failed / Crashed
+- parameters.event = "{{ event }}"
+
+flow 体全容错——本 flow 自身永不进入 Failed 状态,避免自我触发循环。
+event 可能以 dict 或 str(jinja 渲染结果)传入,均做防御性解析;
+解析出 flow run id 后再查 Prefect API 补全权威信息(名称/错误消息/时间)。
+"""
+
+from __future__ import annotations
+
+import json
+import os
+import re
+from datetime import datetime
+from typing import Any
+
+import httpx
+from prefect import flow
+
+from flows.shared import send_markdown
+
+UI_BASE = os.environ.get("PREFECT_UI_URL", "http://192.168.124.92:4200")
+
+
+def _parse_event(event: Any) -> dict:
+    if isinstance(event, dict):
+        return event
+    if isinstance(event, str) and event.strip():
+        # jinja "{{ event }}" 可能渲染成 JSON 串或 Python repr,统一抠除关键字段
+        rid = re.search(r"prefect\.flow-run\.[0-9a-fA-F-]{36}", event)
+        ev = re.search(r"prefect\.flow-run\.(?:Failed|Crashed)", event)
+        name = re.search(r"prefect\.resource\.name['\"]?\s*[:=]\s*['\"]([^'\"]+)", event)
+        occurred = re.search(r"occurred['\"]?\s*[:=]\s*['\"]([^'\"]+)", event)
+        return {
+            "event": ev.group(0) if ev else "",
+            "resource": {
+                "prefect.resource.id": rid.group(0) if rid else "",
+                "prefect.resource.name": name.group(1) if name else "",
+            },
+            "occurred": occurred.group(1) if occurred else "",
+        }
+    return {}
+
+
+def _flow_run_id(event: dict) -> str | None:
+    rid = (event.get("resource") or {}).get("prefect.resource.id", "")
+    m = re.match(r"^prefect\.flow-run\.(.+)$", rid)
+    return m.group(1) if m else None
+
+
+def _fetch_run_detail(run_id: str) -> dict:
+    """查 Prefect API 补全 flow run 权威信息,失败返回空 dict。"""
+    api_url = os.environ.get("PREFECT_API_URL", "http://localhost:4200/api")
+    try:
+        resp = httpx.get(f"{api_url}/flow_runs/{run_id}", timeout=10)
+        resp.raise_for_status()
+        return resp.json()
+    except Exception as err:  # noqa: BLE001
+        print(f"[wecom-alert] 查询 flow run {run_id} 失败: {err}")
+        return {}
+
+
+def _fmt_ts(value: Any) -> str:
+    if not value:
+        return "—"
+    try:
+        return datetime.fromisoformat(value.replace("Z", "+00:00")).astimezone().strftime("%m-%d %H:%M")
+    except (ValueError, TypeError):
+        return str(value)[:19]
+
+
+@flow(name="wecom-alert", log_prints=True)
+def wecom_alert_failure(event: dict | str | None = None) -> dict:
+    """Prefect 事件触发:flow run 失败/崩溃 → 发企微群报警。永不抛异常。"""
+    result = {"sent": False, "flowName": None, "runId": None}
+    try:
+        ev = _parse_event(event)
+        event_name = ev.get("event", "prefect.flow-run.?")
+        resource_name = (ev.get("resource") or {}).get("prefect.resource.name", "")
+        run_id = _flow_run_id(ev)
+
+        flow_name, run_name, state_msg, occurred = "", resource_name, "", ev.get("occurred", "")
+        if run_id:
+            detail = _fetch_run_detail(run_id)
+            flow_name = detail.get("name", "") or resource_name
+            run_name = detail.get("name", "")
+            state = detail.get("state") or {}
+            state_msg = (state.get("message") or "").strip()
+            ended = detail.get("end_time") or detail.get("expected_start_time")
+            occurred = occurred or ended
+
+        lines = [
+            "🔴 **Prefect 流程失败报警**",
+            f"- 事件: `{event_name}`",
+            f"- Flow: {flow_name or resource_name or '未知'}",
+        ]
+        if run_name and run_name != flow_name:
+            lines.append(f"- Run: {run_name}")
+        if state_msg:
+            lines.append(f"- 错误: {state_msg[:200].replace(chr(10), ' ')}")
+        lines.append(f"- 时间: {_fmt_ts(occurred)}")
+        if run_id:
+            lines.append(f"- 详情: [Prefect UI]({UI_BASE}/runs/flow-run/{run_id})")
+        else:
+            raw = json.dumps(ev, ensure_ascii=False)[:300] if ev else "(事件内容解析失败)"
+            lines.append(f"- 原始事件: {raw}")
+
+        result["sent"] = send_markdown("\n".join(lines))
+        result.update({"flowName": flow_name or None, "runId": run_id})
+        print(f"[wecom-alert] 已发送失败报警: {flow_name or resource_name} ({event_name})")
+    except Exception as err:  # noqa: BLE001
+        print(f"[wecom-alert] 报警流程异常(已吞掉,避免自我触发): {err}")
+    return result
+
+
+if __name__ == "__main__":
+    # 本地手测:模拟一条 Failed 事件
+    wecom_alert_failure({
+        "event": "prefect.flow-run.Failed",
+        "resource": {
+            "prefect.resource.id": "prefect.flow-run.00000000-0000-0000-0000-000000000000",
+            "prefect.resource.name": "riin-order-sync/manual",
+        },
+        "occurred": datetime.now().astimezone().isoformat(),
+    })