feat: auto-heal pipeline — detect+restart gateway/bot/ejabberd on failure

server.py

@@ -1315,6 +1315,20 @@ def api_xmpp_stats():
         return jsonify({"today": {"sent": 0, "failed": 0, "latency_avg": 0}, "week": {"sent": 0, "failed": 0}})
 
 
+@app.route("/api/xmpp/autoheal", methods=["GET", "POST"])
+def api_xmpp_autoheal():
+    """自愈：检测异常并自动修复。GET=查看状态, POST=执行修复"""
+    try:
+        from xmpp_logger import auto_heal
+        if request.method == "POST":
+            result = auto_heal()
+            return jsonify(result)
+        else:
+            return jsonify({"usage": "POST to trigger auto-heal"})
+    except ImportError:
+        return jsonify({"error": "xmpp_logger not available"}), 500
+
+
 # 注册提示词管理路由
 register_routes(app)
 

xmpp_logger.py

@@ -9,6 +9,7 @@
 """
 import json
 import time as _time
+import subprocess as _sp
 from pathlib import Path
 from datetime import datetime, timedelta
 
@@ -230,3 +231,46 @@ def health():
         elif la > 0: result["status"] = "ok"
 
     return result
+
+
+def auto_heal():
+    """自愈：检测并尝试修复 XMPP 通信问题。"""
+    h = health()
+    actions = []
+
+    # 1. LLM Provider 超时 → 重启 Hermes Gateway
+    if h.get("llm_provider", {}).get("status") in ("timeout", "error"):
+        gw = h.get("gateways", {}).get("zhiwei", {})
+        if gw.get("alive"):
+            try:
+                r = _sp.run(["sudo", "-n", "systemctl", "restart", "hermes-gw-pa"],
+                           capture_output=True, timeout=30, text=True)
+                ok = r.returncode == 0
+                actions.append({"action": "restart_hermes_gateway", "target": "position-analyst",
+                              "success": ok, "detail": "restarted" if ok else r.stderr[:100]})
+            except Exception as e:
+                actions.append({"action": "restart_hermes_gateway", "target": "position-analyst",
+                              "success": False, "detail": str(e)[:100]})
+
+    # 2. Bot 无出站 → 重启知微 Bot
+    ba = h.get("bot_activity", {})
+    if ba.get("inbound", 0) > 0 and ba.get("outbound", 0) == 0 and ba.get("errors", 0) > 0:
+        try:
+                            r = _sp.run(["sudo", "-n", "systemctl", "restart", "xmpp-zhiwei"], capture_output=True, timeout=30, text=True)
+            actions.append({"action": "restart_zhiwei_bot", "target": "xmpp-zhiwei",
+                          "success": r.returncode == 0, "detail": "restarted" if r.returncode == 0 else r.stderr[:100]})
+        except Exception as e:
+            actions.append({"action": "restart_zhiwei_bot", "target": "xmpp-zhiwei",
+                          "success": False, "detail": str(e)[:100]})
+
+    # 3. ejabberd → 重启容器
+    if h.get("ejabberd", "") and "Up" not in str(h["ejabberd"]):
+        try:
+                            r = _sp.run(["sudo", "-n", "docker", "restart", "ejabberd"], capture_output=True, timeout=30, text=True)
+            actions.append({"action": "restart_ejabberd", "target": "ejabberd",
+                          "success": r.returncode == 0, "detail": "restarted" if r.returncode == 0 else r.stderr[:100]})
+        except Exception as e:
+            actions.append({"action": "restart_ejabberd", "target": "ejabberd",
+                          "success": False, "detail": str(e)[:100]})
+
+    return {"actions": actions, "status": h.get("status", "unknown")}