fix: 恢复 HarnessRL HWPOD 重启幂等

tools/hwlab-node.py

@@ -11,6 +11,7 @@
 from __future__ import annotations
 
 import base64
+import calendar
 import ctypes
 import ctypes.wintypes as wt
 import hashlib
@@ -61,6 +62,7 @@ LOG_DIR = CONFIG_DIR / "logs"
 UPDATE_DIR = CONFIG_DIR / "update"
 CONFIG_PATH = CONFIG_DIR / "config.json"
 STATE_PATH = CONFIG_DIR / "state.json"
+OPERATION_LEDGER_PATH = CONFIG_DIR / "hwpod-operations.json"
 LOG_PATH = LOG_DIR / "hwlab-node.log"
 
 DEFAULT_CONFIG = {
@@ -832,6 +834,7 @@ class WebSocketNodeClient:
         self.manual_stop = False
         self.connected = False
         self.diagnostic_cursor = 0
+        self.operation_retention_seconds = 0
 
     def start(self) -> None:
         if self.thread and self.thread.is_alive():
@@ -972,18 +975,52 @@ class WebSocketNodeClient:
         message = json.loads(text)
         if message.get("type") == "ack" and message.get("requestId") == "register":
             self.connected = message.get("ok") is True
+            self.operation_retention_seconds = safe_int(message.get("operationRetentionSeconds"), 0)
             self.bus.emit("connection", "connected" if self.connected else "error")
             self.bus.emit("log", f"注册结果: ok={message.get('ok')} message={message.get('message')}")
             return
         if message.get("type") != "hwpod-node-ops":
             return
         request_id = str(message.get("requestId") or "req_unknown")
+        plan = message.get("plan") or {}
+        plan_id = str(plan.get("planId") or request_id)
+        ledger = self._operation_ledger()
+        existing = ledger.get(plan_id)
+        self._send_json({"type": "hwpod-node-ops-accepted", "nodeId": self.executor.node_id, "requestId": request_id, "planId": plan_id, "observedAt": utc_now()})
+        if isinstance(existing, dict) and existing.get("status") == "completed" and isinstance(existing.get("result"), dict):
+            self._send_json({"type": "hwpod-node-ops-result", "nodeId": self.executor.node_id, "requestId": request_id, "result": existing["result"], "observedAt": utc_now()})
+            return
+        ledger[plan_id] = {"status": "accepted", "acceptedAt": utc_now(), "result": None}
+        self._save_operation_ledger(ledger)
         try:
-            result = self.executor.execute(message.get("plan") or {})
+            result = self.executor.execute(plan)
         except Exception as exc:
             result = {"ok": False, "status": "failed", "results": [], "blocker": {"code": "hwpod_node_ws_run_failed", "layer": "hwpod-node", "retryable": True, "summary": str(exc)}}
+        ledger[plan_id] = {"status": "completed", "acceptedAt": ledger[plan_id]["acceptedAt"], "completedAt": utc_now(), "result": result}
+        self._save_operation_ledger(ledger)
         self._send_json({"type": "hwpod-node-ops-result", "nodeId": self.executor.node_id, "requestId": request_id, "result": result, "observedAt": utc_now()})
 
+    def _operation_ledger(self) -> dict:
+        try:
+            value = json.loads(OPERATION_LEDGER_PATH.read_text(encoding="utf-8"))
+            ledger = value if isinstance(value, dict) else {}
+        except Exception:
+            ledger = {}
+        if self.operation_retention_seconds > 0:
+            cutoff = time.time() - self.operation_retention_seconds
+            ledger = {key: value for key, value in ledger.items() if not isinstance(value, dict) or self._operation_timestamp(value) >= cutoff}
+        return ledger
+
+    def _save_operation_ledger(self, ledger: dict) -> None:
+        atomic_write_json(OPERATION_LEDGER_PATH, ledger)
+
+    def _operation_timestamp(self, value: dict) -> float:
+        timestamp = str(value.get("completedAt") or value.get("acceptedAt") or "")
+        try:
+            return calendar.timegm(time.strptime(timestamp, "%Y-%m-%dT%H:%M:%SZ"))
+        except Exception:
+            return time.time()
+
     def _flush_diagnostics(self) -> None:
         for event in DIAGNOSTICS.since(self.diagnostic_cursor, 20):
             self._send_json({"type": "hwpod-node-diagnostic", "nodeId": self.executor.node_id, "diagnostic": event, "observedAt": utc_now()})