feat(api): Redis 冪等鎖防止重複修復 (Sprint 3 T4)

雙層鎖設計: in-process asyncio.Lock (必定生效) + Redis 分散式鎖 (跨 Pod best-effort)
同一 URI 的第二次修復呼叫立即返回 "already running" 錯誤

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

apps/api/src/services/host_repair_agent.py

@@ -145,6 +145,17 @@ def build_repair_command(component: str) -> str:
 class HostRepairAgent:
     """透過 SSH 執行主機層修復命令。"""
 
+    def __init__(self) -> None:
+        # in-process 鎖表 — key: lock_key → asyncio.Lock
+        # 2026-04-06 Claude Code: Sprint 3 T4
+        self._in_process_locks: dict[str, asyncio.Lock] = {}
+
+    def _get_in_process_lock(self, lock_key: str) -> asyncio.Lock:
+        """取得或建立指定 key 的 in-process 鎖。"""
+        if lock_key not in self._in_process_locks:
+            self._in_process_locks[lock_key] = asyncio.Lock()
+        return self._in_process_locks[lock_key]
+
     async def repair(self, layer: str, component: str) -> HostRepairResult:
         """執行修復並回傳結果。"""
         try:
@@ -193,37 +204,89 @@ class HostRepairAgent:
         """
         根據 URI scheme 路由至對應的執行路徑。
         2026-04-06 Claude Code: Sprint 3 T3
+        2026-04-06 Claude Code: Sprint 3 T4 — Redis 冪等鎖防止重複修復
         """
         try:
             uri = parse_uri_command(command)
         except ValueError as e:
             return HostRepairResult(success=False, layer="", component="", error=str(e))
 
-        if uri.scheme == "openclaw":
-            return await self._execute_openclaw(uri.host_or_layer, uri.payload)
+        # 冪等鎖 — 防止同一 component 並發修復
+        # 雙層鎖: in-process asyncio.Lock (必定生效) + Redis 分散式鎖 (best-effort)
+        # 2026-04-06 Claude Code: Sprint 3 T4
+        lock_key = f"repair_lock:ssh_command:{uri.scheme}:{uri.host_or_layer}:{uri.payload}"
+        in_process_lock = self._get_in_process_lock(lock_key)
 
-        if uri.scheme == "ansible":
+        async def _execute() -> HostRepairResult:
+            if uri.scheme == "openclaw":
+                return await self._execute_openclaw(uri.host_or_layer, uri.payload)
+
+            if uri.scheme == "ansible":
+                try:
+                    validate_ansible_playbook(uri.payload)
+                except ValueError as e:
+                    return HostRepairResult(success=False, layer="ansible", component=uri.payload, error=str(e))
+                return await self._execute_ansible(uri.host_or_layer, uri.payload)
+
+            if uri.scheme == "ssh":
+                if not approved:
+                    return HostRepairResult(
+                        success=False,
+                        layer="ssh",
+                        component=uri.payload,
+                        error="ssh:// scheme requires_approval=True — must be explicitly approved",
+                    )
+                try:
+                    validate_shell_safety(uri.payload)
+                except ValueError as e:
+                    return HostRepairResult(success=False, layer="ssh", component=uri.payload, error=str(e))
+                return await self._execute_ssh_direct(uri.host_or_layer, uri.payload)
+
+            return HostRepairResult(success=False, layer="", component="", error=f"Unhandled scheme: {uri.scheme}")
+
+        # in-process 鎖: locked() 代表正在進行，立即拒絕重複
+        if in_process_lock.locked():
+            return HostRepairResult(
+                success=False,
+                layer=uri.scheme,
+                component=uri.payload,
+                error=f"Repair already running for {uri.scheme}://{uri.host_or_layer}/{uri.payload}",
+            )
+
+        async with in_process_lock:
+            # Redis 分散式鎖 (best-effort，跨 Pod)
+            # blocking_timeout=0: 立即失敗，不等待
             try:
-                validate_ansible_playbook(uri.payload)
-            except ValueError as e:
-                return HostRepairResult(success=False, layer="ansible", component=uri.payload, error=str(e))
-            return await self._execute_ansible(uri.host_or_layer, uri.payload)
+                from src.core.redis_client import RedisLock
+                redis_lock: RedisLock | None = RedisLock(lock_key, timeout=120, blocking_timeout=0)
+            except Exception:
+                redis_lock = None  # Redis 未連線，fail open
 
-        if uri.scheme == "ssh":
-            if not approved:
+            if redis_lock is None:
+                return await _execute()
+
+            try:
+                acquired = await redis_lock.acquire()
+            except Exception:
+                # Redis 不可用，fail open
+                return await _execute()
+
+            if not acquired:
+                # Redis 鎖已被其他 Pod 持有
                 return HostRepairResult(
                     success=False,
-                    layer="ssh",
+                    layer=uri.scheme,
                     component=uri.payload,
-                    error="ssh:// scheme requires_approval=True — must be explicitly approved",
+                    error=f"Repair already running for {uri.scheme}://{uri.host_or_layer}/{uri.payload}",
                 )
-            try:
-                validate_shell_safety(uri.payload)
-            except ValueError as e:
-                return HostRepairResult(success=False, layer="ssh", component=uri.payload, error=str(e))
-            return await self._execute_ssh_direct(uri.host_or_layer, uri.payload)
 
-        return HostRepairResult(success=False, layer="", component="", error=f"Unhandled scheme: {uri.scheme}")
+            try:
+                return await _execute()
+            finally:
+                try:
+                    await redis_lock.release()
+                except Exception:
+                    pass
 
     async def _execute_openclaw(self, layer: str, component: str) -> HostRepairResult:
         """openclaw:// — 呼叫現有的 repair(layer, component) 邏輯"""