perf(api): HostAggregator 效能優化 — probe timeout 縮短 + 30 秒記憶體快取

apps/api/src/services/host_aggregator.py

@@ -197,7 +197,7 @@ def get_baseline_context_for_llm(metrics: HostMetrics, host_name: str) -> str:
 # Real Host Probing
 # =============================================================================
 
-async def _tcp_probe(ip: str, port: int, timeout: float = 3.0) -> tuple[bool, float | None, str | None]:
+async def _tcp_probe(ip: str, port: int, timeout: float = 1.5) -> tuple[bool, float | None, str | None]:
     """
     Real TCP port probe using asyncio.open_connection
 
@@ -236,7 +236,7 @@ async def _http_probe(
     ip: str,
     port: int,
     path: str,
-    timeout: float = 5.0,
+    timeout: float = 2.0,
     https: bool = False
 ) -> tuple[bool, float | None, str | None]:
     """
@@ -328,8 +328,18 @@ class HostAggregator:
 
     Uses asyncio.gather for parallel fetching of all host statuses.
     Performs real TCP/HTTP probes to determine service availability.
+
+    2026-04-08 Claude Code: Sprint 5 效能優化
+    - TCP timeout: 3.0s → 1.5s
+    - HTTP timeout: 5.0s → 2.0s
+    - 記憶體快取: 上次成功結果 + 30 秒 TTL
     """
 
+    # 記憶體快取 (避免 probe 超時時前端空白)
+    _cache: AggregatedStatus | None = None
+    _cache_time: float = 0
+    _CACHE_TTL = 30  # 秒
+
     @classmethod
     async def _probe_service(
         cls,
@@ -425,6 +435,7 @@ class HostAggregator:
             metrics=metrics,
         )
 
+    @classmethod
     @classmethod
     async def fetch_all(cls) -> AggregatedStatus:
         """
@@ -432,7 +443,15 @@ class HostAggregator:
 
         Uses asyncio.gather for maximum concurrency.
         Always performs real probing - no mock data.
+        快取: 30 秒內返回快取結果，避免前端等待超時
         """
+        import time as _time
+
+        # 快取命中: 30 秒內直接回傳
+        now = _time.monotonic()
+        if cls._cache is not None and (now - cls._cache_time) < cls._CACHE_TTL:
+            return cls._cache
+
         logger.info("aggregator_fetch_start", mode="real_probing")
 
         # Fetch all hosts in parallel
@@ -480,7 +499,7 @@ class HostAggregator:
             host_statuses={h.ip: h.status for h in hosts},
         )
 
-        return AggregatedStatus(
+        result = AggregatedStatus(
             timestamp=datetime.now(UTC),
             environment=settings.ENVIRONMENT,
             mock_mode=False,  # Always real mode
@@ -488,6 +507,12 @@ class HostAggregator:
             hosts=hosts,
         )
 
+        # 更新快取
+        cls._cache = result
+        cls._cache_time = _time.monotonic()
+
+        return result
+
     @classmethod
     async def fetch_single(cls, ip: str) -> HostStatus | None:
         """Fetch status from a single host"""