perf: server/status接口优化(cpu非阻塞+pm2直接调用+COUNT估算+5秒缓存)

backend/main.py

@@ -458,11 +458,40 @@ async def get_signal_trades(
 
 import shutil, subprocess, psutil
 
+# 服务器状态缓存（避免重复调用慢操作）
+_server_cache: dict = {"data": None, "ts": 0}
+_PM2_BIN = None
+
+def _find_pm2_bin():
+    """找到pm2二进制路径，避免每次走npx"""
+    global _PM2_BIN
+    if _PM2_BIN:
+        return _PM2_BIN
+    import shutil as _sh
+    for p in ["/home/fzq1228/.local/bin/pm2", "/usr/local/bin/pm2", "/usr/bin/pm2"]:
+        if os.path.exists(p):
+            _PM2_BIN = p
+            return p
+    found = _sh.which("pm2")
+    if found:
+        _PM2_BIN = found
+        return found
+    return "npx pm2"
+
+# 启动时初始化CPU采样（首次调用不阻塞）
+psutil.cpu_percent(interval=None)
+
 @app.get("/api/server/status")
 async def get_server_status(user: dict = Depends(get_current_user)):
     """服务器全状态：CPU/内存/硬盘/负载/PM2进程/PG数据库/回补进度"""
-    # CPU
+
-    cpu_percent = psutil.cpu_percent(interval=0.5)
+    # 5秒缓存，避免频繁调用慢操作
+    now = time.time()
+    if _server_cache["data"] and (now - _server_cache["ts"]) < 5:
+        return _server_cache["data"]
+
+    # CPU（非阻塞，取上次采样间隔的值）
+    cpu_percent = psutil.cpu_percent(interval=None)
     cpu_count = psutil.cpu_count()
 
     # 内存
@@ -482,12 +511,14 @@ async def get_server_status(user: dict = Depends(get_current_user)):
     # 网络IO
     net = psutil.net_io_counters()
 
-    # PM2进程状态
+    # PM2进程状态（直接调pm2二进制，不走npx）
     pm2_procs = []
     try:
+        pm2_bin = _find_pm2_bin()
+        cmd = [pm2_bin, "jlist"] if not pm2_bin.startswith("npx") else ["npx", "pm2", "jlist"]
         result = subprocess.run(
-            ["npx", "pm2", "jlist"],
+            cmd,
-            capture_output=True, text=True, timeout=10
+            capture_output=True, text=True, timeout=5
         )
         import json as _json
         procs = _json.loads(result.stdout)
@@ -504,7 +535,7 @@ async def get_server_status(user: dict = Depends(get_current_user)):
     except Exception:
         pm2_procs = []
 
-    # PG数据库大小 + agg_trades条数
+    # PG数据库大小 + agg_trades条数（用估算值，快1000倍）
     pg_info = {}
     try:
         row = await async_fetchrow(
@@ -512,10 +543,15 @@ async def get_server_status(user: dict = Depends(get_current_user)):
         )
         pg_info["db_size_mb"] = round(row["db_size"] / 1024 / 1024, 1) if row else 0
 
-        row2 = await async_fetchrow("SELECT COUNT(*) as cnt FROM agg_trades")
+        # 用PG统计信息估算行数（毫秒级，而非COUNT(*)的秒级全表扫描）
-        pg_info["agg_trades_count"] = row2["cnt"] if row2 else 0
+        row2 = await async_fetchrow(
+            "SELECT SUM(n_live_tup)::bigint as cnt FROM pg_stat_user_tables WHERE relname LIKE 'agg_trades%'"
+        )
+        pg_info["agg_trades_count"] = row2["cnt"] if row2 and row2["cnt"] else 0
 
-        row3 = await async_fetchrow("SELECT COUNT(*) as cnt FROM rate_snapshots")
+        row3 = await async_fetchrow(
+            "SELECT n_live_tup::bigint as cnt FROM pg_stat_user_tables WHERE relname = 'rate_snapshots'"
+        )
         pg_info["rate_snapshots_count"] = row3["cnt"] if row3 else 0
 
         # 各symbol最新数据时间
@@ -542,7 +578,7 @@ async def get_server_status(user: dict = Depends(get_current_user)):
     except Exception:
         pass
 
-    return {
+    result = {
         "timestamp": int(time.time() * 1000),
         "cpu": {
             "percent": cpu_percent,
@@ -574,3 +610,6 @@ async def get_server_status(user: dict = Depends(get_current_user)):
         "postgres": pg_info,
         "backfill_running": backfill_running,
     }
+    _server_cache["data"] = result
+    _server_cache["ts"] = now
+    return result