Update main.py

main.py

@@ -271,9 +271,7 @@ async def _engine_call(engine, coro, timeout_sec: float):
 
 # ─── Background Memory Cleanup Task ───────────────────────────────────────────
 _RAM_CLEANUP_THRESHOLD_MB = float(os.environ.get("RAM_CLEANUP_THRESHOLD_MB", "300"))
-_RAM_RESTART_THRESHOLD_MB = float(os.environ.get("RAM_RESTART_THRESHOLD_MB", "1400"))
-_RAM_CLEANUP_INTERVAL_SEC = int(os.environ.get("RAM_CLEANUP_INTERVAL_SEC", "30"))
-_ENGINE_NODES_LIMIT = int(os.environ.get("ENGINE_NODES_LIMIT", "1500000"))
+_RAM_CLEANUP_INTERVAL_SEC = int(os.environ.get("RAM_CLEANUP_INTERVAL_SEC", "60"))
 _CLEAR_HASH_AFTER_MOVE = os.environ.get("CLEAR_HASH_AFTER_MOVE", "1").strip().lower() not in {"0", "false", "no", "off"}
 _RESTART_ENGINE_AFTER_MOVE = os.environ.get("RESTART_ENGINE_AFTER_MOVE", "1").strip().lower() not in {"0", "false", "no", "off"}
 
@@ -288,61 +286,27 @@ async def memory_cleanup_task():
         try:
             process = psutil.Process(os.getpid())
             mem_mb = process.memory_info().rss / 1024 / 1024
-            
-            # Sum up all children (chess engines) memory too
-            total_mb = mem_mb
-            child_count = 0
-            for child in process.children(recursive=True):
-                try:
-                    total_mb += child.memory_info().rss / 1024 / 1024
-                    child_count += 1
-                except (psutil.NoSuchProcess, psutil.AccessDenied):
-                    pass
-            
-            # ─── RESTART CHECK (The Last Resort) ─────────────────────────────
-            if total_mb > _RAM_RESTART_THRESHOLD_MB:
-                active_count = len(manager.active_connections)
-                print(f"[CRITICAL] RAM at {total_mb:.1f}MB — RESTARTING SERVER (Limit: {_RAM_RESTART_THRESHOLD_MB}MB, Active: {active_count})")
-                # Immediate exit: Hugging Face or Docker will automatically restart the process/container
-                # This clears all leaky memory and hung engine processes at once.
-                os._exit(0)
-
-            # ─── REGULAR CLEANUP ─────────────────────────────────────────────
-            if total_mb > _RAM_CLEANUP_THRESHOLD_MB:
-                print(f"[CLEANUP] total_tree_ram={total_mb:.1f}MB (threshold {_RAM_CLEANUP_THRESHOLD_MB}MB) — purging engines & GC")
-                
-                # 1. Purge global engine
+
+            if mem_mb > _RAM_CLEANUP_THRESHOLD_MB:
+                print(f"[CLEANUP] RAM at {mem_mb:.1f}MB (threshold {_RAM_CLEANUP_THRESHOLD_MB}MB) — clearing engine hash")
                 engine = _GLOBAL_DEEPCASTLE_ENGINE
                 if engine is not None:
                     try:
-                        async with _ENGINE_IO_LOCK:
-                            async with _ENGINE_LOCK:
-                                _GLOBAL_DEEPCASTLE_ENGINE = None
-                        await asyncio.wait_for(engine.quit(), timeout=3.0)
-                    except Exception:
-                        pass
-                
-                # 2. Hard purge any orphaned engine processes
-                for child in process.children(recursive=True):
-                    try:
-                        if "deepcastle" in child.name().lower() or "stockfish" in child.name().lower():
-                            print(f"[CLEANUP] Killing orphaned engine: {child.pid}")
-                            child.kill()
+                        if not engine.is_terminated():
+                            async with _ENGINE_IO_LOCK:
+                                await _clear_engine_hash(engine)
                     except Exception:
                         pass
-
-                # 3. Release memory
                 force_memory_release()
                 after_mb = process.memory_info().rss / 1024 / 1024
-                print(f"[CLEANUP] Finished. RAM: {total_mb:.1f}MB → {after_mb:.1f}MB")
+                print(f"[CLEANUP] Done. RAM: {mem_mb:.1f}MB → {after_mb:.1f}MB")
             else:
-                # Basic maintenance
+                # Always nudge GC + malloc_trim even when RAM is fine
                 force_memory_release()
-                if total_mb > _RAM_CLEANUP_THRESHOLD_MB * 0.7:
-                    print(f"[CLEANUP] Maintenance. RAM {total_mb:.1f}MB (OK)")
+                print(f"[CLEANUP] RAM at {mem_mb:.1f}MB — OK")
 
         except Exception as e:
-            print(f"[CLEANUP] Task error: {e}")
+            print(f"[CLEANUP] Error during cleanup: {e}")
 
 
 @asynccontextmanager
@@ -454,7 +418,6 @@ def ram_usage():
         "child_process_count": child_count,
         "vms_mb": round(mem.vms / 1024 / 1024, 2),
         "threshold_mb": _RAM_CLEANUP_THRESHOLD_MB,
-        "restart_threshold_mb": _RAM_RESTART_THRESHOLD_MB,
         "cleanup_interval_sec": _RAM_CLEANUP_INTERVAL_SEC,
         "status": "high" if total_mb > _RAM_CLEANUP_THRESHOLD_MB else "ok",
         "active_rooms": len(manager.active_connections),
@@ -812,7 +775,7 @@ async def analyze_game(request: AnalyzeRequest):
     try:
         engine = await get_stockfish_engine(hash_mb=2048)
         board = chess.Board(request.start_fen) if request.start_fen else chess.Board()
-        limit = chess.engine.Limit(time=request.time_per_move, nodes=_ENGINE_NODES_LIMIT)
+        limit = chess.engine.Limit(time=request.time_per_move)
 
         analysis_results = []
         ply_timeout = _analyze_ply_timeout(request.time_per_move)