⚡ Bolt: Optimize RequestMetrics.to_dict for faster serialization

.jules/bolt.md

@@ -0,0 +1,3 @@
+## 2025-03-11 - Optimize dataclass serialization with __slots__
+**Learning:** `dataclasses.asdict` can be a significant performance bottleneck when dealing with high-frequency serialization of dataclasses, especially nested ones. For dataclasses that use `slots=True` (like `RequestMetrics`), iterating over `__slots__` and using `getattr` to manually construct the dictionary is significantly faster than using `asdict`. Nested dataclasses can still be supported efficiently by dynamically checking for `hasattr(v, '__dataclass_fields__')` and falling back to `asdict(v)` only for the nested components.
+**Action:** When optimizing frequent dataclass serialization in Python, check if the dataclass uses `slots=True` and prefer a custom `to_dict` method using `__slots__` iteration over `dataclasses.asdict()`.

fastdeploy/engine/request.py

@@ -895,7 +895,10 @@ def to_dict(self):
         """
         Convert the RequestMetrics object to a dictionary.
         """
-        return {k: v for k, v in asdict(self).items()}
+        # Optimized: Avoiding dataclasses.asdict() because its recursive deepcopy
+        # and validation overhead is a significant bottleneck for high-frequency calls.
+        # Iterating over __slots__ is significantly faster for flat structs.
+        return {k: getattr(self, k) for k in self.__slots__}
 
     def record_recv_first_token(self):
         cur_time = time.time()