perf(disaggregation): slice decode waiting queue for prebuilt batches

python/sglang/srt/disaggregation/decode.py

@@ -1212,23 +1212,16 @@ class SchedulerDisaggregationDecodeMixin:
 
         num_not_used_batch = batch_size - curr_batch_size
 
-        # pop req from waiting queue
-        can_run_list: List[Req] = []
-        waiting_queue: List[Req] = []
-
-        for i in range(len(self.waiting_queue)):
-            req = self.waiting_queue[i]
-            # we can only add at least `num_not_used_batch` new batch to the running queue
-            if i < num_not_used_batch:
-                can_run_list.append(req)
-                req.init_next_round_input(self.tree_cache)
-            else:
-                waiting_queue.append(req)
-
-        self.waiting_queue = waiting_queue
-        if len(can_run_list) == 0:
+        num_new_reqs = min(num_not_used_batch, len(self.waiting_queue))
+        if num_new_reqs <= 0:
             return None
 
+        can_run_list = self.waiting_queue[:num_new_reqs]
+        self.waiting_queue = self.waiting_queue[num_new_reqs:]
+
+        for req in can_run_list:
+            req.init_next_round_input(self.tree_cache)
+
         set_time_batch(can_run_list, "set_forward_entry_time")
 
         # construct a schedule batch with those requests and mark as decode