Upload 4 files

app.py

@@ -144,7 +144,12 @@ def is_within_rate_limit(api_key):
     while request_counts[api_key] and request_counts[api_key][0] < now - timedelta(seconds=LIMIT_WINDOW):
         request_counts[api_key].popleft()
 
-    return len(request_counts[api_key]) < MAX_REQUESTS
+    if len(request_counts[api_key]) >= MAX_REQUESTS:
+        earliest_request_time = request_counts[api_key][0]
+        wait_time = (earliest_request_time + timedelta(seconds=LIMIT_WINDOW)) - now
+        return False, wait_time.total_seconds()
+    else:
+        return True, 0
 
 def increment_request_count(api_key):
     now = datetime.now()
@@ -211,8 +216,9 @@ def chat_completions():
       return jsonify(error_response), 400
 
     def do_request(current_api_key, attempt):
-        if not is_within_rate_limit(current_api_key):
-            logger.warning(f"{current_api_key[:11]} → 暂时超过限额")
+        isok, time = is_within_rate_limit(current_api_key)
+        if not isok:
+            logger.warning(f"{current_api_key[:11]} → 暂时超过限额，请在 {time} 秒后重试...")
             switch_api_key()
             return False, None