fix try/finally with continue, fix wrong subclass signature

llm_server/routes/openai_request_handler.py

@@ -74,7 +74,7 @@ class OpenAIRequestHandler(RequestHandler):
         # Ratelimited: you are only allowed to have {opts.simultaneous_requests_per_ip} simultaneous requests at a time. Please complete your other requests before sending another.
         return 'Ratelimited', 429
 
-    def handle_error(self, error_msg: str) -> Tuple[flask.Response, int]:
+    def handle_error(self, error_msg: str, error_type: str = 'error') -> Tuple[flask.Response, int]:
         # TODO: return a simulated OpenAI error message
         return jsonify({
             "error": {

llm_server/workers/blocking.py

@@ -16,19 +16,13 @@ def worker():
         increment_ip_count(client_ip, 'processing_ips')
         redis.incr('active_gen_workers')
 
-        try:
         if not request_json_body:
             # This was a dummy request from the websocket handler.
             # We're going to let the websocket handler decrement processing_ips and active_gen_workers.
             continue
 
-            start_time = time.time()
+        try:
             success, response, error_msg = generator(request_json_body)
-            end_time = time.time()
-
-            elapsed_time = end_time - start_time
-            # redis.rpush('generation_elapsed', json.dumps((end_time, elapsed_time)))
-
             event = DataEvent(event_id)
             event.set((success, response, error_msg))
         finally: