apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-09-28 21:32:29 +08:00
Code Issues Actions 6 Packages Projects Releases Wiki Activity

[Fix] Do not drop result when request result slowly (#3704)

Browse Source 
* [Fix] Do not drop result when request result slowly

* set default FD_ZMQ_SNDHWM to 64k
This commit is contained in:
chenjian
2025-09-01 10:14:04 +08:00
committed by GitHub
parent b9af800edd
commit 10a95f8ed5
1 changed files with 13 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
fastdeploy/inter_communicator/zmq_server.py
View File

@@ -97,8 +97,17 @@ class ZmqServerBase(ABC):
with self.response_token_lock: with self.response_token_lock:
client, _, request_id = self.socket.recv_multipart(flags=zmq.NOBLOCK) client, _, request_id = self.socket.recv_multipart(flags=zmq.NOBLOCK)
req_id_str = request_id.decode("utf-8") req_id_str = request_id.decode("utf-8")
need_send_after_finished_inference = False
with self.mutex: with self.mutex:
self.req_dict[req_id_str] = client self.req_dict[req_id_str] = client
if req_id_str in self.cached_results:
if self.cached_results[req_id_str][-1][-1].finished:
need_send_after_finished_inference = True
if need_send_after_finished_inference:
self.send_response(req_id_str, [])
llm_logger.info(f"send_multipart finished, req_id: {req_id_str}")
self.req_dict.pop(req_id_str, None)
except zmq.Again: except zmq.Again:
time.sleep(0.001) time.sleep(0.001)
continue continue
@@ -144,15 +153,12 @@ class ZmqServerBase(ABC):
except Exception as e: except Exception as e:
llm_logger.error(f"Send result to zmq client failed: {e}") llm_logger.error(f"Send result to zmq client failed: {e}")
if data[-1].finished: if data and data[-1].finished:
with self.mutex: with self.mutex:
if req_id not in self.req_dict: if req_id in self.req_dict:
llm_logger.warning(f"req_id {req_id} finished but no result handle, drop it")
if req_id in self.cached_results:
del self.cached_results[req_id]
else:
llm_logger.info(f"send_multipart finished, req_id: {req_id}") llm_logger.info(f"send_multipart finished, req_id: {req_id}")
self.req_dict.pop(req_id, None) self.req_dict.pop(req_id, None)
@abstractmethod @abstractmethod
def close(self): def close(self):