EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

[PD] Propagate internal server errors from aborted requests to clients instead of blindly returning 200's (#8936)

Browse Source
This commit is contained in:
datdo-msft
2025-08-18 14:23:46 -07:00
committed by GitHub
parent 6805f6da40
commit 98b44e9e56
4 changed files with 10 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
python/sglang/srt/disaggregation/decode.py
View File

@@ -259,7 +259,7 @@ class DecodePreallocQueue:
if len(req.origin_input_ids) > self.max_total_num_tokens:
message = f"Request {req.rid} exceeds the maximum number of tokens: {len(req.origin_input_ids)} > {self.max_total_num_tokens}"
logger.error(message)
prepare_abort(req, message)
prepare_abort(req, message, status_code=HTTPStatus.BAD_REQUEST)
self.scheduler.stream_output([req], req.return_logprob)
return True
return False

2
python/sglang/srt/disaggregation/prefill.py
View File

@@ -178,7 +178,7 @@ class PrefillBootstrapQueue:
if len(req.origin_input_ids) > self.max_total_num_tokens:
message = f"Request {req.rid} exceeds the maximum number of tokens: {len(req.origin_input_ids)} > {self.max_total_num_tokens}"
logger.error(message)
prepare_abort(req, message)
prepare_abort(req, message, status_code=HTTPStatus.BAD_REQUEST)
self.scheduler.stream_output([req], req.return_logprob)
return True
return False