xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-20 06:35:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[V1][Bugfix] Clean up requests when aborted (#9629)

Browse Source 
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
This commit is contained in:
Woosuk Kwon 2024-10-23 23:33:22 -07:00 committed by GitHub
parent 836e8ef6ee
commit 3770071eb4
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 12 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
vllm/v1/engine/llm_engine.py
View File

@ -300,6 +300,7 @@ class LLMEngine:
def abort_request(self, request_id: Union[str, Iterable[str]]) -> None: def abort_request(self, request_id: Union[str, Iterable[str]]) -> None:
self.scheduler.finish_requests(request_id, self.scheduler.finish_requests(request_id,
RequestStatus.FINISHED_ABORTED) RequestStatus.FINISHED_ABORTED)
self._free_request(request_id)
def get_num_unfinished_requests(self) -> int: def get_num_unfinished_requests(self) -> int:
"""Gets the number of unfinished requests.""" """Gets the number of unfinished requests."""
@ -361,6 +362,11 @@ class LLMEngine:
num_reqs = len(detokenizer_output.req_ids) num_reqs = len(detokenizer_output.req_ids)
for i in range(num_reqs): for i in range(num_reqs):
req_id = detokenizer_output.req_ids[i] req_id = detokenizer_output.req_ids[i]
if req_id not in self.requests:
# The request has been aborted while the detokenizer was
# processing the outputs.
continue
req = self.requests[req_id] req = self.requests[req_id]
req.output_text += detokenizer_output.detokenized_texts[i] req.output_text += detokenizer_output.detokenized_texts[i]
@ -373,9 +379,7 @@ class LLMEngine:
req_outputs.append(req_output) req_outputs.append(req_output)
if finished: if finished:
del self.requests[req_id] self._free_request(req_id)
del self.num_lagged_steps[req_id]
del self.request_outputs[req_id]
return req_outputs return req_outputs
def terminate_detokenizer(self) -> None: def terminate_detokenizer(self) -> None:
@ -440,6 +444,11 @@ class LLMEngine:
req_output.finished = finished req_output.finished = finished
return req_output return req_output
def _free_request(self, request_id: str) -> None:
self.requests.pop(request_id, None)
self.num_lagged_steps.pop(request_id, None)
self.request_outputs.pop(request_id, None)
def check_health(self) -> None: def check_health(self) -> None:
if self.tokenizer: if self.tokenizer:
self.tokenizer.check_health() self.tokenizer.check_health()