[Core] Add a random suffix to frontend-provided request IDs (#27987)

Signed-off-by: Mark McLoughlin <markmc@redhat.com>
Signed-off-by: Nick Hill <nhill@redhat.com>
Co-authored-by: Nick Hill <nhill@redhat.com>
This commit is contained in:
Mark McLoughlin
2025-12-23 13:05:39 -08:00
committed by GitHub
co-authored by Nick Hill
parent 34916ae37f
commit f790068600
16 changed files with 328 additions and 154 deletions
+9 -4
View File
@@ -6,6 +6,7 @@ from typing import Optional, cast
from vllm.outputs import CompletionOutput
from vllm.sampling_params import RequestOutputKind, SamplingParams
from vllm.v1.engine import EngineCoreRequest
from vllm.v1.metrics.stats import IterationStats
@@ -17,6 +18,7 @@ class ParentRequest:
"""
request_id: str
external_req_id: str
sampling_params: SamplingParams
# To track the completion of child requests
@@ -31,8 +33,11 @@ class ParentRequest:
# To efficiently obtain child sampling params
cached_child_sampling_params: SamplingParams | None
def __init__(self, request_id: str, sampling_params: SamplingParams) -> None:
self.request_id = request_id
def __init__(self, request: EngineCoreRequest) -> None:
assert request.external_req_id is not None
sampling_params = request.params
self.request_id = request.request_id
self.external_req_id = request.external_req_id
self.sampling_params = sampling_params
self.child_requests = set()
@@ -96,7 +101,7 @@ class ParentRequest:
self,
child_request_id: str,
completion_output: CompletionOutput,
) -> tuple[str, list[CompletionOutput], bool]:
) -> tuple[list[CompletionOutput], bool]:
already_finished_and_returned: bool = False
if completion_output.finished():
if child_request_id in self.child_requests:
@@ -118,7 +123,7 @@ class ParentRequest:
outputs = [] if self.child_requests else self.output_aggregator
finished = not self.child_requests
return self.request_id, outputs, finished
return outputs, finished
def observe_num_generation_tokens(self, num_generation_tokens: int):
self.max_num_generation_tokens = max(