[Bugfix] Keep async video jobs queued until inference and abort on delete (#6759)

Signed-off-by: ZhengWG <zwg0606@gmail.com>
Signed-off-by: wuli666 <421774554@qq.com>
Co-authored-by: wuli666 <421774554@qq.com>
This commit is contained in:
Zheng Wengangandwuli666 authored and GitHub committed 2026-09-15 20:43:30 -07:00
1 parent 5dbae2b182
commit 275720b378
22 files changed
+549 -44

No files matched your search

+22 -2
View File
@@ -26,6 +26,8 @@ from vllm.v1.engine.exceptions import EngineDeadError
from vllm.v1.kv_cache_interface import KVCacheConfig
from vllm_omni.diffusion.data import (
DIFFUSION_REQUEST_LIFECYCLE_KEY,
DIFFUSION_REQUEST_STARTED,
DiffusionOutput,
DiffusionRequestAbortedError,
OmniDiffusionConfig,
@@ -55,7 +57,7 @@ from vllm_omni.diffusion.registry import (
)
from vllm_omni.diffusion.request import DUMMY_DIFFUSION_REQUEST_ID, OmniDiffusionRequest
from vllm_omni.diffusion.sched import BaseScheduler, RequestScheduler, StepScheduler
from vllm_omni.diffusion.sched.interface import DiffusionRequestStatus
from vllm_omni.diffusion.sched.interface import DiffusionRequestStatus, DiffusionSchedulerOutput
from vllm_omni.diffusion.worker.utils import BaseRunnerOutput, BatchRunnerOutput, RunnerOutput
from vllm_omni.errors import client_error_from_metadata, is_client_error_status
from vllm_omni.inputs.data import OmniDiffusionSamplingParams, OmniTextPrompt
@@ -533,7 +535,12 @@ class DiffusionEngine:
error_type=output.error_type,
)
raise RuntimeError(output.error)
logger.debug("Generation completed successfully.")
if output.request_started:
return format_empty_diffusion_outputs(
request,
finished=False,
custom_output={DIFFUSION_REQUEST_LIFECYCLE_KEY: DIFFUSION_REQUEST_STARTED},
)
if output.media is not None:
if output.output is not None:
@@ -599,6 +606,8 @@ class DiffusionEngine:
sched_output = self.scheduler.schedule()
self._scheduler_num_waiting_reqs = max(int(sched_output.num_waiting_reqs), 0)
self._emit_request_started_outputs(sched_output)
if sched_output.is_empty:
self._emit_finished_outputs(sched_output.finished_req_ids, None)
continue
@@ -629,6 +638,17 @@ class DiffusionEngine:
# Engine is stopping: fail any RPCs still queued so callers don't hang.
self._fail_pending_rpcs(RuntimeError("DiffusionEngine is shutting down."))
def _emit_request_started_outputs(self, sched_output: DiffusionSchedulerOutput) -> None:
"""Notify opted-in callers when requests leave the scheduler queue."""
for new_req in sched_output.scheduled_new_reqs:
request = getattr(new_req, "req", None)
sampling_params = getattr(request, "sampling_params", None)
if getattr(sampling_params, "emit_request_lifecycle", False):
self._put_output(
new_req.request_id,
DiffusionOutput(finished=False, request_started=True),
)
def _wait_for_admission_if_needed_locked(self) -> None:
"""Apply scheduler admission policy while holding the engine condition.