[KV Offload] Pass ScheduleEndContext to on_schedule_end hook (#46450)

Signed-off-by: Ronen Schaffer <ronen.schaffer@ibm.com>
Co-authored-by: Or Ozeri <oro@il.ibm.com>
This commit is contained in:
Ronen Schaffer
2026-06-30 17:07:12 +03:00
committed by GitHub
co-authored by Or Ozeri
parent 62c7d8009f
commit 0fc2512094
11 changed files with 42 additions and 14 deletions
@@ -13,7 +13,7 @@ from types import SimpleNamespace
import numpy as np
from vllm.v1.kv_offload.base import LookupResult, ReqContext
from vllm.v1.kv_offload.base import LookupResult, ReqContext, ScheduleEndContext
from vllm.v1.kv_offload.tiering.base import JobMetadata, JobResult
from vllm.v1.kv_offload.tiering.p2p import manager as manager_module
from vllm.v1.kv_offload.tiering.p2p.manager import (
@@ -1279,7 +1279,12 @@ class TestOnScheduleEnd:
mgr = _make_manager()
before_sessions = dict(mgr._sessions)
before_jobs = list(mgr._finished_jobs)
assert mgr.on_schedule_end() is None
assert (
mgr.on_schedule_end(
ScheduleEndContext(new_req_ids=[], preempted_req_ids=())
)
is None
)
assert mgr._sessions == before_sessions
assert mgr._finished_jobs == before_jobs
+2 -1
View File
@@ -22,6 +22,7 @@ from vllm.v1.kv_offload.base import (
LookupResult,
OffloadKey,
ReqContext,
ScheduleEndContext,
make_offload_key,
)
from vllm.v1.kv_offload.tiering.base import JobMetadata
@@ -107,7 +108,7 @@ def lookup_and_wait(
"""Perform a full async lookup cycle and return resolved results."""
for k in keys:
tier.lookup(k, ctx)
tier.on_schedule_end()
tier.on_schedule_end(ScheduleEndContext(new_req_ids=[], preempted_req_ids=()))
deadline = time.monotonic() + timeout
while time.monotonic() < deadline:
if not tier._lookup_manager._pending_results.empty():
+2 -1
View File
@@ -21,6 +21,7 @@ from vllm.v1.kv_offload.base import (
LookupResult,
OffloadKey,
ReqContext,
ScheduleEndContext,
make_offload_key,
)
from vllm.v1.kv_offload.tiering.base import JobMetadata, JobResult
@@ -222,7 +223,7 @@ def lookup_and_wait(
"""Perform a full async lookup cycle and return resolved results."""
for k in keys:
tier.lookup(k, ctx)
tier.on_schedule_end()
tier.on_schedule_end(ScheduleEndContext(new_req_ids=[], preempted_req_ids=()))
deadline = time.monotonic() + timeout
while time.monotonic() < deadline:
if not tier._lookup_manager._pending_results.empty():
@@ -27,6 +27,7 @@ from vllm.v1.kv_offload.base import (
OffloadPolicy,
ReqContext,
RequestOffloadingContext,
ScheduleEndContext,
make_offload_key,
)
from vllm.v1.kv_offload.tiering.base import (
@@ -233,7 +234,8 @@ class TestTieringOffloadingManager:
def _simulate_on_schedule_end(self):
"""Simulate end of scheduler step: lifecycle flush + drain events."""
self.manager.on_schedule_end()
ctx = ScheduleEndContext(new_req_ids=[], preempted_req_ids=())
self.manager.on_schedule_end(ctx)
list(self.manager.take_events())
def _start_request(self, req_context: ReqContext = _CTX):