A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 10, 17:45 UTC
[Cleanup] Remove deprecated fields from CachedRequestData class (#31734)
Signed-off-by: njhill <nickhill123@gmail.com>
Nick Hill committed
Jan 5, 2026 at 13:07 UTC
32f4e4db00f79188e2c4b5986391a38d0bac0558
2 files changed
+12
-30
tests/v1/core/test_scheduler.py
+12
-14
@@ -2284,7 +2284,6 @@ def test_priority_scheduling_preemption_and_resumption_when_out_of_kv(
2284
# 4th Schedule - this should trigger the resumption
2285
output = scheduler.schedule()
2286
scheduled_cached_reqs = output.scheduled_cached_reqs
2287
- resumed_from_preemption = scheduled_cached_reqs.resumed_from_preemption
2287
2288
assert len(output.scheduled_new_reqs) == 0
2289
assert scheduled_cached_reqs.num_reqs == 1
@@ -2292,14 +2291,14 @@ def test_priority_scheduling_preemption_and_resumption_when_out_of_kv(
2291
assert len(scheduler.running) == 1
2292
2293
# Preempted request resumed in scheduled_cached_reqs
2295
- assert len(resumed_from_preemption) == 1
2296
- assert len(scheduled_cached_reqs.resumed_req_token_ids) == 1
2297
- assert resumed_from_preemption[0]
2294
+ assert len(scheduled_cached_reqs.resumed_req_ids) == 1
2295
+ assert len(scheduled_cached_reqs.all_token_ids) == 1
2296
assert scheduled_cached_reqs.req_ids[0] == request_low.request_id
2299
- assert scheduled_cached_reqs.resumed_req_token_ids[0] is not None
2297
+ assert request_low.request_id in scheduled_cached_reqs.resumed_req_ids
2298
+ assert request_low.request_id in scheduled_cached_reqs.all_token_ids
2299
# Resumed tokens include 30 prompt tokens and 2 decoded tokens
2301
- assert len(scheduled_cached_reqs.resumed_req_token_ids[0]) == 32
2302
- assert scheduled_cached_reqs.resumed_req_token_ids[0][31] == 100
2300
+ assert len(scheduled_cached_reqs.all_token_ids[request_low.request_id]) == 32
2301
+ assert scheduled_cached_reqs.all_token_ids[request_low.request_id][31] == 100
2302
2303
2304
@pytest.mark.parametrize(
@@ -3122,7 +3121,6 @@ def test_priority_scheduling_ec_connector_preemption_and_resumption(
3121
# 4th Schedule - this should trigger req_low resumption from waiting
3122
output = scheduler.schedule()
3123
scheduled_cached_reqs = output.scheduled_cached_reqs
3125
- resumed_from_preemption = scheduled_cached_reqs.resumed_from_preemption
3124
3125
assert len(output.scheduled_new_reqs) == 0
3126
assert scheduled_cached_reqs.num_reqs == 1
@@ -3130,14 +3128,14 @@ def test_priority_scheduling_ec_connector_preemption_and_resumption(
3128
assert len(scheduler.running) == 1
3129
3130
# Preempted request resumed in scheduled_cached_reqs
3133
- assert len(resumed_from_preemption) == 1
3134
- assert len(scheduled_cached_reqs.resumed_req_token_ids) == 1
3135
- assert resumed_from_preemption[0]
3131
+ assert len(scheduled_cached_reqs.resumed_req_ids) == 1
3132
+ assert len(scheduled_cached_reqs.all_token_ids) == 1
3133
assert scheduled_cached_reqs.req_ids[0] == request_low.request_id
3137
- assert scheduled_cached_reqs.resumed_req_token_ids[0] is not None
3134
+ assert request_low.request_id in scheduled_cached_reqs.resumed_req_ids
3135
+ assert request_low.request_id in scheduled_cached_reqs.all_token_ids
3136
## Resumed tokens include 94 prompt tokens and 2 decoded tokens
3139
- assert len(scheduled_cached_reqs.resumed_req_token_ids[0]) == 96
3140
- assert scheduled_cached_reqs.resumed_req_token_ids[0][95] == 100
3137
+ assert len(scheduled_cached_reqs.all_token_ids[request_low.request_id]) == 96
3138
+ assert scheduled_cached_reqs.all_token_ids[request_low.request_id][95] == 100
3139
assert scheduler.running[0].request_id == request_low.request_id
3140
assert request_high.request_id in output.finished_req_ids
3141
vllm/v1/core/sched/output.py
-16
@@ -2,11 +2,8 @@
2
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
3
4
from dataclasses import dataclass
5
-from functools import cached_property
5
from typing import TYPE_CHECKING
6
8
-from typing_extensions import deprecated
9
-
7
from vllm._bc_linter import bc_linter_include
8
9
if TYPE_CHECKING:
@@ -129,19 +126,6 @@ class CachedRequestData:
126
def num_reqs(self) -> int:
127
return len(self.req_ids)
128
132
- @cached_property
133
- @deprecated("This will be removed in v0.14, use `resumed_req_ids` instead.")
134
- def resumed_from_preemption(self) -> list[bool]:
135
- return [req_id in self.resumed_req_ids for req_id in self.req_ids]
136
-
137
- @cached_property
138
- @deprecated("This will be removed in v0.14, use `all_token_ids` instead.")
139
- def resumed_req_token_ids(self) -> list[list[int] | None]:
140
- return [
141
- self.all_token_ids[req_id] if req_id in self.resumed_req_ids else None
142
- for req_id in self.req_ids
143
- ]
144
-
129
@classmethod
130
def make_empty(cls) -> "CachedRequestData":
131
return cls(