Files
FastDeploy/tests/cache_manager/v1/test_cache_utils.py
T
kevin 7707be8384 [Feature][KVCache] Implement Cache Manager V1 with GPU + CPU Cache Support (1/n) (#7097)
* [Feature][KVCache] Support cache manager v1 architecture

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Update cache manager and related modules

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* chore: update cache_manager and related modules

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* fix: add node to evictable set in complete_swap_to_device

When a node transitions from SWAP_TO_DEVICE to DEVICE via
complete_swap_to_device, it was not being added to the
_evictable_device set. This caused nodes with ref_count=0 to
become "orphaned" - not appearing in any evictable set despite
having cache_status=DEVICE.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* feat: update cache manager v1 and related modules

- Add new cache_manager.py with cache management functionality
- Add radix_tree.py for prefix caching
- Update block_pool.py and metadata.py
- Update request.py and resource_manager_v1.py for scheduling
- Update gpu_model_runner.py for GPU model execution

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* feat(cache): add cache controller v1 implementation

- Add CacheController class for cache management
- Update config.py with cache related configurations
- Refactor gpu_model_runner.py for improved cache handling

* feat(cache_manager): update cache manager v1

* fix(cache_manager): 修复 swap_cache H2D/D2H 方向的 block_ids 逻辑并清理 ForwardMeta

## Motivation

修复 swap_cache_optimized.cu 中 H2D 方向时 src/dst block_ids 使用错误的问题,
并清理 ForwardMeta 中已废弃的 cache_controller 字段。

## Modifications

- fix: swap_cache_optimized.cu 中根据 D2H 模板参数正确选取 src/dst block_ids,
  修复 H2D 方向 src/dst 倒置 bug(同时修复 SwapCachePerLayerImpl 和 SwapCacheAllLayersBatchImpl)
- refactor: cache_manager/v1/__init__.py 将 LayerSwapTimeoutError 导入从
  cache_controller 改为 cache_utils(正确来源)
- refactor: ForwardMeta 移除废弃的 cache_controller 字段
- refactor: gpu_model_runner.py 移除对应的 cache_controller 赋值语句
- test: 新增 tests/cache_manager/v1/test_swap_cache_ops.py 单元测试

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(cache_manager): refactor cache manager v1 and optimize swap ops

## Motivation

对 cache manager v1 进行重构和优化,精简代码结构,提升可维护性。

## Modifications

- 重构 transfer_manager.py,大幅精简代码逻辑
- 优化 swap_cache_optimized.cu GPU 算子实现
- 调整 cache_manager.py、cache_controller.py 逻辑,修复 free_device_blocks 方法缺失问题
- 更新 block_pool.py、cache_utils.py、metadata.py、radix_tree.py
- 精简 gpu_model_runner.py、forward_meta.py、attention.py 中相关调用
- 更新对应单元测试(test_cache_controller、test_swap_cache_ops、test_transfer_manager)
- 调整 config.py 中相关配置项

* [KVCache][MTP] 支持 cache_manager_v1 下的 MTP KV Cache 初始化及多模态 hash

## Motivation

在 enable_cache_manager_v1 路径下,MTP(speculative decode)的 KV Cache 需要由
CacheController 统一管理,以复用 swap/transfer 能力,同时修复多模态场景下 block
hash 未携带 multimodal extra_keys 的问题。

## Modifications

- `cache_controller.py`
  - 新增 `initialize_mtp_kv_cache`:通过 CacheController 初始化 MTP KV Cache,
    并将其注册到 cache_kvs_map,使 transfer_manager 自动覆盖 MTP 层
  - `initialize_host_cache` 中的 num_layers 改为包含 MTP 额外 cache 层数,保证
    Host Cache 也为 MTP 分配足够空间
  - `_free_gpu_cache` 改名为 `free_gpu_cache`(对外可调用)

- `cache_utils.py`
  - 新增 `get_block_hash_extra_keys`:提取单个 block 内的多模态 hash 信息,
    对齐 PrefixCacheManager 的 multimodal extra_keys 逻辑
  - `get_request_block_hasher` 中在 hash_block_tokens 时携带 extra_keys,
    修复多模态场景 prefix cache 命中率不准的问题

- `spec_decode/mtp.py`
  - `update_mtp_block_num` 新增 `skip_cache_init` 参数,避免 v1 cache manager
    路径下重复初始化 MTP KV Cache

- `gpu_model_runner.py`
  - `initialize_kv_cache(v1)` 路径:在主模型 cache 初始化后,调用
    `cache_controller.initialize_mtp_kv_cache` 完成 MTP cache 创建
  - `clear_cache` / `wakeup` / `reset` 等路径:respect `enable_cache_manager_v1`
    标志,跳过重复的 proposer.initialize_kv_cache 调用

## Usage or Command

```bash
# 启动支持 MTP + cache_manager_v1 的推理服务(示例)
bash run.sh
```

* fix(cache_manager): multi-GPU fix, mm hash boundary fix, and remove batch ops

1. Fix CuPy stream/event creation for multi-GPU: wrap all stream operations
   with cp.cuda.Device(device_id) context to ensure streams/events are bound
   to the correct device, preventing cross-device errors in multi-GPU setups.

2. Remove cudaSetDevice from SwapCacheAllLayers (handled by cupy context now).

3. Remove swap_cache_all_layers_batch op: simplified the implementation by
   removing the batch upload variant; all-layer transfers now use the standard
   swap_cache_all_layers with cupy device context.

4. Fix mm hash boundary comparison in get_block_hash_extra_keys: change
   strict less-than (<) to less-than-or-equal (<=) so that multimodal items
   ending exactly at block start are correctly excluded.

5. Extract config fields to KVCacheBase: model_config, cache_config,
   quant_config, parallel_config are now set in the base class __init__ to
   avoid duplication in CacheController and CacheManager subclasses.

6. Translate metadata.py docstrings from Chinese to English for broader
   contributor accessibility.

7. Add test_cache_utils.py: comprehensive unit tests for
   get_block_hash_extra_keys covering all boundary and overlap scenarios.

8. Expand test suite: test_request.py cache fields tests, test_radix_tree.py
   backup candidate tests, test_transfer_manager.py and test_cache_manager.py
   multi-GPU and concurrent operation tests.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* [BugFix][KVCache] fix List import and move write_policy normalization to CacheManager

## Motivation

修复两处问题:
1. `fastdeploy/engine/request.py` 中 `List` 未导入导致 pre-commit F821 报错
2. `write_policy` 归一化逻辑(`write_through` → `write_through_selective`)不应放在 `FDConfig`,移至 `CacheManager.__init__` 中,使其只影响 Cache Manager V1 的内部逻辑

## Modifications

- `fastdeploy/engine/request.py`: 在 `typing` 导入中补充 `List`,删除重复的 `CacheSwapMetadata` TYPE_CHECKING 导入,修复 F821/F811
- `fastdeploy/config.py`: 删除 `write_policy` 归一化逻辑
- `fastdeploy/cache_manager/v1/cache_manager.py`: 将归一化逻辑移入 `CacheManager.__init__`

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* [BugFix][KVCache] fix pre-commit code style issues

## Motivation

修复 CI pre-commit 代码风格检查失败问题。

## Modifications

- `fastdeploy/engine/common_engine.py`: black 格式化
- `fastdeploy/worker/worker_process.py`: black 格式化 + isort 修复
- `fastdeploy/cache_manager/v1/storage/__init__.py`: isort 修复
- `fastdeploy/worker/gpu_worker.py`: isort 修复

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* [Feature][KVCache] update cache_manager_v1 modules

## Motivation

更新 Cache Manager V1 相关模块,完善版权信息、改进模块结构与可维护性。

## Modifications

- `fastdeploy/cache_manager/v1/` 系列模块:补充版权 header,优化代码结构
- `fastdeploy/config.py`:配置项更新
- `fastdeploy/engine/sched/resource_manager_v1.py`:调度相关更新

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* [Feature][KVCache] add BatchRequest.from_tasks and refactor worker task parsing

## Motivation

将 worker_process 中重复的 task 解析逻辑收敛到 BatchRequest,减少代码冗余,提升可维护性。

## Modifications

- `fastdeploy/engine/request.py`:新增 `BatchRequest.from_tasks()` 类方法,统一将 task_queue 任务分类为推理请求和控制请求
- `fastdeploy/worker/worker_process.py`:使用 `BatchRequest.from_tasks()` 替代内联解析逻辑,并修复重复的 control_reqs 处理块

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* [Feature][KVCache] add NUMA affinity for host cache and skip swap cache tests

## Motivation

优化 Host cache 内存分配的 NUMA 亲和性,减少跨 NUMA 访问延迟;
同时跳过 swap cache ops 测试(当前环境不支持)。

## Modifications

- `fastdeploy/cache_manager/v1/cache_controller.py`:
  - 新增 `_get_numa_node_for_gpu()` 方法,通过 nvidia-smi 或 sysfs 获取 GPU 对应的 NUMA 节点
  - 新增 `_bind_to_closest_numa_node()` 方法,绑定当前线程到 GPU 最近的 NUMA 节点
  - 在 `initialize_host_cache()` 中调用 NUMA 绑定,优化 H2D 传输性能
- `tests/cache_manager/v1/test_swap_cache_ops.py`:跳过所有测试类(`TestSwapCacheAllLayersCorrectness`、`TestSwapCacheAllLayersPerformance`、`TestSwapCacheRandomBlockIndices`)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* [BugFix][KVCache] fix unittest failures for cache_manager_v1

三个单测因接口变更或 Mock 方式问题导致失败,需修复。

- tests/distributed/chunked_moe.py:`setup_model_runner` 使用 `__new__` 跳过 `__init__`,补加 `enable_cache_manager_v1 = False`,修复 `AttributeError`
- tests/engine/test_resource_manager.py:`PrefixCacheManager` 为局部导入,`patch` 路径改为定义位置 `fastdeploy.cache_manager.prefix_cache_manager.PrefixCacheManager`
- tests/v1/test_resource_manager_v1.py:`_trigger_preempt` 第四参数已由 `list` 改为 `BatchRequest`,更新测试传参和断言

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* [BugFix][KVCache] remove debug logging code

## Modifications

- fastdeploy/engine/request.py:删除调试用 logger 及 prompt_hashes 中的 debug 日志
- fastdeploy/worker/worker_process.py:删除 __main__ 中的调试 import 和 print 语句

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* [BugFix][KVCache] fix cupy device id caching and pickle for _match_result

## Motivation

修复两个 bug:
1. `transfer_manager.py` 中每次调用 `cp.cuda.runtime.getDevice()` 存在隐患,应在初始化时缓存为实例变量,保证后续操作使用一致的设备 ID。
2. `request.py` 的 `__getstate__` 未跳过 `_match_result`,该字段包含 BlockNode 树的父子循环引用,pickle 时会触发 `RecursionError`;同时补充 `__setstate__` 确保 unpickle 后字段恢复为安全默认值。

## Modifications

- `transfer_manager.py`:初始化时调用 `cp.cuda.runtime.getDevice()` 并缓存到 `self._cupy_device_id`,后续 `with cp.cuda.Device(...)` 和日志均使用该缓存值。
- `request.py`:
  - `__getstate__` 中将 `_match_result` 加入跳过集合 `_SKIP_KEYS`,避免循环引用导致 pickle 失败。
  - 新增 `__setstate__`,unpickle 后将 `_block_hasher` 和 `_match_result` 恢复为 `None`。

## Usage or Command

* fix(test): fix unit test errors for _trigger_preempt and wakeup with MTP

- Use BatchRequest instead of list in test_trigger_preempt_records_tasks
- Add missing enable_cache_manager_v1 attr in TestSleepWakeupBehavior._make_runner

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* [BugFix][KVCache] fix gpu_free_block_list returning wrong block IDs

## Motivation

`gpu_free_block_list` 的兼容 property 中误用了 `list(range(N))`,
将 `available_blocks()` 的返回值当作整数传给 `range()`,
导致返回 `[0, 1, ..., N-1]` 的假列表,而非真实的空闲 block ID。

## Modifications

- `cache_manager/v1/cache_manager.py`:将 `list(range(self._device_pool.available_blocks()))` 改为 `list(self._device_pool.available_blocks())`

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* [BugFix][KVCache] 修复 gpu_free_block_list 返回 int 导致 TypeError

## Motivation

gpu_free_block_list 属性中调用 BlockPool.available_blocks(),
该方法返回 int(空闲块数量),用 list() 包装 int 会触发
TypeError: 'int' object is not iterable。

## Modifications

将 list(self._device_pool.available_blocks()) 改为
list(self._device_pool._free_blocks),直接返回空闲块索引列表。

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* [KVCache][CacheManager] 适配 V1 CacheManager 的 pause/sleep/free_cache 操作

## Motivation

V1 CacheManager 引入了新的 reset_cache() 接口,pause 和 sleep 操作需要适配,
同时 free_cache 需要支持可选的 clear_storage 参数。

## Modifications

- cache_controller.py: free_cache 新增 clear_storage 参数(默认 False),
  仅当 clear_storage=True 时才调用 _clear_storage(),避免不必要的 storage 清空
- common_engine.py: pause 和 sleep 操作中,当 ENABLE_V1_KVCACHE_MANAGER 时
  使用 cache_manager.reset_cache() 替代旧的 reset() 和 pause_transfer 逻辑
- gpu_model_runner.py: sleep 时仅在非 V1 cache manager 下执行 MTP cache 清除

## Usage or Command

# 启动服务(V1 CacheManager)
python -m fastdeploy.entrypoints.openai.api_server \
  --enable-v1-kvcache-manager \
  ...

* [BugFix][KVCache] fix missing enable_cache_manager_v1 in test mocks and remove unused select_blocks_for_backup

- Remove unused `select_blocks_for_backup` method from radix_tree.py
- Fix `match_prefix` default param `skip_storage=True` and log order in cache_manager.py
- Sync test_gpu_model_runner.py with upstream/develop (add TestInsertTasksV1SplitwiseSuffix)
- Add `enable_cache_manager_v1=False` to all mock runners to fix AttributeError in CI

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* [BugFix][KVCache] simplify _free_blocks in ResourceManagerV1 for non-v1 path

Remove redundant prefix_caching branch in else path; always call
recycle_gpu_blocks with full block_tables for non-cache-manager-v1 case.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* [KVCache][Optimization][BugFix] fix and optimize block_pool, cache_manager, transfer_manager, request

## Motivation

修复 cache_manager v1 中若干代码质量问题,提升性能并消除潜在的类型不一致 Bug。

## Modifications

1. **block_pool.py**:`BlockPool.allocate` 将逐个 pop 循环替换为切片 + 批量 set.update,消除 Python 循环开销,O(n) → O(k)(C 层批量操作)
2. **cache_manager.py**:`match_prefix` 在 prefix caching 关闭时提前 return 前写入空 `MatchResult()`,避免调用方解引用 `_match_result=None` 崩溃
3. **transfer_manager.py**:`_build_device_layer_indices` 在 `_cache_kvs_map` 为空时也重置四个层索引列表,防止残留旧 tensor 被 swap 算子使用
4. **request.py**:`BatchRequest.append_swap_metadata` / `append_evict_metadata` 构造 `CacheSwapMetadata` 时将 `src_type`/`dst_type` 从字符串改为 `CacheLevel` 枚举,与字段类型声明一致;补充 `CacheLevel` 导入;`match_result` 属性返回类型标注修正为 `Optional[MatchResult]`
5. **resource_manager_v1.py**:`_allocate_gpu_blocks` 日志从 `INFO` 降级为 `DEBUG`,消除高频调度路径的日志噪音
6. **tests/engine/test_request.py**:同步更新 `src_type`/`dst_type` 断言为 `CacheLevel` 枚举值,补充 `CacheLevel` 导入

## Usage or Command

单元测试:
```bash
source .venv/py310/bin/activate
cd baidu/FastDeploy
python -m pytest tests/cache_manager/v1/test_cache_manager.py -v
python -m pytest tests/cache_manager/v1/test_transfer_manager.py -v
python -m pytest tests/engine/test_request.py -v
```

* [BugFix][KVCache] Fix BlockPool.allocate returns all blocks when num_blocks=0

## Motivation

当 `allocate(num_blocks=0)` 被调用时,Python 负索引陷阱导致严重错误:
`-0 == 0`,所以 `self._free_blocks[-0:]` 等价于 `self._free_blocks[0:]`,
会返回并清空整个空闲块列表,而非返回空列表。

## Modifications

在 `BlockPool.allocate` 中增加对 `num_blocks == 0` 的提前判断,直接返回 `[]`,
避免触发 Python 负索引陷阱。

## Usage or Command

```bash
# 运行相关单元测试验证修复
python -m pytest tests/cache_manager/v1/test_cache_manager.py -vv -s
```

* [KVCache][Test] add unit tests for cache_manager v1 modules

## Motivation

补全 cache_manager/v1 各模块的单测覆盖,确保核心方法有完整的测试保障。

## Modifications

新增/补充以下测试文件,全部 326 个用例通过:

- tests/cache_manager/v1/test_block_pool.py(新建)
  覆盖 BlockPool.get_metadata/set_metadata/resize、DeviceBlockPool/HostBlockPool
- tests/cache_manager/v1/test_metadata.py(新建)
  覆盖 BlockNode、RadixTreeStats、MatchResult、CacheSwapMetadata、AsyncTaskHandler
- tests/cache_manager/v1/test_cache_utils.py(补充)
  新增 hash_block_tokens、get_request_block_hasher、LayerDoneCounter 时间追踪及内部辅助方法
- tests/cache_manager/v1/test_radix_tree.py(补充)
  新增 TestCompleteSwapToDevice 专项测试类(6 个用例)
- tests/cache_manager/v1/test_cache_manager.py(补充)
  新增 offload_to_host、load_from_host、pending backup 系列、prepare_prefetch_metadata
- tests/cache_manager/v1/test_transfer_manager.py(补充)
  新增 _swap_single_layer 校验路径、sync_input/output_stream、record_input_stream_event

## Usage or Command

```bash
# 运行所有新增单测
source .venv/py310/bin/activate
python -m pytest tests/cache_manager/v1/test_block_pool.py \
  tests/cache_manager/v1/test_metadata.py \
  tests/cache_manager/v1/test_cache_utils.py \
  tests/cache_manager/v1/test_radix_tree.py \
  tests/cache_manager/v1/test_cache_manager.py \
  tests/cache_manager/v1/test_transfer_manager.py -v
# 期望结果:326 passed
```

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: Jiang-Jia-Jun <163579578+Jiang-Jia-Jun@users.noreply.github.com>
2026-04-21 14:39:00 +08:00

682 lines
28 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
"""
# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
Unit tests for get_block_hash_extra_keys in
fastdeploy/cache_manager/v1/cache_utils.py.
Tests mirror the style used in
tests/cache_manager/test_prefix_cache_manager.py and cover:
- Early return paths (None input, missing keys, empty mm_positions)
- Fast-exit path (last item ends before block start)
- Image entirely before the block (skip via continue)
- Image entirely after the block (stop via return)
- Image fully contained in block
- Image spanning the right block boundary
- Image spanning the entire block (starts before, ends after)
- Multiple images: only overlapping ones included
- Sequential multi-block scan using the returned mm_idx
- Single-token block and single-token image edge cases
"""
import time
import unittest
from types import SimpleNamespace
from fastdeploy.cache_manager.v1.cache_utils import get_block_hash_extra_keys
def _req(mm_positions, mm_hashes):
"""Build a minimal request-like object with multimodal_inputs."""
return SimpleNamespace(
multimodal_inputs={
"mm_positions": [SimpleNamespace(offset=o, length=l) for o, l in mm_positions],
"mm_hashes": list(mm_hashes),
}
)
class TestGetBlockHashExtraKeysEarlyReturn(unittest.TestCase):
"""Tests for the guard / early-return paths at the top of the function."""
def test_multimodal_inputs_none(self):
"""multimodal_inputs=None → (mm_idx, []) unchanged."""
req = SimpleNamespace(multimodal_inputs=None)
mm_idx, keys = get_block_hash_extra_keys(req, start_idx=0, end_idx=4, mm_idx=0)
self.assertEqual((mm_idx, keys), (0, []))
def test_multimodal_inputs_attribute_missing(self):
"""Object without multimodal_inputs attribute → treated as None."""
req = SimpleNamespace()
mm_idx, keys = get_block_hash_extra_keys(req, start_idx=0, end_idx=4, mm_idx=0)
self.assertEqual((mm_idx, keys), (0, []))
def test_mm_positions_key_missing(self):
"""mm_positions key absent → early return."""
req = SimpleNamespace(multimodal_inputs={"mm_hashes": ["h"]})
mm_idx, keys = get_block_hash_extra_keys(req, start_idx=0, end_idx=4, mm_idx=0)
self.assertEqual((mm_idx, keys), (0, []))
def test_mm_hashes_key_missing(self):
"""mm_hashes key absent → early return."""
req = SimpleNamespace(multimodal_inputs={"mm_positions": [SimpleNamespace(offset=0, length=2)]})
mm_idx, keys = get_block_hash_extra_keys(req, start_idx=0, end_idx=4, mm_idx=0)
self.assertEqual((mm_idx, keys), (0, []))
def test_mm_positions_empty_list(self):
"""mm_positions=[] → early return."""
req = SimpleNamespace(multimodal_inputs={"mm_positions": [], "mm_hashes": []})
mm_idx, keys = get_block_hash_extra_keys(req, start_idx=0, end_idx=4, mm_idx=0)
self.assertEqual((mm_idx, keys), (0, []))
def test_fast_exit_last_item_ends_exactly_at_block_start(self):
"""
Fast-exit: last item offset+length == start_idx
(item ends exactly where block begins → no overlap).
"""
# img [0,4), block [4,8) → 4 <= 4 → fast exit
req = _req([(0, 4)], ["h"])
mm_idx, keys = get_block_hash_extra_keys(req, start_idx=4, end_idx=8, mm_idx=0)
self.assertEqual((mm_idx, keys), (0, []))
def test_fast_exit_last_item_ends_before_block_start(self):
"""Fast-exit: all items end strictly before block start."""
# img [0,3), block [4,8)
req = _req([(0, 3)], ["h"])
mm_idx, keys = get_block_hash_extra_keys(req, start_idx=4, end_idx=8, mm_idx=0)
self.assertEqual((mm_idx, keys), (0, []))
def test_fast_exit_preserves_mm_idx(self):
"""Fast-exit returns the original mm_idx unchanged."""
req = _req([(0, 2)], ["h"])
mm_idx, keys = get_block_hash_extra_keys(req, start_idx=5, end_idx=9, mm_idx=0)
self.assertEqual(mm_idx, 0)
self.assertEqual(keys, [])
class TestGetBlockHashExtraKeysSingleImage(unittest.TestCase):
"""Tests with exactly one multimodal item and one block."""
# ------------------------------------------------------------------
# Item entirely before block → skip (continue), reaches end of loop
# ------------------------------------------------------------------
def test_item_ends_before_block_start(self):
"""img [0,2) is entirely before block [3,7)."""
req = _req([(0, 2)], ["h"])
mm_idx, keys = get_block_hash_extra_keys(req, start_idx=3, end_idx=7, mm_idx=0)
self.assertEqual((mm_idx, keys), (0, []))
def test_item_ends_exactly_at_block_start(self):
"""img [0,3) ends exactly at block start 3 → 3<=3 → skip."""
req = _req([(0, 3)], ["h"])
mm_idx, keys = get_block_hash_extra_keys(req, start_idx=3, end_idx=7, mm_idx=0)
self.assertEqual((mm_idx, keys), (0, []))
# ------------------------------------------------------------------
# Item entirely after block → stop (return img_idx, [])
# ------------------------------------------------------------------
def test_item_starts_at_block_end(self):
"""img [8,10) starts exactly at block end 8 → offset>=end_idx → stop."""
req = _req([(8, 2)], ["h"])
mm_idx, keys = get_block_hash_extra_keys(req, start_idx=4, end_idx=8, mm_idx=0)
self.assertEqual((mm_idx, keys), (0, []))
def test_item_starts_after_block_end(self):
"""img [10,3) starts strictly after block [4,8)."""
req = _req([(10, 3)], ["h"])
mm_idx, keys = get_block_hash_extra_keys(req, start_idx=4, end_idx=8, mm_idx=0)
self.assertEqual((mm_idx, keys), (0, []))
# ------------------------------------------------------------------
# Item spans beyond block right boundary
# ------------------------------------------------------------------
def test_item_spans_right_boundary(self):
"""img [6,4) → [6,10) spans block [4,8) right boundary."""
req = _req([(6, 4)], ["hash-cross"])
mm_idx, keys = get_block_hash_extra_keys(req, start_idx=4, end_idx=8, mm_idx=0)
self.assertEqual((mm_idx, keys), (0, ["hash-cross"]))
def test_item_spans_entire_block(self):
"""img [3,6) → [3,9) wraps the whole block [4,8)."""
req = _req([(3, 6)], ["hash-span"])
mm_idx, keys = get_block_hash_extra_keys(req, start_idx=4, end_idx=8, mm_idx=0)
self.assertEqual((mm_idx, keys), (0, ["hash-span"]))
def test_item_starts_at_block_start_spans_right(self):
"""img starts at block start, extends past block end."""
req = _req([(4, 6)], ["h"])
mm_idx, keys = get_block_hash_extra_keys(req, start_idx=4, end_idx=8, mm_idx=0)
self.assertEqual((mm_idx, keys), (0, ["h"]))
# ------------------------------------------------------------------
# Item fully contained within block
# ------------------------------------------------------------------
def test_item_fully_inside_block(self):
"""img [2,2) → [2,4) fully inside block [0,8)."""
req = _req([(2, 2)], ["hash-inside"])
mm_idx, keys = get_block_hash_extra_keys(req, start_idx=0, end_idx=8, mm_idx=0)
self.assertIn("hash-inside", keys)
def test_item_fills_block_exactly(self):
"""img occupies exactly the block [4,8)."""
req = _req([(4, 4)], ["h-exact"])
mm_idx, keys = get_block_hash_extra_keys(req, start_idx=4, end_idx=8, mm_idx=0)
self.assertEqual((mm_idx, keys), (0, ["h-exact"]))
# ------------------------------------------------------------------
# Single-token edge cases
# ------------------------------------------------------------------
def test_single_token_block_single_token_item_inside(self):
"""Block [5,6), img [5,1) → item fills the single-token block."""
req = _req([(5, 1)], ["h1"])
mm_idx, keys = get_block_hash_extra_keys(req, start_idx=5, end_idx=6, mm_idx=0)
self.assertIn("h1", keys)
def test_single_token_block_item_starts_after(self):
"""Block [5,6), img [6,1) → starts at block end, not included."""
req = _req([(6, 1)], ["h1"])
mm_idx, keys = get_block_hash_extra_keys(req, start_idx=5, end_idx=6, mm_idx=0)
self.assertEqual(keys, [])
class TestGetBlockHashExtraKeysMultipleImages(unittest.TestCase):
"""Tests with multiple multimodal items."""
def test_only_overlapping_items_included(self):
"""
3 images; only the one overlapping the block should be in hash_keys.
img0: [0,2) → before block [4,8)
img1: [5,2) → inside block [4,8)
img2: [9,2) → after block [4,8)
"""
req = _req([(0, 2), (5, 2), (9, 2)], ["h0", "h1", "h2"])
mm_idx, keys = get_block_hash_extra_keys(req, start_idx=4, end_idx=8, mm_idx=0)
self.assertNotIn("h0", keys)
self.assertIn("h1", keys)
self.assertNotIn("h2", keys)
def test_multiple_items_all_inside_block(self):
"""Two images both inside the block → both hashes collected."""
req = _req([(1, 2), (4, 2)], ["hA", "hB"])
mm_idx, keys = get_block_hash_extra_keys(req, start_idx=0, end_idx=8, mm_idx=0)
self.assertEqual(keys, ["hA", "hB"])
def test_no_item_overlaps_block(self):
"""All images are before the block → empty keys."""
req = _req([(0, 2), (2, 1)], ["h0", "h1"])
mm_idx, keys = get_block_hash_extra_keys(req, start_idx=5, end_idx=9, mm_idx=0)
self.assertEqual(keys, [])
def test_mm_idx_skips_already_processed_items(self):
"""
When mm_idx=1, item at index 0 is not scanned at all.
"""
req = _req([(0, 2), (5, 2)], ["h0", "h1"])
# Start scanning from mm_idx=1, so h0 must never appear
mm_idx, keys = get_block_hash_extra_keys(req, start_idx=4, end_idx=8, mm_idx=1)
self.assertNotIn("h0", keys)
self.assertIn("h1", keys)
def test_returned_mm_idx_points_to_spanning_item(self):
"""
When an item spans the block right boundary, returned mm_idx points
to that item (so the next block can re-examine it).
img0 [2,7): offset+length=9 > end_idx=8 → spans right boundary
→ include hA, return img_idx=0 immediately (img1 never reached).
"""
# img0 offset=2, length=7 → end=9 > end_idx=8 → spans right boundary
req = _req([(2, 7), (10, 2)], ["hA", "hB"])
mm_idx, keys = get_block_hash_extra_keys(req, start_idx=4, end_idx=8, mm_idx=0)
self.assertEqual(mm_idx, 0) # still points to img0 (not fully consumed)
self.assertIn("hA", keys)
self.assertNotIn("hB", keys)
def test_returned_mm_idx_stops_at_after_item(self):
"""
When an item starts after the block, returned mm_idx points to it
so the next block can start scanning from there.
"""
req = _req([(2, 2), (9, 1)], ["hA", "hB"])
# img1 [9,10) is after block [4,8)
mm_idx, keys = get_block_hash_extra_keys(req, start_idx=4, end_idx=8, mm_idx=1)
self.assertEqual(mm_idx, 1)
self.assertEqual(keys, [])
class TestGetBlockHashExtraKeysSequentialScan(unittest.TestCase):
"""
Simulates a full multi-block scan, reusing the returned mm_idx as the
next call's mm_idx mirroring the exact pattern used in
test_prefix_cache_manager.py.
Data layout (block_size=4):
tokens: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
img0: [=====] [2,5) hash-0
img1: [========] [8,12) hash-1
img2: [==] [14,16) hash-2
blocks: [0,4) [4,8) [8,12) [12,16)
"""
def setUp(self):
self.req = SimpleNamespace(
multimodal_inputs={
"mm_positions": [
SimpleNamespace(offset=2, length=3), # [2,5)
SimpleNamespace(offset=8, length=4), # [8,12)
SimpleNamespace(offset=14, length=2), # [14,16)
],
"mm_hashes": ["hash-0", "hash-1", "hash-2"],
}
)
def test_block_0_4(self):
"""Block [0,4): img0 [2,5) spans right boundary → hash-0, mm_idx=0."""
mm_idx, keys = get_block_hash_extra_keys(self.req, start_idx=0, end_idx=4, mm_idx=0)
self.assertEqual((mm_idx, keys), (0, ["hash-0"]))
def test_block_4_8_using_returned_mm_idx(self):
"""Block [4,8): carry mm_idx=0 from previous call → img0 tail, then img1 stops."""
mm_idx, keys = get_block_hash_extra_keys(self.req, start_idx=4, end_idx=8, mm_idx=0)
self.assertEqual((mm_idx, keys), (1, ["hash-0"]))
def test_block_8_12_using_returned_mm_idx(self):
"""Block [8,12): img1 [8,12) exactly fills block → hash-1, mm_idx advances."""
mm_idx, keys = get_block_hash_extra_keys(self.req, start_idx=8, end_idx=12, mm_idx=1)
self.assertEqual((mm_idx, keys), (2, ["hash-1"]))
def test_block_12_16_using_returned_mm_idx(self):
"""Block [12,16): img2 [14,16) fully inside → hash-2."""
mm_idx, keys = get_block_hash_extra_keys(self.req, start_idx=12, end_idx=16, mm_idx=2)
self.assertEqual((mm_idx, keys), (2, ["hash-2"]))
def test_full_sequential_scan(self):
"""Run all four blocks sequentially, feeding mm_idx forward."""
mm_idx = 0
expected = [
((0, 4), (0, ["hash-0"])),
((4, 8), (1, ["hash-0"])),
((8, 12), (2, ["hash-1"])),
((12, 16), (2, ["hash-2"])),
]
for (s, e), (exp_mm_idx, exp_keys) in expected:
mm_idx, keys = get_block_hash_extra_keys(self.req, start_idx=s, end_idx=e, mm_idx=mm_idx)
self.assertEqual((mm_idx, keys), (exp_mm_idx, exp_keys), msg=f"block [{s},{e})")
class TestGetBlockHashExtraKeysBoundaryPrecision(unittest.TestCase):
"""Exact boundary conditions: <= vs < matters at edges."""
def test_item_end_equals_start_idx_not_included(self):
"""
offset+length == start_idx → item ends exactly where block starts
→ condition `<= start_idx` is True → skip (not included).
"""
# img [0,4), block [4,8): 0+4=4 == start_idx=4 → skip
req = SimpleNamespace(
multimodal_inputs={
"mm_positions": [SimpleNamespace(offset=0, length=4), SimpleNamespace(offset=10, length=1)],
"mm_hashes": ["h-boundary", "h-other"],
}
)
_, keys = get_block_hash_extra_keys(req, start_idx=4, end_idx=8, mm_idx=0)
self.assertNotIn("h-boundary", keys)
def test_item_offset_equals_end_idx_not_included(self):
"""
offset == end_idx → item starts exactly where block ends
→ condition `>= end_idx` is True → stop (not included).
"""
# img [8,2), block [4,8): offset=8 == end_idx=8 → stop
req = SimpleNamespace(
multimodal_inputs={
"mm_positions": [SimpleNamespace(offset=8, length=2)],
"mm_hashes": ["h-boundary"],
}
)
_, keys = get_block_hash_extra_keys(req, start_idx=4, end_idx=8, mm_idx=0)
self.assertNotIn("h-boundary", keys)
def test_item_end_one_past_block_end_included(self):
"""
offset+length == end_idx+1 → item end is 1 past block end
→ condition `> end_idx` is True → included and mm_idx stays.
"""
# img [6,3) → [6,9), block [4,8): 6+3=9 > 8 → spans right boundary
req = SimpleNamespace(
multimodal_inputs={
"mm_positions": [SimpleNamespace(offset=6, length=3)],
"mm_hashes": ["h-one-past"],
}
)
mm_idx, keys = get_block_hash_extra_keys(req, start_idx=4, end_idx=8, mm_idx=0)
self.assertIn("h-one-past", keys)
self.assertEqual(mm_idx, 0)
def test_item_end_equals_end_idx_fully_contained(self):
"""
offset+length == end_idx → item ends exactly at block end
→ condition `> end_idx` is False → fully contained, included.
"""
# img [4,4) → [4,8), block [4,8): 4+4=8 == end_idx=8 → contained
req = SimpleNamespace(
multimodal_inputs={
"mm_positions": [SimpleNamespace(offset=4, length=4)],
"mm_hashes": ["h-exact-end"],
}
)
_, keys = get_block_hash_extra_keys(req, start_idx=4, end_idx=8, mm_idx=0)
self.assertIn("h-exact-end", keys)
# ---------------------------------------------------------------------------
# hash_block_tokens
# ---------------------------------------------------------------------------
class TestHashBlockTokens(unittest.TestCase):
"""Direct tests for hash_block_tokens."""
def setUp(self):
from fastdeploy.cache_manager.v1.cache_utils import hash_block_tokens
self.hash_block_tokens = hash_block_tokens
def test_returns_hex_string(self):
h = self.hash_block_tokens([1, 2, 3])
self.assertIsInstance(h, str)
self.assertEqual(len(h), 64) # SHA256 hex digest length
def test_same_input_same_hash(self):
h1 = self.hash_block_tokens([1, 2, 3])
h2 = self.hash_block_tokens([1, 2, 3])
self.assertEqual(h1, h2)
def test_different_tokens_different_hash(self):
h1 = self.hash_block_tokens([1, 2, 3])
h2 = self.hash_block_tokens([1, 2, 4])
self.assertNotEqual(h1, h2)
def test_parent_hash_none_and_empty_string_differ(self):
"""None and '' parent hash should both work; chaining is the key."""
h_none = self.hash_block_tokens([1, 2], parent_block_hash=None)
h_empty = self.hash_block_tokens([1, 2], parent_block_hash="")
# Both produce valid hashes; they may or may not be equal depending on
# implementation, but must be deterministic.
self.assertEqual(h_none, self.hash_block_tokens([1, 2], parent_block_hash=None))
self.assertEqual(h_empty, self.hash_block_tokens([1, 2], parent_block_hash=""))
def test_chained_hash_differs_from_unchained(self):
parent = self.hash_block_tokens([0])
h_chained = self.hash_block_tokens([1, 2], parent_block_hash=parent)
h_no_parent = self.hash_block_tokens([1, 2])
self.assertNotEqual(h_chained, h_no_parent)
def test_extra_keys_affect_hash(self):
h1 = self.hash_block_tokens([1, 2], extra_keys=None)
h2 = self.hash_block_tokens([1, 2], extra_keys=("image_hash",))
self.assertNotEqual(h1, h2)
def test_empty_token_ids(self):
h = self.hash_block_tokens([])
self.assertIsInstance(h, str)
self.assertEqual(len(h), 64)
# ---------------------------------------------------------------------------
# get_request_block_hasher
# ---------------------------------------------------------------------------
class TestGetRequestBlockHasher(unittest.TestCase):
"""Tests for the factory function get_request_block_hasher."""
def setUp(self):
from fastdeploy.cache_manager.v1.cache_utils import get_request_block_hasher
self.block_size = 4
self.hasher = get_request_block_hasher(self.block_size)
def _make_request(self, prompt_tokens, existing_hashes=None, output_tokens=None):
req = SimpleNamespace(
prompt_token_ids=prompt_tokens,
output_token_ids=output_tokens or [],
_prompt_hashes=existing_hashes if existing_hashes is not None else [],
multimodal_inputs=None,
)
return req
def test_returns_callable(self):
from fastdeploy.cache_manager.v1.cache_utils import get_request_block_hasher
hasher = get_request_block_hasher(4)
self.assertTrue(callable(hasher))
def test_single_complete_block(self):
req = self._make_request(prompt_tokens=[1, 2, 3, 4])
hashes = self.hasher(req)
self.assertEqual(len(hashes), 1)
self.assertIsInstance(hashes[0], str)
def test_two_complete_blocks(self):
req = self._make_request(prompt_tokens=list(range(8)))
hashes = self.hasher(req)
self.assertEqual(len(hashes), 2)
def test_incomplete_last_block_not_hashed(self):
# 5 tokens with block_size=4 → 1 complete block, 1 incomplete
req = self._make_request(prompt_tokens=list(range(5)))
hashes = self.hasher(req)
self.assertEqual(len(hashes), 1)
def test_existing_hashes_skip_computed_blocks(self):
# First compute 1 block
req = self._make_request(prompt_tokens=list(range(4)))
first_hashes = self.hasher(req)
# Now add more tokens, provide existing hashes so they aren't recomputed
req2 = self._make_request(
prompt_tokens=list(range(8)),
existing_hashes=first_hashes,
)
new_hashes = self.hasher(req2)
self.assertEqual(len(new_hashes), 1) # only the second block
def test_chained_hashes_differ_between_blocks(self):
req = self._make_request(prompt_tokens=list(range(8)))
hashes = self.hasher(req)
self.assertNotEqual(hashes[0], hashes[1])
def test_deterministic_across_calls(self):
req1 = self._make_request(prompt_tokens=[1, 2, 3, 4])
req2 = self._make_request(prompt_tokens=[1, 2, 3, 4])
self.assertEqual(self.hasher(req1), self.hasher(req2))
def test_empty_tokens_returns_empty(self):
req = self._make_request(prompt_tokens=[])
hashes = self.hasher(req)
self.assertEqual(hashes, [])
def test_output_tokens_included_in_hash(self):
# With only prompt tokens filling one block
req_prompt_only = self._make_request(
prompt_tokens=[1, 2],
output_tokens=[3, 4],
)
# The same tokens purely as prompt
req_prompt_full = self._make_request(prompt_tokens=[1, 2, 3, 4])
h1 = self.hasher(req_prompt_only)
h2 = self.hasher(req_prompt_full)
# Both should produce a hash for the first complete block
self.assertEqual(len(h1), 1)
self.assertEqual(len(h2), 1)
# ---------------------------------------------------------------------------
# LayerDoneCounter time-tracking and cleanup
# ---------------------------------------------------------------------------
class TestLayerDoneCounterTimeTracking(unittest.TestCase):
"""Tests for get_layer_complete_time, get_layer_wait_time, get_all_layer_times, get_elapsed_time."""
def setUp(self):
from fastdeploy.cache_manager.v1.cache_utils import LayerDoneCounter
self.LayerDoneCounter = LayerDoneCounter
def test_get_layer_complete_time_none_before_done(self):
counter = self.LayerDoneCounter(num_layers=3)
self.assertIsNone(counter.get_layer_complete_time(0))
def test_get_layer_complete_time_after_mark_done(self):
counter = self.LayerDoneCounter(num_layers=3)
before = time.time()
counter.mark_layer_done(0)
after = time.time()
t = counter.get_layer_complete_time(0)
self.assertIsNotNone(t)
self.assertGreaterEqual(t, before)
self.assertLessEqual(t, after + 0.01)
def test_get_layer_wait_time_none_before_done(self):
counter = self.LayerDoneCounter(num_layers=3)
self.assertIsNone(counter.get_layer_wait_time(1))
def test_get_layer_wait_time_is_non_negative(self):
counter = self.LayerDoneCounter(num_layers=3)
counter.mark_layer_done(2)
wait_time = counter.get_layer_wait_time(2)
self.assertIsNotNone(wait_time)
self.assertGreaterEqual(wait_time, 0.0)
def test_get_all_layer_times_empty_before_any_done(self):
counter = self.LayerDoneCounter(num_layers=4)
times = counter.get_all_layer_times()
self.assertEqual(times, {})
def test_get_all_layer_times_after_mark_all_done(self):
counter = self.LayerDoneCounter(num_layers=4)
counter.mark_all_done()
times = counter.get_all_layer_times()
self.assertEqual(set(times.keys()), {0, 1, 2, 3})
def test_get_all_layer_times_returns_copy(self):
counter = self.LayerDoneCounter(num_layers=2)
counter.mark_layer_done(0)
times = counter.get_all_layer_times()
times[999] = 0.0 # mutate the returned dict
# Should not affect internal state
self.assertNotIn(999, counter.get_all_layer_times())
def test_get_elapsed_time_increases(self):
counter = self.LayerDoneCounter(num_layers=2)
t1 = counter.get_elapsed_time()
time.sleep(0.02)
t2 = counter.get_elapsed_time()
self.assertGreater(t2, t1)
class TestLayerDoneCounterGetNumLayers(unittest.TestCase):
def test_get_num_layers(self):
from fastdeploy.cache_manager.v1.cache_utils import LayerDoneCounter
counter = LayerDoneCounter(num_layers=7)
self.assertEqual(counter.get_num_layers(), 7)
class TestLayerDoneCounterSetLayerEvent(unittest.TestCase):
"""Tests for set_layer_event (no real CUDA event needed)."""
def test_set_layer_event_stores_value(self):
from fastdeploy.cache_manager.v1.cache_utils import LayerDoneCounter
counter = LayerDoneCounter(num_layers=3)
mock_event = object()
counter.set_layer_event(1, mock_event)
self.assertIs(counter._cuda_events[1], mock_event)
def test_set_layer_event_out_of_range_is_safe(self):
from fastdeploy.cache_manager.v1.cache_utils import LayerDoneCounter
counter = LayerDoneCounter(num_layers=3)
# Should not raise
counter.set_layer_event(99, object())
class TestLayerDoneCounterCleanup(unittest.TestCase):
def test_cleanup_clears_events(self):
from fastdeploy.cache_manager.v1.cache_utils import LayerDoneCounter
counter = LayerDoneCounter(num_layers=2)
counter.mark_all_done()
# No waiters, all done → cleanup should succeed
counter.cleanup()
self.assertEqual(len(counter._cuda_events), 0)
def test_cleanup_with_active_waiter_is_noop(self):
from fastdeploy.cache_manager.v1.cache_utils import LayerDoneCounter
counter = LayerDoneCounter(num_layers=2)
# Manually increment wait count to simulate an active waiter
counter._increment_wait_count()
counter.cleanup()
# Should NOT have cleared events (waiter still active)
self.assertEqual(len(counter._cuda_events), 2)
counter._decrement_wait_count()
class TestLayerDoneCounterInternalHelpers(unittest.TestCase):
def setUp(self):
from fastdeploy.cache_manager.v1.cache_utils import LayerDoneCounter
self.LayerDoneCounter = LayerDoneCounter
def test_increment_and_decrement_wait_count(self):
counter = self.LayerDoneCounter(num_layers=2)
counter._increment_wait_count()
self.assertEqual(counter._wait_count, 1)
counter._decrement_wait_count()
self.assertEqual(counter._wait_count, 0)
def test_decrement_does_not_go_below_zero(self):
counter = self.LayerDoneCounter(num_layers=2)
counter._decrement_wait_count()
self.assertEqual(counter._wait_count, 0)
def test_should_cleanup_false_when_not_all_done(self):
counter = self.LayerDoneCounter(num_layers=3)
self.assertFalse(counter._should_cleanup())
def test_should_cleanup_true_when_all_done_no_waiters(self):
counter = self.LayerDoneCounter(num_layers=2)
counter.mark_all_done()
self.assertTrue(counter._should_cleanup())
def test_should_cleanup_false_when_waiter_present(self):
counter = self.LayerDoneCounter(num_layers=2)
counter.mark_all_done()
counter._increment_wait_count()
self.assertFalse(counter._should_cleanup())
counter._decrement_wait_count()
if __name__ == "__main__":
unittest.main()