061792db1c
- 用例①两章无冲突全自动：两 accepted 章 + 两 digest 行 + job done written=[1,2] - 用例②注入冲突→interrupt→job awaiting→resume 带裁决→续跑→accept； interrupt+resume 横跨两次端点调用，单进程单 MemorySaver（覆盖 get_checkpointer_factory） - 负向断言：正文/token 不入 job result/状态（HTTP + DB 真源双查） - 守不变量 #1/#3/#4/#5/#9
430 lines
18 KiB
Python
430 lines
18 KiB
Python
"""C4 多章工作流链 端到端（chain-workflow §9 C4 / §10）——真 pg + mock 网关零 token。
|
||
|
||
证明杀手锏闭环（不变量 #1/#3/#4/#5/#9）：批量量产，但每章过 `write→四审→decide→accept`，
|
||
遇冲突 interrupt 交人裁决再 resume 续跑——星月架构做不到的一致性闸。
|
||
|
||
用例 1 · 两章无冲突全自动
|
||
`POST /projects` → `POST .../chains/draft_volume/run {start=1,count=2}` → 202 {job_id}；
|
||
BackgroundTask `run_chain_job` 自建独立 session 驱动链图（mock 网关零冲突）→ 两章
|
||
`write→四审(无冲突)→accept` 自动跑完。DB 真源逐项断言：
|
||
- `chapters` 第 1/2 章各一行 `accepted`（write 落 draft → accept 晋升终稿）；
|
||
- `chapter_digests` 两行（digest 从终稿提炼，不变量 #4）；
|
||
- `GET /jobs/{id}` status=done，result.written=[1,2]、completed=True、awaiting_chapter=None。
|
||
负向：job result/状态绝不含正文/prompt/token（§5）。
|
||
|
||
用例 2 · 注入冲突 → interrupt → awaiting → resume → accept
|
||
续审第 1 章报冲突 → `decide` 命中 → `accept` 节点 `interrupt()` 暂停 → job=awaiting_input、
|
||
result.awaiting_chapter=1、`chapters` 第 1 章仍 `draft`（未晋升）、`chapter_digests` 空。
|
||
`POST .../chains/runs/{job_id}/resume {decisions:[...]}` → 202 → resume 续跑 → accept →
|
||
job done、第 1 章 `accepted`、digest 一行。
|
||
interrupt+resume 横跨**两次端点调用**，靠单进程单 `MemorySaver`（经依赖覆盖注入
|
||
`get_checkpointer_factory` 的同一 saver 实例）持久控制流位置 + pending handle。
|
||
|
||
确定性 & 零成本（同 M3/M4）：真 `Gateway` + 假适配器（据 `req.output_schema` 分支返固定
|
||
parsed/text，绝不联网）+ 真 `SqlAlchemyLedgerSink`。检查点用 MemorySaver（绝不连真 Postgres
|
||
检查点）。无 pg → skip。
|
||
|
||
坑（见 memory/gotchas）：
|
||
- `get_checkpointer_factory` 必 override 成产**同一** MemorySaver 的工厂——run 与 resume 两次
|
||
端点调用须落同一 thread_id 的同一检查点存储，否则 resume 找不到 interrupt 现场。
|
||
- `get_session_factory` 必 override → `e2e_sm`（真 sessionmaker，同测试 engine/loop）；
|
||
BackgroundTask 在请求 session 关闭后才跑。
|
||
- `get_digest_gateway_builder` 必 override → 返「按 session 建真 Gateway 包假适配器」的 builder
|
||
（链 accept 节点终稿提炼用 light 档），否则会从空凭据建真 OpenAI 适配器。
|
||
- ASGITransport 下 `await client.post` 等 ASGI app（含 background task）跑完才返回。
|
||
"""
|
||
|
||
from __future__ import annotations
|
||
|
||
import uuid
|
||
from collections.abc import AsyncIterator, Awaitable, Callable
|
||
from contextlib import AbstractAsyncContextManager, asynccontextmanager
|
||
from typing import Annotated, Any
|
||
|
||
import httpx
|
||
import pytest
|
||
from asgi_lifespan import LifespanManager
|
||
from fastapi import Depends
|
||
from langgraph.checkpoint.memory import MemorySaver
|
||
from sqlalchemy import delete, select
|
||
from sqlalchemy.ext.asyncio import AsyncSession, async_sessionmaker
|
||
from ww_agents import (
|
||
Conflict,
|
||
ContinuityReview,
|
||
ForeshadowReview,
|
||
PaceReview,
|
||
StyleDriftReview,
|
||
)
|
||
from ww_api.services.digest_extraction import ChapterDigestFacts
|
||
from ww_db import get_session, get_sessionmaker
|
||
from ww_db.models import (
|
||
Chapter,
|
||
ChapterDigest,
|
||
ChapterReview,
|
||
Job,
|
||
Project,
|
||
UsageLedger,
|
||
)
|
||
from ww_llm_gateway import Gateway, SqlAlchemyLedgerSink, resolve_route
|
||
from ww_llm_gateway.adapters.base import (
|
||
Capabilities,
|
||
ProviderResult,
|
||
ProviderUsage,
|
||
StreamChunk,
|
||
)
|
||
from ww_llm_gateway.types import LlmRequest
|
||
|
||
# 三档（writer/analyst/light）默认都路由 deepseek（config.tier_defaults）。
|
||
_PROVIDER = "deepseek"
|
||
|
||
# 链写章收集版正文（非 SSE，经 gateway.run → adapter.complete，output_schema is None）。
|
||
_CHAPTER_TEXT = "灵气如潮，少年握剑而立，山门之上风雷大作。"
|
||
|
||
# 各档假用量（喂记账；证明各档调用各落一条 usage_ledger）。
|
||
_USAGE = ProviderUsage(input_tokens=13, output_tokens=7)
|
||
|
||
# 三审固定无冲突产物（让四审图跑通；continuity 的冲突由 fixture 参数注入）。
|
||
_FORESHADOW_REVIEW = ForeshadowReview(planted=[], resolved=[])
|
||
_PACE_REVIEW = PaceReview(water=[], hook=True, beat_map=[1, 2, 3])
|
||
_STYLE_DRIFT = StyleDriftReview(score=88, segments=[])
|
||
|
||
|
||
class _FakeChainAdapter:
|
||
"""实现 `ProviderAdapter`：据 `req.output_schema` 分支返 parsed/text，绝不联网。
|
||
|
||
- `output_schema is None` → 链写章收集版纯文本（writer 档，`gateway.run`）。
|
||
- 四审 schema → 各审 parsed（continuity 注入冲突，其余无冲突）。
|
||
- `ChapterDigestFacts` → 终稿 digest 提炼（light 档）。
|
||
`stream()` 实现但链不走（收集版量产非 SSE）。
|
||
"""
|
||
|
||
provider = _PROVIDER
|
||
|
||
def __init__(self, *, continuity_conflicts: list[Conflict]) -> None:
|
||
self._continuity = ContinuityReview(conflicts=continuity_conflicts)
|
||
|
||
def capabilities(self) -> Capabilities:
|
||
return Capabilities(structured_output=True)
|
||
|
||
async def complete(self, req: LlmRequest, model: str) -> ProviderResult:
|
||
schema = req.output_schema
|
||
if schema is None:
|
||
return ProviderResult(text=_CHAPTER_TEXT, parsed=None, usage=_USAGE)
|
||
if schema is ContinuityReview:
|
||
return ProviderResult(
|
||
text=self._continuity.model_dump_json(), parsed=self._continuity, usage=_USAGE
|
||
)
|
||
if schema is ForeshadowReview:
|
||
return ProviderResult(
|
||
text=_FORESHADOW_REVIEW.model_dump_json(), parsed=_FORESHADOW_REVIEW, usage=_USAGE
|
||
)
|
||
if schema is PaceReview:
|
||
return ProviderResult(
|
||
text=_PACE_REVIEW.model_dump_json(), parsed=_PACE_REVIEW, usage=_USAGE
|
||
)
|
||
if schema is StyleDriftReview:
|
||
return ProviderResult(
|
||
text=_STYLE_DRIFT.model_dump_json(), parsed=_STYLE_DRIFT, usage=_USAGE
|
||
)
|
||
# 否则视为终稿 digest 提炼（ChapterDigestFacts）。
|
||
facts = ChapterDigestFacts(summary="终稿摘要", events=["少年出山"], locations=["山门"])
|
||
return ProviderResult(text=facts.model_dump_json(), parsed=facts, usage=_USAGE)
|
||
|
||
async def stream(self, req: LlmRequest, model: str) -> AsyncIterator[StreamChunk]:
|
||
yield StreamChunk(text=_CHAPTER_TEXT)
|
||
yield StreamChunk(usage=_USAGE)
|
||
|
||
|
||
@pytest.fixture
|
||
async def e2e_sm() -> AsyncIterator[async_sessionmaker[AsyncSession]]:
|
||
"""真 DB session 工厂；无 pg 时跳过（每测试清缓存重建 engine、结束 dispose）。"""
|
||
get_sessionmaker.cache_clear()
|
||
maker = get_sessionmaker()
|
||
try:
|
||
async with maker() as probe:
|
||
await probe.execute(select(1))
|
||
except Exception:
|
||
pytest.skip("postgres not reachable")
|
||
yield maker
|
||
await maker.kw["bind"].dispose()
|
||
get_sessionmaker.cache_clear()
|
||
|
||
|
||
def _gateway_for(adapter: _FakeChainAdapter, session: AsyncSession) -> Gateway:
|
||
"""真 Gateway + 给定假适配器 + 真 ledger（绑给定 session）。"""
|
||
return Gateway(
|
||
adapters={_PROVIDER: adapter},
|
||
ledger=SqlAlchemyLedgerSink(session),
|
||
resolver=resolve_route,
|
||
)
|
||
|
||
|
||
def _chain_gateway_override(
|
||
adapter: _FakeChainAdapter,
|
||
) -> Callable[[AsyncSession], Gateway]:
|
||
"""`get_chain_gateway` 覆盖：请求阶段凭据探测 → 真 Gateway 包假适配器（绕过 503）。"""
|
||
|
||
def _override(session: Annotated[AsyncSession, Depends(get_session)]) -> Gateway:
|
||
return _gateway_for(adapter, session)
|
||
|
||
return _override
|
||
|
||
|
||
def _digest_builder_override(
|
||
adapter: _FakeChainAdapter,
|
||
) -> Callable[[], Callable[[AsyncSession], Awaitable[Gateway]]]:
|
||
"""`get_digest_gateway_builder` 覆盖：返「按 session 建真 Gateway 包假适配器」的 builder。
|
||
|
||
链 accept 节点在自建短事务里据该 session 现建 light 档 digest 网关（终稿提炼）。
|
||
"""
|
||
|
||
def _get_builder() -> Callable[[AsyncSession], Awaitable[Gateway]]:
|
||
async def _build(session: AsyncSession) -> Gateway:
|
||
return _gateway_for(adapter, session)
|
||
|
||
return _build
|
||
|
||
return _get_builder
|
||
|
||
|
||
def _memsaver_override(
|
||
saver: MemorySaver,
|
||
) -> Callable[[], Callable[[], AbstractAsyncContextManager[MemorySaver]]]:
|
||
"""`get_checkpointer_factory` 覆盖：返产**同一** saver 的工厂（run/resume 跨调用共用）。"""
|
||
|
||
@asynccontextmanager
|
||
async def _ctx() -> AsyncIterator[MemorySaver]:
|
||
yield saver
|
||
|
||
return lambda: lambda: _ctx()
|
||
|
||
|
||
async def _cleanup(e2e_sm: async_sessionmaker[AsyncSession], project_uuid: uuid.UUID) -> None:
|
||
"""按 FK 顺序清理（多数子表 FK CASCADE；显式删保独立）。"""
|
||
async with e2e_sm() as cleanup:
|
||
await cleanup.execute(delete(UsageLedger).where(UsageLedger.project_id == project_uuid))
|
||
await cleanup.execute(delete(Job).where(Job.project_id == project_uuid))
|
||
await cleanup.execute(delete(ChapterDigest).where(ChapterDigest.project_id == project_uuid))
|
||
await cleanup.execute(delete(ChapterReview).where(ChapterReview.project_id == project_uuid))
|
||
await cleanup.execute(delete(Chapter).where(Chapter.project_id == project_uuid))
|
||
await cleanup.execute(delete(Project).where(Project.id == project_uuid))
|
||
await cleanup.commit()
|
||
|
||
|
||
def _build_app(adapter: _FakeChainAdapter, saver: MemorySaver, e2e_sm: Any) -> Any:
|
||
"""组装注入了全 fake 缝的 app（凭据探测/session 工厂/checkpointer/digest builder）。"""
|
||
from ww_api.main import create_app
|
||
from ww_api.services.chain_deps import get_checkpointer_factory
|
||
from ww_api.services.project_deps import (
|
||
get_chain_gateway,
|
||
get_digest_gateway_builder,
|
||
get_session_factory,
|
||
)
|
||
|
||
app = create_app()
|
||
app.dependency_overrides[get_chain_gateway] = _chain_gateway_override(adapter)
|
||
app.dependency_overrides[get_session_factory] = lambda: e2e_sm
|
||
app.dependency_overrides[get_checkpointer_factory] = _memsaver_override(saver)
|
||
app.dependency_overrides[get_digest_gateway_builder] = _digest_builder_override(adapter)
|
||
return app
|
||
|
||
|
||
async def test_chain_two_chapters_no_conflict_full_auto(
|
||
e2e_sm: async_sessionmaker[AsyncSession],
|
||
) -> None:
|
||
"""用例 1：两章无冲突全自动 → 两 accepted 章 + 两 digest 行 + job done written=[1,2]。"""
|
||
adapter = _FakeChainAdapter(continuity_conflicts=[]) # 零冲突 → 全自动
|
||
saver = MemorySaver()
|
||
app = _build_app(adapter, saver, e2e_sm)
|
||
|
||
transport = httpx.ASGITransport(app=app)
|
||
project_uuid: uuid.UUID | None = None
|
||
try:
|
||
async with LifespanManager(app):
|
||
async with httpx.AsyncClient(transport=transport, base_url="http://test") as client:
|
||
create_resp = await client.post("/projects", json={"title": "链 E2E 无冲突作品"})
|
||
assert create_resp.status_code == 201
|
||
project_id = create_resp.json()["id"]
|
||
project_uuid = uuid.UUID(project_id)
|
||
|
||
run_resp = await client.post(
|
||
f"/projects/{project_id}/chains/draft_volume/run",
|
||
json={"start_chapter_no": 1, "count": 2},
|
||
)
|
||
assert run_resp.status_code == 202
|
||
run_body = run_resp.json()
|
||
assert run_body["chain_key"] == "draft_volume"
|
||
job_id = run_body["job_id"]
|
||
|
||
# 链经 BackgroundTask 已跑完（ASGITransport 等其完成）→ job done。
|
||
job_resp = await client.get(f"/jobs/{job_id}")
|
||
assert job_resp.status_code == 200
|
||
job = job_resp.json()
|
||
assert job["status"] == "done", job
|
||
assert job["kind"] == "chain"
|
||
result = job["result"]
|
||
assert result["written"] == [1, 2]
|
||
assert result["completed"] is True
|
||
assert result["awaiting_chapter"] is None
|
||
# 负向：job result 绝不含正文/token（只章号/计数/标志）。
|
||
assert _CHAPTER_TEXT not in str(result)
|
||
assert "终稿摘要" not in str(result)
|
||
|
||
# DB 真源逐项断言。
|
||
assert project_uuid is not None
|
||
async with e2e_sm() as verify:
|
||
chapters = (
|
||
(
|
||
await verify.execute(
|
||
select(Chapter)
|
||
.where(Chapter.project_id == project_uuid)
|
||
.order_by(Chapter.chapter_no, Chapter.version)
|
||
)
|
||
)
|
||
.scalars()
|
||
.all()
|
||
)
|
||
# 第 1/2 章各晋升一个 accepted 版次。
|
||
accepted = [c for c in chapters if c.status == "accepted"]
|
||
assert {c.chapter_no for c in accepted} == {1, 2}
|
||
assert all(c.content == _CHAPTER_TEXT for c in accepted)
|
||
|
||
digests = (
|
||
(
|
||
await verify.execute(
|
||
select(ChapterDigest)
|
||
.where(ChapterDigest.project_id == project_uuid)
|
||
.order_by(ChapterDigest.chapter_no)
|
||
)
|
||
)
|
||
.scalars()
|
||
.all()
|
||
)
|
||
assert [d.chapter_no for d in digests] == [1, 2]
|
||
|
||
job_rows = (
|
||
(await verify.execute(select(Job).where(Job.project_id == project_uuid)))
|
||
.scalars()
|
||
.all()
|
||
)
|
||
assert len(job_rows) == 1
|
||
assert job_rows[0].status == "done"
|
||
assert job_rows[0].kind == "chain"
|
||
# 负向（DB 真源）：job result 列不含正文/token。
|
||
assert _CHAPTER_TEXT not in str(job_rows[0].result)
|
||
finally:
|
||
if project_uuid is not None:
|
||
await _cleanup(e2e_sm, project_uuid)
|
||
|
||
|
||
async def test_chain_conflict_interrupts_then_resume_accepts(
|
||
e2e_sm: async_sessionmaker[AsyncSession],
|
||
) -> None:
|
||
"""用例 2：注入冲突 → interrupt → job awaiting → resume 带裁决 → 续跑 → accept。
|
||
|
||
interrupt+resume 横跨两次端点调用，靠单进程单 MemorySaver（同一 thread_id 检查点）。
|
||
"""
|
||
conflict = Conflict(type="性格漂移", where="第2段", suggestion="统一主角姓名为「萧寒」")
|
||
adapter = _FakeChainAdapter(continuity_conflicts=[conflict])
|
||
saver = MemorySaver() # 单实例横跨 run + resume 两次端点调用
|
||
app = _build_app(adapter, saver, e2e_sm)
|
||
|
||
transport = httpx.ASGITransport(app=app)
|
||
project_uuid: uuid.UUID | None = None
|
||
try:
|
||
async with LifespanManager(app):
|
||
async with httpx.AsyncClient(transport=transport, base_url="http://test") as client:
|
||
create_resp = await client.post("/projects", json={"title": "链 E2E 冲突作品"})
|
||
assert create_resp.status_code == 201
|
||
project_id = create_resp.json()["id"]
|
||
project_uuid = uuid.UUID(project_id)
|
||
|
||
# 1) 发起：第 1 章四审报冲突 → interrupt → job awaiting_input。
|
||
run_resp = await client.post(
|
||
f"/projects/{project_id}/chains/draft_volume/run",
|
||
json={"start_chapter_no": 1, "count": 1},
|
||
)
|
||
assert run_resp.status_code == 202
|
||
job_id = run_resp.json()["job_id"]
|
||
|
||
job = (await client.get(f"/jobs/{job_id}")).json()
|
||
assert job["status"] == "awaiting_input", job
|
||
assert job["result"]["awaiting_chapter"] == 1
|
||
assert job["result"]["completed"] is False
|
||
assert job["result"]["written"] == []
|
||
|
||
# awaiting 时第 1 章仍 draft（未经 accept 晋升），无 digest。
|
||
async with e2e_sm() as mid:
|
||
mid_chapters = (
|
||
(
|
||
await mid.execute(
|
||
select(Chapter).where(Chapter.project_id == project_uuid)
|
||
)
|
||
)
|
||
.scalars()
|
||
.all()
|
||
)
|
||
assert all(c.status == "draft" for c in mid_chapters)
|
||
assert mid_chapters != []
|
||
mid_digests = (
|
||
(
|
||
await mid.execute(
|
||
select(ChapterDigest).where(
|
||
ChapterDigest.project_id == project_uuid
|
||
)
|
||
)
|
||
)
|
||
.scalars()
|
||
.all()
|
||
)
|
||
assert mid_digests == []
|
||
|
||
# 2) resume：作者裁决冲突 → 续跑 accept → done。
|
||
resume_resp = await client.post(
|
||
f"/projects/{project_id}/chains/runs/{job_id}/resume",
|
||
json={
|
||
"decisions": [
|
||
{"conflict_index": 0, "verdict": "ignore", "note": "笔误，忽略"}
|
||
]
|
||
},
|
||
)
|
||
assert resume_resp.status_code == 202
|
||
|
||
job = (await client.get(f"/jobs/{job_id}")).json()
|
||
assert job["status"] == "done", job
|
||
assert job["result"]["written"] == [1]
|
||
assert job["result"]["completed"] is True
|
||
assert job["result"]["awaiting_chapter"] is None
|
||
|
||
# DB 真源断言：resume 后第 1 章 accepted + digest 一行。
|
||
assert project_uuid is not None
|
||
async with e2e_sm() as verify:
|
||
chapters = (
|
||
(await verify.execute(select(Chapter).where(Chapter.project_id == project_uuid)))
|
||
.scalars()
|
||
.all()
|
||
)
|
||
accepted = [c for c in chapters if c.status == "accepted"]
|
||
assert {c.chapter_no for c in accepted} == {1}
|
||
|
||
digests = (
|
||
(
|
||
await verify.execute(
|
||
select(ChapterDigest).where(ChapterDigest.project_id == project_uuid)
|
||
)
|
||
)
|
||
.scalars()
|
||
.all()
|
||
)
|
||
assert [d.chapter_no for d in digests] == [1]
|
||
|
||
# 负向：job 状态/result 不含正文/裁决正文外的 token/prompt。
|
||
job_row = (
|
||
await verify.execute(select(Job).where(Job.project_id == project_uuid))
|
||
).scalar_one()
|
||
assert _CHAPTER_TEXT not in str(job_row.result)
|
||
assert job_row.status == "done"
|
||
finally:
|
||
if project_uuid is not None:
|
||
await _cleanup(e2e_sm, project_uuid)
|