Files
AgentLab/backend/app/api/ledger.py
T
ThearchyHelios b3dcddb52e 召回复核修正(后端):人工节点不再挡召回、读不出召回卡记失败、旧版不再接着算、团队渠道一次召回只推一条、目录核对按数据源合并排队;账本接口带上召回标记,对账结果可以反馈
- W4R-01:召回只查新版图里会不会写外部系统(apps/backfill 抽出 graph_external_write,graph_side_effect 复用它)。召回图只有两版
  的取数和口径卡,图里别处的人工节点、按流程使用的问题碰不到它;口径卡上游的人工节点 derive 本来就认作数据部分不能复用。
  「按新版重新出具」走完整的正式运行,仍按 graph_side_effect 全套拒绝。原来带审批的口径升版后一组都不召回,L6 对它们失效。
- W4R-09:收尾时召回卡读不出、或缺了某个比较的旧值 / 新值,照实记重算失败(读不出召回卡),原答案一个不动。原来两边都是空,
  按渲染比被当成「没变」,给每个原答案永久盖上「已核对,不影响」,收过尾之后再也不核对。
- W4R-12:这一版已经不是当前发布版(之后又发布了更新的版本)时不再发起召回、不给接着算,批次里写 continue_reason,接着算的
  接口 409 带同一句。同启动时接着做只认当前发布版:新版本的召回覆盖更早的所有版本,再按旧版开「口径升到 vN」的待办只是添乱。
- W4R-13:开登录时一组答案几个主人就几条待办,全局渠道却按这一批(recall.batch_of_key:同一次召回、同一个来由)只在最早开的
  那条上推一次,答案、待办按这一批汇总,几条的链到待办箱;别的几条连 task.opened 也不再推全局渠道。个人渠道照旧每人一条。
- W4R-11:目录审阅之后的核对按数据源排队、批量审阅的几张表合成一遍(catalog_pass,同一数据源一把锁);查询快照按数据源在库里
  筛、按运行结束时间分页、每张表凑够 BASIS_MAX_RUNS 次运行就停;冻结的表结构快照这一遍里按工件 id 只读一次。原来批量审阅
  几十张表就同时起几十个后台任务,每个都全表扫查询快照、反复读同一份几 MB 的表结构快照。
- W4R-03(本包那一半):GET /api/runs/{id}/ledger 加 recall:checked(已按 vN 核对,不影响)、basis(依据有变:哪一种、几项、
  新出现几条错误级问题)、tasks(看的这个人手上说的是这个答案、还没处理的 recall 待办)。不放数、不放表名。几种状态都带。
- W4R-10(本包那一半):新事件 ledger.feedback(对账运行上的封存后事件),POST /api/ledger/answers/{run}/reconcile/{rid}/feedback
  {verdict: useful | unclear} 记反馈,同一个人改主意只算最后一次(app/ledger/feedback.py);账本接口的 reconciled 每一项带
  feedback {mine, useful, unclear};统计 E2 的 feedback 数到它,去掉「第 4 波接上」的占位说法。离线核对页、封存后事件摘要
  都认得这个事件。

测试:test_ledger_recall 新增七个用例、改了人工节点那一个(召回照算、重新出具照实拒绝);新增 test_ledger_feedback;
test_ledger_api 的 E2 断言跟上。每个修复都先写失败的测试、去掉修复回跑确认抓得住。后端全量 14237 通过。
2026-10-16 22:16:58 +08:00

256 lines
16 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.
"""账本(BE-3,只读):工作流版本被哪些应用钉住、生效计数;生效记录查询。设计文档 §2.9 L4。
不另建存储:ActionRecord 本身带着工作流、版本、graph_hash、能力版本、payload_hash、回执;「口径 → 应用」就是固定这一版的
应用版本。用在发布弹窗、退役确认、版本列表三处。
口径账本(docs/design/next-round/02-口径账本.md)在这里加的接口:
- GET /api/runs/{run_id}/ledger(runs_router,和证据同一道门):一次运行里每个数的口径标签、整题判断、每条查询的比对、能做的
动作、出具过的正式运行、对过的账(app/ledger/view.py);
- GET /api/ledger/stats(router,搭建者):账本自身给出的统计 E1–E4,从比对工件、运行、审计现算(app/ledger/stats.py);
- POST /api/ledger/answers/{run_id}/issue(answers_router,登录即可,第 3 波):按口径正式出具(app/ledger/issue.py,拍板
L1、L5:参数只从比对工件取,使用者只能按他有权使用的应用固定着的版本出具);
- POST /api/ledger/answers/{run_id}/reconcile(同上):口径对账,点开才跑一次确定性派生的只读图(app/ledger/reconcile.py);
- POST /api/ledger/answers/{run_id}/reconcile/{reconcile_id}/feedback(同上,第 4 波):对账结果上的「说明有用 / 看不懂」
(app/ledger/feedback.py,统计 E2);
- GET /api/ledger/certify-queue、POST /api/ledger/certify-queue/dismiss(router,搭建者,第 3 波):可沉淀的问题、「不是口径」
(app/ledger/certify.py)。
- GET /api/ledger/recall/{workflow_id}/{version}、POST …/continue(router,搭建者,第 4 波):口径升到这一版之后的召回批次——
每组的结论、「另有 M 组未重算」、手动接着算(app/ledger/recall.py)。版本用量(workflow-versions)另加问数回答、导出、召回。
这几个新接口都不放给访问密钥(auth/keys.KEY_ROUTES 不登记)。
"""
from __future__ import annotations
from typing import Any
from datetime import datetime
from fastapi import APIRouter, Depends, HTTPException, Query
from pydantic import BaseModel, Field
from sqlalchemy import select
from sqlalchemy.ext.asyncio import AsyncSession
from app.db.base import SessionLocal, get_session
from app.db.models import Workflow
from app.db.models_process import ActionRecord
router = APIRouter(prefix="/api/ledger", tags=["ledger"])
#: 口径账本里按运行读的那几个(一个答案里每个数的口径标签等,docs/design/next-round/02-口径账本.md):挂在
#: /api/runs/{run_id}/ledger… 下,和证据接口同一道门(main.py 挂 evidence_scope:使用者只读、而且要能读这次运行)
runs_router = APIRouter(prefix="/api/runs", tags=["ledger"])
#: 按口径正式出具、口径对账这类要写的(docs/design/next-round/02-口径账本.md 做法 5、6):使用者也能发起(总方案拍板
#: 账本-L5:对账都可以,正式出具只限有权使用的应用固定着的口径版本),所以 main.py 只挂登录即可,接口里按运行和应用授权再判
answers_router = APIRouter(prefix="/api/ledger/answers", tags=["ledger"])
@router.get("/workflow-versions/{workflow_id}/{version}")
async def workflow_version(workflow_id: str, version: int, session: AsyncSession = Depends(get_session)
) -> dict[str, Any]:
"""一个工作流版本被哪些应用固定、生效记录数、正式运行数;另加口径账本的用量(next-round 02 做法 8):被几次问数回答用过
(answers:比对里等价已证明的探索运行)、导出过几次(exports:这些答案和这一版的正式运行上的报告导出审计)、定时任务几个、
召回(recall:以这一版为新版本召回过几组、其中数变了几组,作为旧版本被召回过几组)。"""
from app.apps.versions import usage
from app.ledger.uses import version_usage
if await session.get(Workflow, workflow_id) is None:
raise HTTPException(404, "工作流不存在,可能已被删除")
base = await usage(session, workflow_id, version)
extra = await version_usage(session, workflow_id, version)
return {**base, "answers": extra["answers"], "exports": extra["exports"], "schedules": extra["schedules"],
"recall": extra["recall"]}
def effect_row(r: ActionRecord, *, receipt: bool = True) -> dict[str, Any]:
return {"id": r.id, "run_id": r.run_id, "node_id": r.node_id, "app_id": r.app_id, "workflow_id": r.workflow_id,
"workflow_version": r.workflow_version, "graph_hash": r.graph_hash, "capability": r.capability,
"capability_version": r.capability_version, "environment": r.environment, "state": r.state,
"attempt": r.attempt, "payload_hash": r.payload_hash, "preview": r.preview,
"receipt": r.receipt if receipt else None, "error": r.error, "resolution": r.resolution,
"created_at": r.created_at.isoformat() if r.created_at else None,
"updated_at": r.updated_at.isoformat() if r.updated_at else None}
@router.get("/effects")
async def effects(workflow_id: str | None = None, version: int | None = None, app_id: str | None = None,
state: str | None = None, capability: str | None = None,
limit: int = Query(default=100, ge=1, le=500), session: AsyncSession = Depends(get_session)
) -> dict[str, Any]:
"""生效记录,新的在前。演练(simulated)照样列出,状态写明是演练。"""
query = select(ActionRecord)
if workflow_id:
query = query.where(ActionRecord.workflow_id == workflow_id)
if version is not None:
query = query.where(ActionRecord.workflow_version == version)
if app_id:
query = query.where(ActionRecord.app_id == app_id)
if state:
query = query.where(ActionRecord.state == state)
if capability:
query = query.where(ActionRecord.capability == capability)
rows = (await session.execute(query.order_by(ActionRecord.created_at.desc()).limit(limit))).scalars()
return {"items": [effect_row(r) for r in rows]}
# --------------------------------------------------------------------------
# 口径账本(next-round 02)
# --------------------------------------------------------------------------
@runs_router.get("/{run_id}/ledger")
async def run_ledger(run_id: str) -> dict[str, Any]:
"""一次运行的口径账本:每个数来自哪一份口径(等价已证明 / 助手选用)、相关口径差在哪、这次的推理、没比对的原因;整题判断;
每条查询和哪几份口径比过。比对在封存之后做(note 写明不在封存范围内);没比对过、能比的运行这时排上按需比对,答复
state=pending。绑定的值是这次运行查到过的遮罩值时不给值,差异条目里的 SQL 片段按同一套遮罩换掉。"""
from app.ledger.view import run_view
return await run_view(run_id)
@router.get("/stats")
async def ledger_stats(since: datetime | None = Query(default=None, description="只算这之后发起的运行(ISO 时间)"),
until: datetime | None = Query(default=None, description="只算这之前发起的运行"),
origin: str | None = Query(default=None, description="逗号分隔的运行来源,缺省是探索运行的几种"),
session: AsyncSession = Depends(get_session)) -> dict[str, Any]:
"""账本自身给出的统计 E1–E4(app/ledger/stats.compute_stats):口径命中、对账、重复度、召回。全部现算,不另存;评测页
「口径命中与重复度」直接调同一个函数。"""
from app.api.deps import current_principal
from app.ledger.stats import compute_stats
origins = [o.strip() for o in (origin or "").split(",") if o.strip()] or None
return await compute_stats(session, since=since, until=until, origins=origins, principal=current_principal())
class IssueIn(BaseModel):
"""按哪份口径出具:只能选口径,参数由服务端从比对工件里取(不收客户端传的参数)。"""
workflow_id: str = Field(min_length=1, max_length=32)
version: int = Field(ge=1)
@answers_router.post("/{run_id}/issue")
async def issue_answer(run_id: str, payload: IssueIn) -> dict[str, Any]:
"""按口径正式出具这次问数据的答案(app/ledger/issue.py):经 launch_run 发起正式运行,来源 ledger_issue。同一运行、同一
口径、同一组参数只发起一次,再点交回已经发起的那一次(created 为假)。不能出具时回 409(整题不等价、版本退役、有人工
节点或写入、缺入口字段……)、403(权限不够:使用者只能按他有权使用的应用固定着的版本出具)、429(超过每日上限),
detail 写明原因。"""
from app.api.coded import CodedHTTPException
from app.api.deps import current_principal
from app.ledger import issue as ledger_issue
try:
run, created = await ledger_issue.issue(run_id, payload.workflow_id, payload.version,
principal=current_principal())
except ledger_issue.IssueRefused as e:
raise CodedHTTPException(e.status_code, e.message, e.code) from e
return {"run_id": run.id, "created": created, "issued": ledger_issue.issued_view(run)}
class ReconcileIn(BaseModel):
"""对哪条查询、和哪份相关口径对账。inputs + confirm:参数要核对时(动作里 needs_check)人核对、改过的参数;pairs:认不出
哪一列对哪个指标时人选的 [{ours: 这次的列, metric: 口径卡上的指标 id}](动作里写明要不要选)。"""
query: str = Field(min_length=1, max_length=64)
workflow_id: str = Field(min_length=1, max_length=32)
version: int = Field(ge=1)
inputs: dict[str, Any] | None = None
confirm: bool = False
pairs: list[dict[str, Any]] | None = None
@answers_router.post("/{run_id}/reconcile", status_code=201)
async def reconcile_answer(run_id: str, payload: ReconcileIn) -> dict[str, Any]:
"""口径对账(app/ledger/reconcile.py):按那一版口径确定性派生只读图,发起一次探索运行(来源 reconcile),差值在对账卡
里算。参数是推断出来的、不全的要先核对(409,code ledger_reconcile_check_inputs);认不出列怎么配的要人选(409,
ledger_reconcile_needs_pairs);多行时维度是遮罩列、键不唯一照实拒绝(409)。"""
from app.api.coded import CodedHTTPException
from app.api.deps import current_principal
from app.ledger import reconcile as ledger_reconcile
try:
run = await ledger_reconcile.reconcile(run_id, ledger_reconcile.Request(
query=payload.query, workflow_id=payload.workflow_id, version=payload.version, inputs=payload.inputs,
confirm=payload.confirm, pairs=payload.pairs), principal=current_principal())
except ledger_reconcile.ReconcileRefused as e:
raise CodedHTTPException(e.status_code, e.message, e.code) from e
return {"run_id": run.id, "reconciled": ledger_reconcile.reconciled_view(run)}
class FeedbackIn(BaseModel):
"""对账结果上的反馈:说明有用(useful)还是看不懂(unclear)。"""
verdict: str = Field(min_length=1, max_length=20)
@answers_router.post("/{run_id}/reconcile/{reconcile_id}/feedback")
async def reconcile_feedback(run_id: str, reconcile_id: str, payload: FeedbackIn) -> dict[str, Any]:
"""对账结果上的「说明有用 / 看不懂」(app/ledger/feedback.py,统计 E2):记成对账运行上的封存后事件,同一个人改主意只算
最后一次。交回这次对账现在的反馈 {mine, useful, unclear}。不认的反馈 422;不是这个答案的对账 404;读不了这个答案 403;
对账还没跑完 409。"""
from app.api.coded import CodedHTTPException
from app.api.deps import current_principal
from app.ledger import feedback
try:
out = await feedback.give(run_id, reconcile_id, payload.verdict, principal=current_principal())
except feedback.FeedbackRefused as e:
raise CodedHTTPException(e.status_code, e.message, e.code) from e
return {"run_id": reconcile_id, "feedback": out}
@router.get("/recall/{workflow_id}/{version}")
async def recall_batches(workflow_id: str, version: int, session: AsyncSession = Depends(get_session)
) -> dict[str, Any]:
"""口径升到这一版之后的召回(app/ledger/recall.batches):每组(旧版本、参数)的结论——数变了 / 没变 / 无法重算:原因 /
重算失败 / 还在算 / 未重算——和「另有 M 组未重算」、能不能接着算。计划从已有的召回运行现算,不另存。"""
from app.ledger import recall
if await session.get(Workflow, workflow_id) is None:
raise HTTPException(404, "工作流不存在,可能已被删除")
return await recall.batches(session, workflow_id, version)
@router.post("/recall/{workflow_id}/{version}/continue")
async def recall_continue(workflow_id: str, version: int, session: AsyncSession = Depends(get_session)
) -> dict[str, Any]:
"""接着算:把还没重算的组再发起一批(每批 recall_limit 组,上限是 0 时按缺省的组数),交回发起了的召回运行和新的批次
情况。都算过了的、这一版已经不是当前发布版的(之后又发布了更新的版本)回 409,detail 写为什么。"""
from app.api.coded import CodedHTTPException
from app.api.deps import current_principal
from app.ledger import recall
if await session.get(Workflow, workflow_id) is None:
raise HTTPException(404, "工作流不存在,可能已被删除")
current = await recall.batches(session, workflow_id, version)
if not current["can_continue"]:
raise CodedHTTPException(409, current.get("continue_reason") or "没有还没重算的组", recall.REFUSED)
launched = await recall.launch(workflow_id, version, batch="manual", actor=current_principal().username)
async with SessionLocal() as fresh:
view = await recall.batches(fresh, workflow_id, version)
return {"launched": launched, **view}
@router.get("/certify-queue")
async def certify_queue(session: AsyncSession = Depends(get_session)) -> dict[str, Any]:
"""可沉淀的问题(认证队列,app/ledger/certify.py):确定性重复的 SQL 形状、问题聚类、被反复用的答案。每项给代表问题、
轮次、(形状的)参数候选和相关口径;点过「不是口径」的不再列。固化走 POST /api/copilot/from-run(freeze: true)。"""
from app.api.deps import current_principal
from app.ledger import certify
return await certify.queue(session, principal=current_principal())
class DismissIn(BaseModel):
item_id: str = Field(min_length=1, max_length=64)
kind: str | None = Field(default=None, max_length=20)
note: str | None = Field(default=None, max_length=500)
@router.post("/certify-queue/dismiss")
async def certify_dismiss(payload: DismissIn, session: AsyncSession = Depends(get_session)) -> dict[str, Any]:
"""「不是口径」:记一条审计(ledger.certify.dismiss),这一项之后不再进队。"""
from app.ledger import certify
if payload.kind is not None and payload.kind not in certify.KINDS:
raise HTTPException(422, "可沉淀的问题只有三种:确定性重复、问题聚类、被反复用的答案")
await certify.dismiss(session, payload.item_id, kind=payload.kind, note=payload.note)
return {"ok": True, "item_id": payload.item_id}