Files
AgentLab/backend/tests/test_ledger_reconcile.py
T
ThearchyHelios 38d9082994 使用者的原始运行记录只给问数据的运行,对账运行经证据接口读(复核 M3)
复核发现:使用者(开登录后的 user 角色)发起的口径对账运行带 MASK_RESULTS、记在本人名下,run_events_readable 只看这两条,
于是他能经 /api/runs/{id}、/events、/graph 读到对账运行的原始图和事件,里面原样是口径取数节点的 SQL 模板。这一波正是以
「口径的 SQL 可能写着遮罩值」为由不让使用者在问数据里看口径、use_caliber,这条读法绕开了证据面板按遮罩换 SQL 的那一层。

做法:api/deps.run_events_readable 给 user 角色再加一条 origin_kind == "chat"——原始记录、事件、运行时的图、实时流只给他
自己的、查询结果在工具边界遮过的问数据运行(原来的说明里就写着「实际上满足的就是它自己的问数据运行」,这次照实收紧)。
对账运行他仍然能经账本和证据接口读(can_read_run 不变,那里展示的 SQL 按遮罩换过),对账的发起也不变。改了
api/deps.py 的这一个函数(没分给哪个包)和 ledger/reconcile 的模块说明。

测试:开账号密码登录,使用者对自己的问数据答案经接口发起对账,对账运行的 /api/runs/{id}、/events、/graph 都回 403,证据
接口 200;他自己的问数据运行这三样照旧 200。去掉这一条回跑,新用例失败;test_auth_access、test_personal_apps 照旧过。
2026-10-16 18:25:43 +08:00

251 lines
15 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.
"""口径对账:差多少(next-round 02 做法 6,app/ledger/reconcile.py、ledger/derive.reconcile_graph)。
真的跑探索运行(虚构店铺库、mock 模型写报告)、等后台比对做完,再点开对账:
- 动作:有相关口径的查询给「看差多少」;参数是按「同一列、同一运算符」猜的,要先核对(needs_check);
- 不核对直接发起回 409(要核对参数);核对过的参数发起:派生图是确定的、过 validate、出口契约 metrics_from 指向对账卡、
不加报告;跑出来的这次、口径、差值、相对差都对,差值的出处链点得开;GET 账本列出对过的账;统计 E2 数到它;
- 认不出哪一列对哪个指标(口径的指标式不是取数的一格):要人选,选了照记进 origin;
- 多行结果:被引用的行按维度列经合并查询对齐,每行一组对账指标;维度是遮罩列、合并键不唯一时照实拒绝。
数据是虚构店铺库。
"""
from __future__ import annotations
import pytest
from httpx import ASGITransport, AsyncClient
from app.db.base import SessionLocal
from app.db.models import Run
from app.engine.runner import run_manager
from app.engine.schema import GraphSpec, validate_graph
from app.ledger import match as ledger_match
from app.ledger import reconcile as ledger_reconcile
from app.main import app
from tests import ledger_samples as S
BY_REGION = ("SELECT region AS 区域, SUM(amount) AS gmv FROM orders WHERE status = 'paid' "
"AND paid_at >= '{{ input.start }}' AND paid_at < '{{ input.end }}' GROUP BY region ORDER BY region")
BY_REGION_ALL = ("SELECT region AS 区域, SUM(amount) AS gmv FROM orders WHERE paid_at >= '2026-09-01' "
"AND paid_at < '2026-09-08' GROUP BY region ORDER BY region")
CHECKED = {"start": "2026-09-01", "end": "2026-09-08"}
@pytest.fixture(autouse=True)
async def engine_up(tmp_path):
await run_manager.setup()
path = tmp_path / "shop.db"
await S.fresh_source(path)
yield path
await ledger_match.drain()
await run_manager.shutdown()
@pytest.fixture
async def client():
async with AsyncClient(transport=ASGITransport(app=app), base_url="http://test") as c:
yield c
async def ledger_of(client, run_id: str) -> dict:
res = await client.get(f"/api/runs/{run_id}/ledger")
assert res.status_code == 200, res.text
return res.json()
async def card_of(run_id: str, node_id: str = "reconcile") -> dict:
from app.api.evidence import _sealed, _sealed_cards
sealed = await _sealed(run_id)
[(_, card)] = [(a, c) for n, a, c in _sealed_cards(sealed) if n == node_id]
return {m["id"]: m for m in card["metrics"]}
async def reconcile(client, run_id: str, body: dict) -> Run:
res = await client.post(f"/api/ledger/answers/{run_id}/reconcile", json=body)
assert res.status_code == 201, res.text
return await S.wait(res.json()["run_id"])
async def test_reconcile_the_refunds_answer_against_the_weekly_caliber(client, monkeypatch):
wf = await S.publish()
row = await S.explore(monkeypatch, S.WITH_REFUNDS)
body = await ledger_of(client, row.id)
[action] = [a for a in body["actions"] if a["kind"] == "reconcile"]
assert action["workflow_id"] == wf and action["version"] == 1 and action["node_id"] == "fetch"
assert action["needs_check"] is True and action["check_reason"] == ledger_reconcile.CHECK_REASON
assert action["binding"] == [{"field": "end", "value": "2026-09-08"}, {"field": "start", "value": "2026-09-01"}]
query = action["query"]
# 参数是猜的:不核对就发起,照实拒绝
res = await client.post(f"/api/ledger/answers/{row.id}/reconcile", json={"query": query, "workflow_id": wf,
"version": 1})
assert res.status_code == 409 and res.json()["code"] == ledger_reconcile.CHECK_INPUTS
assert res.json()["detail"] == ledger_reconcile.CONFIRM_FIRST
done = await reconcile(client, row.id, {"query": query, "workflow_id": wf, "version": 1, "inputs": CHECKED,
"confirm": True})
assert done.status == "succeeded", done.error
assert done.origin_kind == "reconcile" and done.run_class == "exploratory" and done.workflow_id is None
origin = done.origin
assert origin["source_run"] == row.id and origin["query"] == query and origin["inputs_from"] == "checked"
assert origin["pairs"] == [{"ours": "gmv", "metric": "gmv"}] and origin["caliber"] == f"{wf}@1"
assert done.input == {"start": "2026-09-01", "end": "2026-09-08", "store": "all"}
graph = done.graph
ids = [n["id"] for n in graph["nodes"]]
assert ids == ["derived_in", "ours_q1", "fetch", "card", "reconcile", "derived_out"]
nodes = {n["id"]: n for n in graph["nodes"]}
assert nodes["ours_q1"]["data"]["config"]["args"]["sql"] == S.WITH_REFUNDS
assert "{{ input.start }}" in nodes["fetch"]["data"]["config"]["args"]["sql"], "口径的取数保留模板,运行时按入口渲染"
assert nodes["derived_out"]["data"]["config"]["contract"] == {"metrics_from": "reconcile"}
assert not [n for n in graph["nodes"] if n["type"] in ("report", "llm", "agent")]
assert not [i for i in validate_graph(GraphSpec.model_validate(graph)).issues if i.level == "error"]
metrics = await card_of(done.id)
assert metrics["ours_0"]["value"] == 221.0 and metrics["cal_0"]["value"] == 170.5
assert metrics["diff_0"]["value"] == 50.5 and metrics["rel_0"]["value"] == 0.2962
assert metrics["rel_0"]["rendered"] == "29.62%" and metrics["diff_0"]["rendered"] == "50.50元"
# 三个数各自点得开:差值的出处链追到两边的查询
res = await client.get(f"/api/runs/{done.id}/evidence/metrics/reconcile/diff_0")
assert res.status_code == 200, res.text
inputs = [s for s in res.json()["chain"] if s["step"] == "input"]
assert {i["node_id"] for i in inputs} == {"ours_q1", "card"}
listed = (await ledger_of(client, row.id))["reconciled"]
assert [r["run_id"] for r in listed] == [done.id] and listed[0]["inputs_from"] == "checked"
# 统计 E2:对账运行数到了(它不进使用统计,E2 专门数它)
stats = (await client.get("/api/ledger/stats")).json()["e2"]
assert stats["launched"] >= 1 and stats["succeeded"] >= 1
async def test_the_derived_graph_is_deterministic():
from app.ledger import derive
wf = await S.publish()
async with SessionLocal() as session:
pinned = await derive.load_pinned(session, wf, 1, "card")
plan = ledger_reconcile.Plan(pairs=[ledger_reconcile.Pair(ours="gmv", metric="gmv", index=0, name="销售额",
unit="元", decimals=2)])
a = ledger_reconcile.build(pinned, sql=S.WITH_REFUNDS, tool=S.tool(), plan=plan, inputs=CHECKED, fetch="fetch")
b = ledger_reconcile.build(pinned, sql=S.WITH_REFUNDS, tool=S.tool(), plan=plan, inputs=CHECKED, fetch="fetch")
assert a.graph == b.graph and a.kind == "reconcile"
assert a.origin["from"][0] == {"kind": "executed", "node_id": "ours_q1"} and a.origin["from"][1]["node_id"] == "card"
async def test_metrics_that_are_not_a_plain_cell_need_a_person_to_pick(client, monkeypatch):
wf = await S.publish(graph=S.caliber_graph(expression="cell(nodes.fetch, 0, 'gmv') * 1"))
row = await S.explore(monkeypatch, S.WITH_REFUNDS)
[action] = [a for a in (await ledger_of(client, row.id))["actions"] if a["kind"] == "reconcile"]
body = {"query": action["query"], "workflow_id": wf, "version": 1, "inputs": CHECKED, "confirm": True}
res = await client.post(f"/api/ledger/answers/{row.id}/reconcile", json=body)
assert res.status_code == 409 and res.json()["code"] == ledger_reconcile.NEEDS_PAIRS
bad = await client.post(f"/api/ledger/answers/{row.id}/reconcile",
json={**body, "pairs": [{"ours": "nope", "metric": "gmv"}]})
assert bad.status_code == 422 and "没有列「nope」" in bad.json()["detail"]
done = await reconcile(client, row.id, {**body, "pairs": [{"ours": "gmv", "metric": "gmv"}]})
assert done.status == "succeeded", done.error
assert done.origin["pairs_chosen"] is True
assert (await card_of(done.id))["diff_0"]["value"] == 50.5
async def test_multi_row_answers_are_aligned_by_their_dimensions(client, monkeypatch):
wf = await S.publish(graph=S.caliber_graph(BY_REGION))
row = await S.explore(monkeypatch, BY_REGION_ALL, text="东区 [[v:Q1.r0.gmv]],西区 [[v:Q1.r1.gmv]]。")
[action] = [a for a in (await ledger_of(client, row.id))["actions"] if a["kind"] == "reconcile"]
done = await reconcile(client, row.id, {"query": action["query"], "workflow_id": wf, "version": 1,
"inputs": CHECKED, "confirm": True})
assert done.status == "succeeded", done.error
assert done.origin["rows"] == [0, 1] and done.origin["pairs"] == [{"ours": "gmv", "caliber": "gmv"}]
align = next(n for n in done.graph["nodes"] if n["id"] == "align")
assert align["type"] == "merge" and align["data"]["config"]["inputs"] == {"o": "ours_q1", "c": "fetch"}
metrics = await card_of(done.id)
assert metrics["ours_0_0"]["name"] == "区域=east · gmv · 这次"
assert (metrics["ours_0_0"]["value"], metrics["cal_0_0"]["value"], metrics["diff_0_0"]["value"]) == (163.0, 150.5, 12.5)
assert (metrics["ours_1_0"]["value"], metrics["cal_1_0"]["value"], metrics["diff_1_0"]["value"]) == (58.0, 20.0, 38.0)
async def test_masked_dimensions_and_duplicate_keys_are_refused(client, monkeypatch, engine_up):
wf = await S.publish(graph=S.caliber_graph(BY_REGION))
duplicated = ("SELECT region AS 区域, amount AS gmv FROM orders WHERE paid_at >= '2026-09-01' "
"AND paid_at < '2026-09-08' ORDER BY id")
row = await S.explore(monkeypatch, duplicated, text="第一单 [[v:Q1.r0.gmv]],第二单 [[v:Q1.r1.gmv]]。")
[action] = [a for a in (await ledger_of(client, row.id))["actions"] if a["kind"] == "reconcile"]
res = await client.post(f"/api/ledger/answers/{row.id}/reconcile",
json={"query": action["query"], "workflow_id": wf, "version": 1, "inputs": CHECKED,
"confirm": True})
assert res.status_code == 409 and "有重复的行" in res.json()["detail"]
await S.set_source(engine_up, options={"mask_columns": ["region"]})
row = await S.explore(monkeypatch, BY_REGION_ALL, text="东区 [[v:Q1.r0.gmv]],西区 [[v:Q1.r1.gmv]]。")
[action] = [a for a in (await ledger_of(client, row.id))["actions"] if a["kind"] == "reconcile"]
res = await client.post(f"/api/ledger/answers/{row.id}/reconcile",
json={"query": action["query"], "workflow_id": wf, "version": 1, "inputs": CHECKED,
"confirm": True})
assert res.status_code == 409 and "设置了遮罩" in res.json()["detail"]
def test_align_sql_keeps_the_referenced_rows_in_order():
pairs = [ledger_reconcile.Pair(ours="gmv", column="销售额")]
sql = ledger_reconcile.align_sql([("区域", "区域")], pairs, [("west",), ("o'east",)])
assert sql == ('SELECT o."区域" AS "区域", o."gmv" AS "ours_0", c."销售额" AS "cal_0" FROM o LEFT JOIN c ON '
'o."区域" = c."区域" WHERE (o."区域" = \'west\') OR (o."区域" = \'o\'\'east\') '
'ORDER BY CASE WHEN (o."区域" = \'west\') THEN 0 WHEN (o."区域" = \'o\'\'east\') THEN 1 END')
async def test_users_reconcile_their_own_answers_read_only_and_masked(client, monkeypatch):
"""拍板 L5:对账(只读的探索运行)使用者都可以,带上问数据同一套 MASK_RESULTS / READ_ONLY;别人的答案读不到就不给对。"""
from app.api.deps import Principal
from app.tools.datasource import MASK_RESULTS_OPTION, READ_ONLY_OPTION
wf = await S.publish()
row = await S.explore(monkeypatch, S.WITH_REFUNDS)
async with SessionLocal() as session:
(await session.get(Run, row.id)).started_by = "carol"
await session.commit()
[action] = [a for a in (await ledger_of(client, row.id))["actions"] if a["kind"] == "reconcile"]
req = ledger_reconcile.Request(query=action["query"], workflow_id=wf, version=1, inputs=CHECKED, confirm=True)
dave = Principal(username="dave", role="user", verified=True, mode="local")
with pytest.raises(ledger_reconcile.ReconcileRefused) as e:
await ledger_reconcile.reconcile(row.id, req, principal=dave)
assert e.value.status_code == 403
carol = Principal(username="carol", role="user", verified=True, mode="local")
run = await ledger_reconcile.reconcile(row.id, req, principal=carol)
assert run.started_by == "carol"
assert run.options[MASK_RESULTS_OPTION] is True and run.options[READ_ONLY_OPTION] is True
done = await S.wait(run.id)
assert done.status == "succeeded", done.error
async def test_users_read_their_reconcile_runs_only_through_the_evidence(client, monkeypatch):
"""使用者发起的对账运行带 MASK_RESULTS、记在他名下,但派生图里有口径取数节点的 SQL 模板(口径的 SQL 可能写着遮罩值,
所以问数据里使用者不看口径、不能 use_caliber):运行的原始记录、事件、运行时的图(/api/runs/{id}、/events、/graph)只给
他自己的问数据运行,对账运行经证据接口读(那里按遮罩渲染)。"""
from app.tools.datasource import MASK_RESULTS_OPTION
from tests.fakes import auth_env
wf = await S.publish()
row = await S.explore(monkeypatch, S.WITH_REFUNDS, origin="chat")
# 当作开登录后使用者 carol 问的:问数据运行在工具边界遮
async with SessionLocal() as session:
source = await session.get(Run, row.id)
source.started_by, source.options = "carol", {MASK_RESULTS_OPTION: True}
await session.commit()
[action] = [a for a in (await ledger_of(client, row.id))["actions"] if a["kind"] == "reconcile"]
await auth_env.reset()
try:
admin = await auth_env.turn_on_local()
await auth_env.make_user("carol", "user")
assert (await admin.put("/api/auth/settings", json={"user_can_chat": True})).status_code == 200
carol = await auth_env.login("carol")
res = await carol.post(f"/api/ledger/answers/{row.id}/reconcile", json={
"query": action["query"], "workflow_id": wf, "version": 1, "inputs": CHECKED, "confirm": True})
assert res.status_code == 201, res.text
done = await S.wait(res.json()["run_id"])
assert done.status == "succeeded" and done.started_by == "carol" and done.options[MASK_RESULTS_OPTION] is True
for rest in ("", "/events", "/graph"):
assert (await carol.get(f"/api/runs/{done.id}{rest}")).status_code == 403, rest
assert (await carol.get(f"/api/runs/{row.id}{rest}")).status_code == 200, f"自己的问数据运行照旧:{rest}"
assert (await carol.get(f"/api/runs/{done.id}/evidence")).status_code == 200
finally:
await auth_env.reset()