"""口径卡的输入取自「查询区间里这张表没有数据」的列时,指标照实标出来(no_data),出具按缺口降档,复核给出重来的指示。 部署上撞过:一张周报的到访事件表只存最新一天,6 列全是 0,口径卡里 10 个指标是 0 或 0%,周报照写「上周到访人数为 0 人」。 查询工具现在会在结果和快照里记 no_data(data/coverage.py);这里守它往下走的几步: - 口径卡按输入的快照 id + 列认(按列:同一条查询里有数的列不受影响),给指标标 no_data、写明原因;清单文字注明; 运行日志一条 metric_no_data; - 有出具契约的成果按缺口降档; - 复核把 metric_no_data / query_no_data 记成缺口(DEGRADED),并给出「按复核重来」的指示:换一张有这段数据的表再建一次, 大概率就好。 库是现造的 SQLite 文件,表名、列名、数字都是虚构的。 """ from __future__ import annotations import asyncio import sqlite3 import uuid import pytest from sqlalchemy import select from app.core import artifact_store from app.data.engine import engines from app.db.base import SessionLocal from app.db.models import DataSource, Run, RunEvent from app.engine.runner import run_manager WINDOW = "{c} >= strftime('%Y%m%d', 'now', '-{a} days') AND {c} < strftime('%Y%m%d', 'now', '-{b} days')" KPI_SQL = ("SELECT (SELECT SUM(pv) FROM daily_pv WHERE day >= date('now', '-14 days')) AS 客流PV, " "(SELECT COUNT(DISTINCT member_id) FROM gate_events WHERE " + WINDOW.format(c="event_date", a=14, b=7) + ") AS 上周到访人数, (SELECT COUNT(DISTINCT member_id) FROM gate_events WHERE " + WINDOW.format(c="event_date", a=21, b=14) + ") AS 上上周到访人数") WOW = ("round((cell(nodes.q_kpi, 0, '上周到访人数') - cell(nodes.q_kpi, 0, '上上周到访人数')) " "/ cell(nodes.q_kpi, 0, '上上周到访人数'), 4) if cell(nodes.q_kpi, 0, '上上周到访人数') else None") @pytest.fixture(autouse=True) async def engine_up(): await run_manager.setup() yield await run_manager.shutdown() @pytest.fixture async def plant(tmp_path): path = tmp_path / f"{uuid.uuid4().hex[:8]}.db" with sqlite3.connect(path) as db: db.executescript("CREATE TABLE gate_events (event_date TEXT, member_id INTEGER);" "INSERT INTO gate_events VALUES ('20200101', 1), ('20200101', 2);" "CREATE TABLE daily_pv (day TEXT, pv INTEGER);") for back in range(30): db.execute("INSERT INTO daily_pv VALUES (date('now', ?), 100)", (f"-{back} days",)) name = f"plant_{uuid.uuid4().hex[:6]}" async with SessionLocal() as session: row = DataSource(name=name, kind="sqlite", database=str(path), readonly=True, enabled=True) session.add(row) await session.commit() source_id = row.id yield name await engines.invalidate(source_id) def node(nid, ntype, label=None, **config): return {"id": nid, "type": ntype, "position": {"x": 0, "y": 0}, "data": {"label": label or nid, "config": config}} def board(source: str) -> dict: nodes = [ node("start", "input", "开始"), node("q_kpi", "tool", "取数", tool=f"db_query__{source}", args={"sql": KPI_SQL}), node("card", "metrics", "周报口径卡", caliber="周报口径", caliber_version="v1", metrics=[ {"id": "pv", "name": "客流PV", "unit": "次", "expression": "cell(nodes.q_kpi, 0, '客流PV')"}, {"id": "visitors", "name": "上周到访人数", "unit": "人", "expression": "cell(nodes.q_kpi, 0, '上周到访人数')"}, {"id": "wow", "name": "到访周环比", "decimals": 4, "format": "percent_of_ratio", "expression": WOW}, ]), node("out", "output", "成果", fields=[{"name": "指标", "value": "{{ nodes.card.text }}"}], contract={"metrics_from": ["card"], "required": ["pv", "visitors"]}), ] return {"nodes": nodes, "edges": [{"source": a["id"], "target": b["id"]} for a, b in zip(nodes, nodes[1:])]} async def finished(graph: dict) -> tuple[Run, list[RunEvent]]: run = await run_manager.start(graph=graph, input_payload={}) for _ in range(400): await asyncio.sleep(0.05) async with SessionLocal() as session: row = await session.get(Run, run.id) if row.status in ("succeeded", "failed") and row.manifest_seq is not None: assert row.status == "succeeded", row.error async with SessionLocal() as session: events = list((await session.execute( select(RunEvent).where(RunEvent.run_id == run.id).order_by(RunEvent.seq))).scalars()) return row, events raise AssertionError(f"运行 {run.id} 没跑完:{row.status}") def _card(events: list[RunEvent]) -> dict: [event] = [e for e in events if e.type == "node.finished" and e.node_id == "card"] [entry] = event.data["evidence"] return artifact_store.load(entry["artifact"]) async def test_metrics_fed_by_columns_without_data_are_flagged(plant): row, events = await finished(board(plant)) metrics = {m["id"]: m for m in _card(events)["metrics"]} visitors, wow, pv = metrics["visitors"], metrics["wow"], metrics["pv"] assert visitors["value"] == 0 and visitors["no_data"] is True assert visitors["no_data_reason"].startswith("所依据的查询区间里这张表没有数据:") assert "gate_events" in visitors["no_data_reason"] [source] = visitors["no_data_sources"] assert source["column"] == "上周到访人数" and source["table"] == "gate_events" and source["artifact"] # 周环比两个输入都在没有数据的列上:照样标;它本身算不出来(分母为 0),两件事都照实记 assert wow["no_data"] is True and wow["undefined"] is True and wow["value"] is None # 同一条查询里有数的列(客流 PV)不受影响 assert "no_data" not in pv and pv["value"] > 0 lines = row.output["指标"].split("\n") assert "- visitors(上周到访人数)= 0人(这段时间表里没有数据:不要当成 0 解读)" in lines assert not any("pv(客流PV)" in ln and "没有数据" in ln for ln in lines) # 运行日志:查询一条 query_no_data,口径卡一条 metric_no_data(受影响的指标都列出来) codes = [(e.data or {}).get("code") for e in events if e.type == "log"] assert "query_no_data" in codes [logged] = [e.data for e in events if e.type == "log" and (e.data or {}).get("code") == "metric_no_data"] assert logged["level"] == "warn" and logged["metrics"] == ["visitors", "wow"] async def test_the_contract_degrades_on_a_metric_without_data(plant): row, _ = await finished(board(plant)) issuance = row.output["_issuance"] assert issuance["tier"] == "degraded" gaps = [g for g in issuance["gaps"] if "没有数据" in g] assert any("上周到访人数" in g for g in gaps) and any("到访周环比" in g for g in gaps), issuance["gaps"] async def test_review_marks_it_and_asks_for_a_rebuild(plant): from app.chat import turn from app.engine import review as rv row, events = await finished(board(plant)) signals = rv.scan(events, row.output) kinds = {s.kind: s.severity for s in signals} assert kinds.get("metric_no_data") == rv.DEGRADED and kinds.get("query_no_data") == rv.DEGRADED assert rv.should_retry(signals), "换一张有这段数据的表再建一次大概率就好" result = rv.ReviewResult(verdict="annotated", note="…", retry=True, signals=signals, severity=rv.worst(signals)) assert turn.retry_after_review(result.as_dict(), 0) and not turn.retry_after_review(result.as_dict(), 1) # 别的降级信号照旧不重来 other = [rv.Signal(kind="node_retry", detail="重试过一次", severity=rv.DEGRADED)] assert not rv.should_retry(other) assert not turn.retry_after_review(rv.ReviewResult(verdict="annotated", retry=True, signals=other, severity=rv.DEGRADED).as_dict(), 0) async def test_the_writer_is_told_and_given_the_rule(plant): """写作目录里这样的指标、查询格子里这样的列都写清楚;写作规则加一句。""" from app.engine.evidence import NO_DATA_RULE, build_catalog, writer_catalog row, events = await finished(board(plant)) card = _card(events) [query] = [e for e in events if e.type == "node.finished" and e.node_id == "q_kpi"] ledger = [*query.data["evidence"], *[e for e in events if e.type == "node.finished" and e.node_id == "card"][0].data["evidence"]] catalog = build_catalog(nodes={"card": {**card, "artifact": ledger[-1]["artifact"]}}, ledger=ledger, inputs={}) assert catalog["m:visitors"]["no_data"] is True and "no_data" not in catalog["m:pv"] text = writer_catalog(catalog).text visitors = next(ln for ln in text.splitlines() if "[[m:visitors]]" in ln) assert "这段时间表里没有数据" in visitors and "不要当成 0 解读" in visitors assert "这段时间表里没有数据" not in next(ln for ln in text.splitlines() if "[[m:pv]]" in ln) assert "(注意:「上周到访人数」" in text and "引用这几列时照实写这段时间表里没有数据" in text assert "照实写这段时间表里没有数据" in NO_DATA_RULE and "0%" in NO_DATA_RULE async def test_the_report_prompt_carries_the_rule(plant, monkeypatch): from app.engine.evidence import NO_DATA_RULE from tests.test_report_node import script seen = script(monkeypatch, "上周到访人数这段时间表里没有数据。") graph = board(plant) out = graph["nodes"].pop() graph["nodes"] += [node("write", "report", "撰写", instructions="写一段周报"), {**out, "data": {**out["data"], "config": {"fields": [{"name": "答案", "value": "{{ nodes.write.text }}"}]}}}] graph["edges"] = [{"source": a["id"], "target": b["id"]} for a, b in zip(graph["nodes"], graph["nodes"][1:])] await finished(graph) prompt = "\n".join(str(m.content) for m in seen[0]) assert NO_DATA_RULE in prompt and "这段时间表里没有数据" in prompt def test_an_ungated_result_says_its_gaps_on_the_stamp(): """成果没配出具契约时,实例页的印章照实写这些缺口(配了契约的已经按缺口降档)。""" from app.apps.resolve import ungated_notes cards = [{"metrics": [{"id": "visitors", "name": "上周到访人数", "no_data": True}, {"id": "pv", "name": "客流PV"}]}] dual = {"issuance_gaps": ["指标「销售额」无法复算:盲算已达每项最多查询次数(1 次)"]} notes = ungated_notes(cards, dual) assert notes[0].startswith("指标「上周到访人数」所依据的查询区间里表里没有数据") and "没有配出具契约" in notes[0] assert notes[1].startswith("双盲复算有缺口:指标「销售额」无法复算") assert ungated_notes([{"metrics": [{"id": "pv", "name": "客流PV"}]}], None) == []