mirror of
https://github.com/ThearchyHelios/AgentLab.git
synced 2026-11-17 22:31:57 +00:00
数据目录阶段 2B:助手自查接上 SQL 检查——工具节点里写死的 SQL 查出 error 进修正轮,warning 和 info 留在交付的问题清单
This commit is contained in:
@@ -16,6 +16,7 @@ from sqlalchemy.ext.asyncio import AsyncSession
|
||||
from app.api.coded import DATASOURCE_SCOPE_EMPTY, CodedHTTPException
|
||||
# 起别名:本文件底下有个叫 explain 的接口,generate 里又有个局部变量叫 raw,同名会互相盖掉
|
||||
from app.core.errors import explain as explain_error, graph_error, not_configured, raw as raw_error
|
||||
from app.data.sqlcheck import CHECK_CODES
|
||||
from app.db.base import get_session
|
||||
from app.db.models import Workflow
|
||||
from app.engine.layout import CORRIDOR_MIN, MIN_ROW_GAP, NODE_W, _height, auto_layout
|
||||
@@ -1081,6 +1082,7 @@ def _blocking_issues(
|
||||
nodes: dict[str, dict[str, Any]], edges: list[dict[str, Any]], scope: set[str] | None = None,
|
||||
*, sources: list[Any] | None = None, baseline: list[dict[str, Any]] | None = None,
|
||||
level: str | None = None, defaults: dict[str, Any] | None = None,
|
||||
checkers: dict[str, Any] | None = None,
|
||||
) -> list[dict[str, Any]] | None:
|
||||
"""按运行时同一套校验,挑出会挡住运行的问题。图本身不成形时返回 None。
|
||||
|
||||
@@ -1098,6 +1100,9 @@ def _blocking_issues(
|
||||
|
||||
defaults 是全图默认(graph.defaults):报告撰写节点的 numbers / claims、审批策略都跟随它,不带上的话
|
||||
自查看到的问题和真正的门禁对不上。
|
||||
|
||||
checkers 是各数据源基于数据目录的 SQL 检查器(data/sqlcheck.load_checkers):调用工具节点里写死的 SQL
|
||||
查出 error 级的问题(一对多关联后重复计算、存量跨期求和)也算挡住,交回模型改;warning、info 不挡。
|
||||
"""
|
||||
try:
|
||||
spec = GraphSpec.model_validate({"nodes": list(nodes.values()), "edges": edges,
|
||||
@@ -1106,7 +1111,7 @@ def _blocking_issues(
|
||||
return None
|
||||
out = ([i.model_dump() for i in validate_graph(spec).issues if i.level == "error"]
|
||||
+ _issue_dicts(scope_issues(list(nodes.values()), scope), "datasource_out_of_scope")
|
||||
+ authored_issues(spec, sources or [], baseline))
|
||||
+ [i for i in authored_issues(spec, sources or [], baseline, checkers=checkers) if i["level"] == "error"])
|
||||
if level:
|
||||
from app.engine.governance import lint_for_publish
|
||||
|
||||
@@ -1181,7 +1186,7 @@ def _unsourced_reports(spec: GraphSpec, changed: set[str]) -> list[dict[str, Any
|
||||
|
||||
|
||||
def authored_issues(spec: GraphSpec, sources: list[Any], baseline: list[dict[str, Any]] | None = None,
|
||||
) -> list[dict[str, Any]]:
|
||||
*, checkers: dict[str, Any] | None = None) -> list[dict[str, Any]]:
|
||||
"""助手搭图时要打回、运行时却不挡的写法,按 error 交回模型改:
|
||||
|
||||
- sql_unknown_column:调用工具节点的 SQL 里有数据源结构(schema_cache)里查不到的列——凭空
|
||||
@@ -1189,6 +1194,11 @@ def authored_issues(spec: GraphSpec, sources: list[Any], baseline: list[dict[str
|
||||
- parse_model_text:整形节点按 JSON 解析 agent / llm 写的文字(validate 里只是 warning:旧图照跑)
|
||||
- report_no_source:报告撰写节点的上游没有任何证据来源(运行时只给一条警告,报告里的数全没有出处)。
|
||||
metrics_from 指向不是口径卡的节点,validate 本来就报 error,自查照样交回去
|
||||
|
||||
另外,给了 checkers(各数据源基于数据目录的 SQL 检查器)时,调用工具节点里写死的 SQL 再对照目录查一遍
|
||||
(code 是 data/sqlcheck.CHECK_CODES 里的规则编号),级别照检查结果:error 交回模型改,warning、info
|
||||
由调用方放进交付的问题清单。列名都对不上的 SQL 先改列名,这一轮不叠加目录检查。Agent 节点的 SQL 要到
|
||||
运行时才写出来,这里拿不到,由数据源查询工具在执行后检查。
|
||||
"""
|
||||
changed = _changed(spec, baseline)
|
||||
out: list[dict[str, Any]] = []
|
||||
@@ -1212,6 +1222,7 @@ def authored_issues(spec: GraphSpec, sources: list[Any], baseline: list[dict[str
|
||||
continue
|
||||
found = sql_unknown_columns(sql, source)
|
||||
if not found:
|
||||
out += _catalog_issues(node.id, sql, (checkers or {}).get(source.name))
|
||||
continue
|
||||
unknown, known = found
|
||||
listed = "、".join(known[:20]) + (f" 等 {len(known)} 列" if len(known) > 20 else "")
|
||||
@@ -1225,6 +1236,14 @@ def authored_issues(spec: GraphSpec, sources: list[Any], baseline: list[dict[str
|
||||
return out
|
||||
|
||||
|
||||
def _catalog_issues(node_id: str, sql: str, checker: Any) -> list[dict[str, Any]]:
|
||||
"""一个调用工具节点的 SQL 对照数据目录查出来的问题,形状和别的自查问题一样(node_id、field 都在),
|
||||
另带 table、column、relation_id、sql_excerpt 给界面定位。"""
|
||||
if checker is None:
|
||||
return []
|
||||
return [{**check, "node_id": node_id, "edge_id": None, "field": "args.sql"} for check in checker.check_dicts(sql)]
|
||||
|
||||
|
||||
# --------------------------------------------------------------------------
|
||||
# SQL 里的列名对照数据源结构
|
||||
#
|
||||
@@ -1599,10 +1618,14 @@ async def generate_stream(payload: GenerateIn, session: AsyncSession = Depends(g
|
||||
"""
|
||||
from fastapi.responses import StreamingResponse
|
||||
|
||||
from app.data.sqlcheck import load_checkers
|
||||
|
||||
sources = await _sources(session, payload.datasource_ids)
|
||||
scope = {r.name for r in sources} if payload.datasource_ids else None
|
||||
tool_list = _tool_catalog(sources)
|
||||
datasources = _datasource_section(sources) + _scope_note(sources, scope is not None)
|
||||
# 基于数据目录的 SQL 检查器:在开流之前读好目录(流里不再碰这个会话),自查和收尾共用
|
||||
checkers = await load_checkers(session, sources)
|
||||
system = (
|
||||
"你是一个 agent 工作流编排专家。根据用户需求,以操作流的方式逐步搭出一张可执行的工作流图。\n\n"
|
||||
f"{NODE_REFERENCE}\n\n可用工具:\n{tool_list}{datasources}\n\n"
|
||||
@@ -1721,7 +1744,7 @@ async def generate_stream(payload: GenerateIn, session: AsyncSession = Depends(g
|
||||
# 循环 / 分支条件里套了 {{ }}),前面的步骤白跑,报错还停在半路
|
||||
repaired = 0
|
||||
for round_no in range(1, _SELF_CHECK_ROUNDS + 1):
|
||||
errors = _blocking_issues(nodes, edges, scope, sources=sources, baseline=baseline)
|
||||
errors = _blocking_issues(nodes, edges, scope, sources=sources, baseline=baseline, checkers=checkers)
|
||||
if not errors:
|
||||
break
|
||||
yield _sse({"op": "check", "status": "repairing", "round": round_no, "issues": errors})
|
||||
@@ -1739,7 +1762,7 @@ async def generate_stream(payload: GenerateIn, session: AsyncSession = Depends(g
|
||||
"message": f"自查修正失败:{explain_error(e)[0]}", "detail": raw_error(e)})
|
||||
break
|
||||
repaired = round_no
|
||||
remaining = _blocking_issues(nodes, edges, scope, sources=sources, baseline=baseline)
|
||||
remaining = _blocking_issues(nodes, edges, scope, sources=sources, baseline=baseline, checkers=checkers)
|
||||
# 工具绑定变化不挡运行,但要在自查这一步就说出来:工具被改没了的图照样
|
||||
# 能跑,只是跑出来的是模型「假设」查过库的答案
|
||||
changes = tool_changes(baseline, list(nodes.values()))
|
||||
@@ -1763,9 +1786,11 @@ async def generate_stream(payload: GenerateIn, session: AsyncSession = Depends(g
|
||||
issues = [i.model_dump() for i in validate_graph(spec).issues if i.level != "info"]
|
||||
issues += _issue_dicts(scope_issues(list(nodes.values()), scope),
|
||||
"datasource_out_of_scope")
|
||||
# 自查没改好的猜列名留在画布上:运行时不挡(结构缓存可能旧了),但要看得见
|
||||
issues += [{**i, "level": "warning"} for i in authored_issues(spec, sources, baseline)
|
||||
if i["code"] == "sql_unknown_column"]
|
||||
# 自查没改好的猜列名留在画布上:运行时不挡(结构缓存可能旧了),但要看得见。
|
||||
# 对照数据目录的 SQL 检查照原级别列出:warning、info 本来就不返工,改不好的 error 也在这里看得见
|
||||
authored = authored_issues(spec, sources, baseline, checkers=checkers)
|
||||
issues += [{**i, "level": "warning"} for i in authored if i["code"] == "sql_unknown_column"]
|
||||
issues += [i for i in authored if i["code"] in CHECK_CODES]
|
||||
# 跳过的节点要说出来,不能安静地少一步。code 给前端认:这一类要单独
|
||||
# 提示「少了一步」,不能和普通校验警告混在一起
|
||||
issues += [
|
||||
|
||||
@@ -955,5 +955,94 @@ def _known_code(literal: tuple[str, bool], codes: Mapping[str, Any]) -> bool:
|
||||
return number is not None and any(_number(k) == number for k in keys)
|
||||
|
||||
|
||||
# ==========================================================================
|
||||
# 工作流里写死的 SQL、从库里读目录
|
||||
# ==========================================================================
|
||||
|
||||
|
||||
def tool_query(node: Any) -> tuple[str, str] | None:
|
||||
"""调用工具节点里写死的查询:(数据源名, SQL)。不是数据源查询工具、SQL 不是字符串的返回 None。
|
||||
|
||||
Agent 节点的 SQL 要到运行时才由模型写出来,静态拿不到,这里不管;它们由数据源查询工具在执行后检查。
|
||||
"""
|
||||
from app.tools.datasource import QUERY_PREFIX
|
||||
|
||||
if str(getattr(node, "type", "")) != "tool":
|
||||
return None
|
||||
config = getattr(node, "config", None) or {}
|
||||
tool = str(config.get("tool") or "")
|
||||
args = config.get("args")
|
||||
sql = args.get("sql") if isinstance(args, Mapping) else None
|
||||
if not tool.startswith(QUERY_PREFIX) or not isinstance(sql, str) or not sql.strip():
|
||||
return None
|
||||
return tool[len(QUERY_PREFIX):], sql
|
||||
|
||||
|
||||
def graph_sources(nodes: Iterable[Any]) -> set[str]:
|
||||
"""一张图里调用工具节点写死查询用到的数据源名。"""
|
||||
return {found[0] for node in nodes if (found := tool_query(node))}
|
||||
|
||||
|
||||
def graph_checks(nodes: Iterable[Any], checkers: Mapping[str, SqlChecker], *,
|
||||
only: set[str] | None = None) -> list[tuple[Any, SqlCheck]]:
|
||||
"""一张图里调用工具节点写死的 SQL 逐个检查:[(节点, 结果)]。only 给了就只查这些节点。"""
|
||||
out: list[tuple[Any, SqlCheck]] = []
|
||||
for node in nodes:
|
||||
if only is not None and getattr(node, "id", None) not in only:
|
||||
continue
|
||||
found = tool_query(node)
|
||||
checker = checkers.get(found[0]) if found else None
|
||||
if checker is not None:
|
||||
out += [(node, check) for check in checker.check(found[1])]
|
||||
return out
|
||||
|
||||
|
||||
async def load_checkers(session: Any, sources: Iterable[Any]) -> dict[str, SqlChecker]:
|
||||
"""给一批数据源(DataSource 行)各建一个检查器:{源名: SqlChecker}。没有目录的源不建。
|
||||
|
||||
一次查询读出这些源的全部目录。读不到就当没有目录(返回空 dict):检查是附加的,不能挡住自查和发布本身。
|
||||
"""
|
||||
from sqlalchemy import select
|
||||
|
||||
from app.db.models import CatalogNote
|
||||
|
||||
rows = [s for s in sources if getattr(s, "id", None) and dialect_of(getattr(s, "kind", None))]
|
||||
if not rows:
|
||||
return {}
|
||||
try:
|
||||
notes = (await session.execute(
|
||||
select(CatalogNote.source_id, CatalogNote.table_name, CatalogNote.notes)
|
||||
.where(CatalogNote.source_id.in_([r.id for r in rows]))
|
||||
)).all()
|
||||
except Exception: # noqa: BLE001
|
||||
logger.exception("读取数据目录失败,本次不做基于目录的 SQL 检查")
|
||||
return {}
|
||||
by_source: dict[str, dict[str, Any]] = {}
|
||||
for source_id, table, value in notes:
|
||||
if isinstance(value, dict) and value:
|
||||
by_source.setdefault(source_id, {})[table] = value
|
||||
return {r.name: SqlChecker(kind=r.kind, schema_cache=r.schema_cache, notes=by_source[r.id], source_name=r.name)
|
||||
for r in rows if r.id in by_source}
|
||||
|
||||
|
||||
async def load_checkers_by_name(session: Any, names: Iterable[str]) -> dict[str, SqlChecker]:
|
||||
"""按数据源名建检查器(只看启用着的源)。发布门禁用:图里写的是源名。"""
|
||||
from sqlalchemy import select
|
||||
|
||||
from app.db.models import DataSource
|
||||
|
||||
wanted = sorted({n for n in names if n})
|
||||
if not wanted:
|
||||
return {}
|
||||
try:
|
||||
rows = list((await session.execute(
|
||||
select(DataSource).where(DataSource.name.in_(wanted), DataSource.enabled.is_(True))
|
||||
)).scalars())
|
||||
except Exception: # noqa: BLE001
|
||||
logger.exception("读取数据源失败,本次不做基于目录的 SQL 检查")
|
||||
return {}
|
||||
return await load_checkers(session, rows)
|
||||
|
||||
|
||||
__all__ = ["CHECK_CODES", "LEVELS", "MAX_CHECKS", "MAX_SQL_CHARS", "SqlCheck", "SqlChecker", "check_sql",
|
||||
"dialect_of"]
|
||||
"dialect_of", "graph_checks", "graph_sources", "load_checkers", "load_checkers_by_name", "tool_query"]
|
||||
|
||||
@@ -0,0 +1,245 @@
|
||||
"""助手搭完图的自查接上基于数据目录的 SQL 检查(data/sqlcheck.py)。
|
||||
|
||||
- 调用工具节点里写死的 SQL 走现有的自查:error 级(一对多关联后重复计算、存量跨期求和)交回模型改,进修正轮;
|
||||
warning、info 不返工,放进 final 的问题清单。
|
||||
- Agent 节点的 SQL 要到运行时才由模型写出来,静态拿不到,不查(运行时由数据源查询工具检查)。
|
||||
- 这一轮没碰的节点不查:用户自己写的 SQL 不替他改(和列名核对同一个规矩)。
|
||||
- 没有目录的数据源什么都不查。
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
import json
|
||||
import uuid
|
||||
from types import SimpleNamespace
|
||||
|
||||
import pytest
|
||||
from httpx import ASGITransport, AsyncClient
|
||||
|
||||
from app.data import catalog
|
||||
from app.data.engine import engines
|
||||
from app.data.introspect import introspect
|
||||
from app.data.sqlcheck import SqlChecker
|
||||
from tests.fixtures.catalog import scenic_notes
|
||||
|
||||
FANOUT_SQL = ("SELECT SUM(o.total_amount) AS gmv FROM orders o JOIN order_items i ON i.order_id = o.id "
|
||||
"WHERE o.status = 1")
|
||||
#: 改好的:对明细求和;没按订单状态筛,留一条 warning
|
||||
FIXED_SQL = "SELECT SUM(i.amount) AS gmv FROM orders o JOIN order_items i ON i.order_id = o.id"
|
||||
CLEAN_SQL = "SELECT SUM(i.amount) AS gmv FROM orders o JOIN order_items i ON i.order_id = o.id WHERE o.status = 1"
|
||||
|
||||
_CACHE: dict[str, dict] = {}
|
||||
|
||||
|
||||
async def _schema(path: str) -> dict:
|
||||
if path not in _CACHE:
|
||||
probe = SimpleNamespace(id=f"probe-{uuid.uuid4().hex[:6]}", name="scenic", kind="sqlite", database=path,
|
||||
host=None, port=None, username=None, password=None, options={}, readonly=True,
|
||||
description="", schema_cache={}, origin="manual")
|
||||
try:
|
||||
_CACHE[path] = await introspect(probe)
|
||||
finally:
|
||||
await engines.invalidate(probe.id)
|
||||
return _CACHE[path]
|
||||
|
||||
|
||||
def _spec(*nodes):
|
||||
from app.engine.schema import GraphSpec
|
||||
|
||||
return GraphSpec.model_validate({"nodes": list(nodes), "edges": [
|
||||
{"source": a["id"], "target": b["id"]} for a, b in zip(nodes, nodes[1:])]})
|
||||
|
||||
|
||||
def _node(nid, ntype, **config):
|
||||
return {"id": nid, "type": ntype, "data": {"label": nid, "config": config}}
|
||||
|
||||
|
||||
def _fetch(sql, source="scenic"):
|
||||
return _node("fetch", "tool", tool=f"db_query__{source}", args={"sql": sql})
|
||||
|
||||
|
||||
async def _checkers(scenic_db, status="confirmed"):
|
||||
cache = await _schema(scenic_db)
|
||||
source = SimpleNamespace(id="s1", name="scenic", kind="sqlite", schema_cache=cache)
|
||||
checker = SqlChecker(kind="sqlite", schema_cache=cache, notes=scenic_notes.notes(status), source_name="scenic")
|
||||
return source, {"scenic": checker}
|
||||
|
||||
|
||||
# --------------------------------------------------------------------------
|
||||
# authored_issues / _blocking_issues
|
||||
# --------------------------------------------------------------------------
|
||||
|
||||
|
||||
async def test_sql_check_errors_block_and_warnings_do_not(scenic_db):
|
||||
from app.api.copilot import _blocking_issues, authored_issues
|
||||
|
||||
source, checkers = await _checkers(scenic_db)
|
||||
spec = _spec(_node("start", "input"), _fetch(FANOUT_SQL), _node("out", "output", fields=[]))
|
||||
[issue] = authored_issues(spec, [source], checkers=checkers)
|
||||
assert issue["code"] == "fanout_sum" and issue["level"] == "error"
|
||||
assert issue["node_id"] == "fetch" and issue["field"] == "args.sql"
|
||||
assert issue["table"] == "orders" and issue["column"] == "total_amount" and issue["sql_excerpt"]
|
||||
assert issue["message"].startswith("「订单」关联「订单明细」是一对多") and "order_items" in issue["for_model"]
|
||||
nodes = {n.id: n.model_dump(mode="json") for n in spec.nodes}
|
||||
assert [i["code"] for i in _blocking_issues(nodes, [], sources=[source], checkers=checkers)] == ["fanout_sum"]
|
||||
|
||||
# 只有 warning:不挡(不返工),但 authored_issues 照样带出来,收尾时放进 final
|
||||
warn = _spec(_node("start", "input"), _fetch(FIXED_SQL), _node("out", "output", fields=[]))
|
||||
[w] = authored_issues(warn, [source], checkers=checkers)
|
||||
assert (w["code"], w["level"]) == ("missing_valid_filter", "warning")
|
||||
nodes = {n.id: n.model_dump(mode="json") for n in warn.nodes}
|
||||
assert _blocking_issues(nodes, [], sources=[source], checkers=checkers) == []
|
||||
|
||||
# 依据只是推断:降为 info,也不挡
|
||||
_, proposed = await _checkers(scenic_db, "proposed")
|
||||
assert {i["level"] for i in authored_issues(spec, [source], checkers=proposed)} == {"info"}
|
||||
# 不给检查器(没有目录):和以前一样什么都不查
|
||||
assert authored_issues(spec, [source]) == []
|
||||
|
||||
|
||||
async def test_unknown_columns_come_first(scenic_db):
|
||||
"""列名都对不上的 SQL 先改列名:这时不再叠加目录检查的结果,免得一轮里塞给模型两套说法。"""
|
||||
from app.api.copilot import authored_issues
|
||||
|
||||
source, checkers = await _checkers(scenic_db)
|
||||
bad = ("SELECT SUM(o.total_amt) AS gmv FROM orders o JOIN order_items i ON i.order_id = o.id "
|
||||
"WHERE o.status = 1")
|
||||
spec = _spec(_node("start", "input"), _fetch(bad), _node("out", "output", fields=[]))
|
||||
assert [i["code"] for i in authored_issues(spec, [source], checkers=checkers)] == ["sql_unknown_column"]
|
||||
|
||||
|
||||
async def test_agent_prompts_and_untouched_nodes_are_not_checked(scenic_db):
|
||||
from app.api.copilot import authored_issues
|
||||
|
||||
source, checkers = await _checkers(scenic_db)
|
||||
agent = _node("ask", "agent", prompt=f"用这条 SQL 查:{FANOUT_SQL}", tools=["db_query__scenic"])
|
||||
spec = _spec(_node("start", "input"), agent, _node("out", "output", fields=[]))
|
||||
assert authored_issues(spec, [source], checkers=checkers) == []
|
||||
# 用户自己写的、这一轮没改的 SQL 不替他改
|
||||
fetch = _fetch(FANOUT_SQL)
|
||||
untouched = _spec(_node("start", "input"), fetch, _node("out", "output", fields=[]))
|
||||
assert authored_issues(untouched, [source], baseline=[fetch], checkers=checkers) == []
|
||||
# 模板占位照常检查其余部分
|
||||
templated = _fetch("SELECT SUM(o.total_amount) FROM orders o JOIN order_items i ON i.order_id = o.id "
|
||||
"WHERE o.status = 1 AND o.ordered_at >= '{{ input.since }}'")
|
||||
out = authored_issues(_spec(_node("start", "input"), templated, _node("out", "output", fields=[])), [source],
|
||||
checkers=checkers)
|
||||
assert [i["code"] for i in out] == ["fanout_sum"]
|
||||
|
||||
|
||||
# --------------------------------------------------------------------------
|
||||
# 端到端:假模型搭图,error 进修正轮,warning 留在 final
|
||||
# --------------------------------------------------------------------------
|
||||
|
||||
|
||||
class _Scripted:
|
||||
"""第一轮吐 first;收到自查请求(human 里有「上线前自查」)时吐 fix。"""
|
||||
|
||||
def __init__(self, first, fix):
|
||||
self.first, self.fix = first, fix
|
||||
self.calls: list[str] = []
|
||||
|
||||
async def astream(self, messages):
|
||||
from langchain_core.messages import AIMessageChunk
|
||||
|
||||
human = next(text for role, text in messages if role == "human")
|
||||
self.calls.append(human)
|
||||
for op in (self.fix if "上线前自查" in human else self.first):
|
||||
yield AIMessageChunk(content=json.dumps(op, ensure_ascii=False) + "\n")
|
||||
|
||||
|
||||
def _ops(source, sql):
|
||||
return [
|
||||
{"op": "plan", "summary": "查订单金额"},
|
||||
{"op": "add_node", "node": {"id": "start", "type": "input", "label": "输入", "config": {}}},
|
||||
{"op": "add_node", "node": {"id": "fetch", "type": "tool", "label": "取数",
|
||||
"config": {"tool": f"db_query__{source}", "args": {"sql": sql}}}},
|
||||
{"op": "add_edge", "edge": {"source": "start", "target": "fetch"}},
|
||||
{"op": "add_node", "node": {"id": "out", "type": "output", "label": "成果",
|
||||
"config": {"fields": [{"name": "r", "value": "{{ nodes.fetch }}"}]}}},
|
||||
{"op": "add_edge", "edge": {"source": "fetch", "target": "out"}},
|
||||
{"op": "done", "explanation": "查一下", "run": True},
|
||||
]
|
||||
|
||||
|
||||
@pytest.fixture
|
||||
async def scenic_source(scenic_db):
|
||||
"""库里一个带目录的景区数据源:表结构真实探查,目录按 scenic_notes 写进 catalog_notes。"""
|
||||
from app.db.base import SessionLocal
|
||||
from app.db.models import DataSource
|
||||
|
||||
name = f"scenic_{uuid.uuid4().hex[:8]}"
|
||||
async with SessionLocal() as session:
|
||||
row = DataSource(name=name, kind="sqlite", database=scenic_db, readonly=True)
|
||||
row.schema_cache = await _schema(scenic_db)
|
||||
session.add(row)
|
||||
await session.commit()
|
||||
for table, notes in scenic_notes.notes().items():
|
||||
await catalog.write_entry(session, row.id, table, notes, if_version=0, actor="王敏")
|
||||
source_id = row.id
|
||||
yield name
|
||||
async with SessionLocal() as session:
|
||||
row = await session.get(DataSource, source_id)
|
||||
row.enabled = False
|
||||
await session.commit()
|
||||
|
||||
|
||||
async def _stream(monkeypatch, model, source):
|
||||
import app.api.copilot as copilot
|
||||
from app.main import app
|
||||
|
||||
async def _model(*_a, **_k):
|
||||
return model, "mock-fast"
|
||||
|
||||
monkeypatch.setattr(copilot, "get_chat_model", _model)
|
||||
events: list[dict] = []
|
||||
async with AsyncClient(transport=ASGITransport(app=app), base_url="http://test") as client:
|
||||
async with client.stream("POST", "/api/copilot/generate-stream",
|
||||
json={"instruction": "上周订单金额多少", "intent": "answer",
|
||||
"datasource_ids": [source]}) as r:
|
||||
async for line in r.aiter_lines():
|
||||
if line.startswith("data: "):
|
||||
events.append(json.loads(line[6:]))
|
||||
return events
|
||||
|
||||
|
||||
async def test_fanout_is_sent_back_and_fixed_before_running(scenic_source, monkeypatch):
|
||||
model = _Scripted(_ops(scenic_source, FANOUT_SQL),
|
||||
[{"op": "update_node", "id": "fetch", "config": {"args": {"sql": FIXED_SQL}}},
|
||||
{"op": "done", "explanation": "改好了"}])
|
||||
events = await _stream(monkeypatch, model, scenic_source)
|
||||
checks = [e for e in events if e["op"] == "check"]
|
||||
assert [c["status"] for c in checks] == ["repairing", "passed"], checks
|
||||
[issue] = checks[0]["issues"]
|
||||
assert issue["code"] == "fanout_sum" and issue["node_id"] == "fetch"
|
||||
# 交回模型的是 for_model:写 SQL 里的真名和改法
|
||||
assert "orders 关联 order_items 是一对多" in model.calls[1] and "「fetch」" in model.calls[1]
|
||||
final = events[-1]
|
||||
assert final["op"] == "final" and final["autorun"] is True
|
||||
sql = next(n for n in final["graph"]["nodes"] if n["id"] == "fetch")["data"]["config"]["args"]["sql"]
|
||||
assert sql == FIXED_SQL
|
||||
# 剩下的 warning 不返工,留在 final 里看得见
|
||||
[warn] = [i for i in final["issues"] if i.get("code") == "missing_valid_filter"]
|
||||
assert warn["level"] == "warning" and warn["node_id"] == "fetch" and "status = 1" in warn["message"]
|
||||
assert not [i for i in final["issues"] if i["level"] == "error"]
|
||||
|
||||
|
||||
async def test_a_fanout_that_is_not_fixed_stops_autorun(scenic_source, monkeypatch):
|
||||
model = _Scripted(_ops(scenic_source, FANOUT_SQL),
|
||||
[{"op": "update_node", "id": "fetch", "config": {"args": {"sql": FANOUT_SQL + " "}}},
|
||||
{"op": "done", "explanation": "改了"}])
|
||||
events = await _stream(monkeypatch, model, scenic_source)
|
||||
checks = [e for e in events if e["op"] == "check"]
|
||||
assert [c["status"] for c in checks] == ["repairing", "repairing", "failed"], checks
|
||||
final = events[-1]
|
||||
assert final["autorun"] is False
|
||||
assert [i["code"] for i in final["issues"] if i["level"] == "error"] == ["fanout_sum"]
|
||||
|
||||
|
||||
async def test_clean_sql_costs_no_extra_call(scenic_source, monkeypatch):
|
||||
model = _Scripted(_ops(scenic_source, CLEAN_SQL), [])
|
||||
events = await _stream(monkeypatch, model, scenic_source)
|
||||
assert len(model.calls) == 1
|
||||
assert [e["status"] for e in events if e["op"] == "check"] == ["passed"]
|
||||
final = events[-1]
|
||||
assert final["autorun"] is True
|
||||
assert not [i for i in final["issues"] if i.get("code") in ("fanout_sum", "missing_valid_filter")]
|
||||
Reference in New Issue
Block a user