Files
AgentLab/backend/tests/test_authoring_recovery.py
T
ThearchyHelios e15e94f93e 口径工作台:结构化语义导入、起草续跑恢复与本体关系编排
- 新增口径结构化语义定义、分批解析与 SQL 生成管道(semantic / semantic_import / semantic_sql)
- 支持起草断点续跑恢复、指标自动匹配对齐与存量口径语义迁移面板(MigrationPanel / SemanticPanel)
- 口径草稿与变更流程接入候选本体边提议、审阅确认与一键修改保护,同步更新可移植 SQLite 基线校验
2026-11-12 12:58:12 +08:00

257 lines
14 KiB
Python

"""Fault injection at graph acceptance and safe missing-definition continuation."""
import asyncio
from sqlalchemy import select
from app.api import ledger
from app.db.base import SessionLocal
from app.db.models import CaliberDraft, DataSource, Workflow
from app.ledger import authoring
from tests.test_ledger_docdraft_api import client, datasource, documents, extract, fake_model, fake_trial, plan_reply
from tests.test_definition_batches import imported, reply
async def test_graph_acceptance_is_atomic_and_resume_does_not_duplicate(
client, documents, datasource, fake_trial, monkeypatch):
doc = await documents('atomic-acceptance')
source = await datasource()
model = fake_model(monkeypatch, plan_reply(f'{doc}-2'))
capture = authoring._capture_field_baseline
async def interrupted(**kwargs):
async with SessionLocal() as session:
saved = await session.get(CaliberDraft, kwargs['draft'].id)
assert saved.workflow_id == kwargs['workflow_id']
assert saved.payload['finalization_pending']
raise asyncio.CancelledError
monkeypatch.setattr(authoring, '_capture_field_baseline', interrupted)
made = await extract(client, doc, source)
draft_id = made['draft_id']
failed = (await client.get(f'/api/ledger/drafts/{draft_id}')).json()
assert failed['status'] == 'failed' and failed['job']['retryable']
workflow_id = failed['workflow_id']
monkeypatch.setattr(authoring, '_capture_field_baseline', capture)
resumed = await client.post(f'/api/ledger/drafts/{draft_id}/retry')
assert resumed.status_code == 200, resumed.text
ready = (await client.get(f'/api/ledger/drafts/{draft_id}')).json()
assert ready['workflow_id'] == workflow_id and ready['status'] == 'ready'
assert not ready['finalization_pending'] and len(model.calls) == 1
async with SessionLocal() as session:
workflows = (await session.execute(select(Workflow))).scalars().all()
assert len([w for w in workflows if (w.origin or {}).get('draft_id') == draft_id]) == 1
async def test_disabled_source_rejected_in_old_new_retry_and_worker(
client, documents, datasource, monkeypatch):
draft_id, source, chunk, items = await imported(client, datasource, monkeypatch, count=1)
async with SessionLocal() as session:
row = await session.get(DataSource, source)
row.enabled = False
await session.commit()
assert (await client.post(f'/api/ledger/drafts/{draft_id}/implement',
json={'datasource_id': source})).status_code == 422
doc = await documents('disabled-source')
assert (await client.post('/api/ledger/drafts/extract',
json={'documents': [doc], 'datasource_id': source})).status_code == 422
async with SessionLocal() as session:
row = await session.get(CaliberDraft, draft_id)
row.status, row.datasource_id = 'drafting', source
await session.commit()
await ledger._extract_in_background(draft_id)
result = (await client.get(f'/api/ledger/drafts/{draft_id}')).json()
assert result['status'] == 'failed' and '停用' in result['error']
assert (await client.post(f'/api/ledger/drafts/{draft_id}/retry')).status_code == 409
async def test_continue_is_idempotent_missing_only_and_preserves_original(
client, datasource, fake_trial, monkeypatch):
draft_id, source, chunk, items = await imported(client, datasource, monkeypatch, count=2)
await client.post(f'/api/ledger/drafts/{draft_id}/implement', json={'datasource_id': source})
fake_model(monkeypatch, reply(items[:1], chunk), '{"groups": [], "open_questions": ["缺少字段"]}')
await ledger._extract_in_background(draft_id)
before = (await client.get(f'/api/ledger/drafts/{draft_id}')).json()
assert before['metric_count'] == 1 and before['status'] == 'questions_open'
async with SessionLocal() as session:
workflow = await session.get(Workflow, before['workflow_id'])
original_graph, version = workflow.graph, workflow.version
body = {'expected_revision': version, 'guidance': '补充:计数使用 orders'}
result = await client.post(f'/api/ledger/drafts/{draft_id}/continue', json=body)
assert result.status_code == 202, result.text
candidate = result.json()['draft_id']
assert candidate != draft_id and result.json()['merge_required']
assert (await client.post(f'/api/ledger/drafts/{draft_id}/continue', json=body)).json()['draft_id'] == candidate
fake_model(monkeypatch, reply(items[1:], chunk))
await ledger._extract_in_background(candidate)
ready = (await client.get(f'/api/ledger/drafts/{candidate}')).json()
assert ready['metric_count'] == 1
assert ready['metrics'][0]['source_definition_id'] == items[1]['id']
async with SessionLocal() as session:
workflow = await session.get(Workflow, before['workflow_id'])
assert workflow.graph == original_graph and workflow.version == version
assert (await client.post(f'/api/ledger/drafts/{draft_id}/continue',
json={**body, 'expected_revision': version + 1})).status_code == 409
merged = await client.post(f'/api/ledger/drafts/{candidate}/merge', json={
'expected_revision': version, 'candidate_revision': ready['version']})
assert merged.status_code == 200, merged.text
combined = merged.json()
assert combined['metric_count'] == 2 and combined['workflow_id'] == before['workflow_id']
assert combined['version'] == version + 1
async with SessionLocal() as session:
workflow = await session.get(Workflow, before['workflow_id'])
old_card = next(n for n in original_graph['nodes'] if n['type'] == 'metrics')
new_card = next(n for n in workflow.graph['nodes'] if n['type'] == 'metrics')
assert new_card['data']['config']['metrics'][0] == old_card['data']['config']['metrics'][0]
assert len({m['id'] for m in new_card['data']['config']['metrics']}) == 2
replay = await client.post(f'/api/ledger/drafts/{candidate}/merge', json={
'expected_revision': version, 'candidate_revision': ready['version']})
assert replay.status_code == 200 and replay.json()['replayed']
assert replay.json()['version'] == combined['version']
async def test_all_blocked_can_continue_and_merge_without_sharing_candidate_graph(
client, datasource, fake_trial, monkeypatch):
draft_id, source, chunk, items = await imported(client, datasource, monkeypatch, count=1)
await client.post(f'/api/ledger/drafts/{draft_id}/implement', json={'datasource_id': source})
empty = '{"groups": [], "open_questions": ["缺少字段"]}'
fake_model(monkeypatch, empty, empty)
await ledger._extract_in_background(draft_id)
before = (await client.get(f'/api/ledger/drafts/{draft_id}')).json()
assert before['status'] == 'questions_open' and not before['workflow_id']
result = await client.post(f'/api/ledger/drafts/{draft_id}/continue', json={
'expected_revision': 0, 'guidance': '已补齐字段'})
assert result.status_code == 202, result.text
candidate = result.json()['draft_id']
fake_model(monkeypatch, reply(items, chunk))
await ledger._extract_in_background(candidate)
ready = (await client.get(f'/api/ledger/drafts/{candidate}')).json()
conflict = await client.post(f'/api/ledger/drafts/{candidate}/merge', json={
'expected_revision': 0, 'candidate_revision': 99})
assert conflict.status_code == 409
merged = await client.post(f'/api/ledger/drafts/{candidate}/merge', json={
'expected_revision': 0, 'candidate_revision': ready['version']})
assert merged.status_code == 200, merged.text
assert merged.json()['metric_count'] == 1
assert merged.json()['workflow_id'] != ready['workflow_id']
def test_merge_keeps_human_edits_and_rejects_conflicting_parameters():
import copy
import pytest
from app.ledger.continuation import append_candidate
from app.ledger.draftplan import build_graph
def graph(key):
return build_graph([{'sql': 'SELECT 1 AS n', 'entry_fields': [
{'name': 'date', 'type': 'string', 'default': 'yesterday'}], 'metrics': [
{'id': 'm1', 'name': key, 'source_definition_id': key, 'expression': "cell(nodes.fetch, 0, 'n')"}]}], name=key)
original, candidate = graph('old'), graph('new')
old_card = next(n for n in original['nodes'] if n['type'] == 'metrics')
old_card['data']['config']['metrics'][0]['expression'] = "cell(nodes.fetch, 0, 'n') * 100"
original_copy = copy.deepcopy(original)
merged = append_candidate(original, candidate, prefix='candidate', allowed_definitions={'new'})
merged_card = next(n for n in merged['nodes'] if n['type'] == 'metrics')
assert original == original_copy
assert merged_card['data']['config']['metrics'][0] == old_card['data']['config']['metrics'][0]
assert 'nodes.candidate_fetch' in merged_card['data']['config']['metrics'][1]['expression']
next(n for n in candidate['nodes'] if n['type'] == 'input')['data']['config']['fields'][0]['default'] = 'today'
with pytest.raises(ValueError, match='参数'):
append_candidate(original, candidate, prefix='candidate', allowed_definitions={'new'})
async def test_continue_supports_subset_datasource_switch_multi_metric_split_and_field_baseline(
client, datasource, fake_trial, monkeypatch):
import json
import uuid
draft_id, source1, chunk, items = await imported(client, datasource, monkeypatch, count=3)
source2 = f'ds-{uuid.uuid4().hex[:10]}'
async with SessionLocal() as session:
session.add(DataSource(
id=source2, name=f'mes_{uuid.uuid4().hex[:6]}', kind='sqlite',
description='制造库', schema_cache={'tables': {'orders': {'columns': [
{'name': 'amount', 'type': 'NUMERIC'}]}}, 'total': 1},
))
await session.commit()
await client.post(f'/api/ledger/drafts/{draft_id}/implement', json={'datasource_id': source1})
empty = '{"groups": [], "open_questions": ["请确认取数表名"]}'
fake_model(monkeypatch, empty)
await ledger._extract_in_background(draft_id)
initial = (await client.get(f'/api/ledger/drafts/{draft_id}')).json()
assert initial['status'] == 'questions_open' and not initial['workflow_id']
bad_subset = await client.post(f'/api/ledger/drafts/{draft_id}/continue', json={
'expected_revision': 0, 'guidance': '先验证第一项', 'selected': ['unknown-def-id'],
})
assert bad_subset.status_code == 422
cont1 = await client.post(f'/api/ledger/drafts/{draft_id}/continue', json={
'expected_revision': 0, 'guidance': '切换到制造库并先处理第一项',
'selected': [items[0]['id']], 'datasource_id': source2,
})
assert cont1.status_code == 202, cont1.text
cand1 = cont1.json()['draft_id']
split_reply = json.dumps({
'sql': 'SELECT COUNT(*) AS a, SUM(amount) AS b FROM orders',
'entry_fields': [],
'metrics': [
{
'id': 'm1', 'name': f"{items[0]['name']}(计划口径)", 'source_definition_id': items[0]['id'],
'expression': "cell(nodes.fetch, 0, 'a')",
'definition': {'text': '计划口径', 'citations': [{'chunk': chunk, 'quote': '按日汇总计数,不取明细第一行'}]},
},
{
'id': 'm2', 'name': f"{items[0]['name']}(产线口径)", 'source_definition_id': items[0]['id'],
'expression': "cell(nodes.fetch, 0, 'b')",
'definition': {'text': '产线口径', 'citations': [{'chunk': chunk, 'quote': '按日汇总计数,不取明细第一行'}]},
},
],
'uncovered': [], 'open_questions': [],
}, ensure_ascii=False)
fake_model(monkeypatch, split_reply)
await ledger._extract_in_background(cand1)
cand1_view = (await client.get(f'/api/ledger/drafts/{cand1}')).json()
assert cand1_view['metric_count'] == 2
merged1 = await client.post(f'/api/ledger/drafts/{cand1}/merge', json={
'expected_revision': 0, 'candidate_revision': cand1_view['version'],
})
assert merged1.status_code == 200, merged1.text
after1 = merged1.json()
assert after1['datasource_id'] == source2
assert after1['metric_count'] == 2
assert after1['implementation_progress']['completed'] == 1
assert after1['implementation_progress']['blocked'] == 2
assert after1['field_baseline']['status'] == 'known'
assert after1['field_baseline']['coverage']['status'] == 'complete'
# Switching datasource is rejected once the draft owns a workflow graph.
switch_rejected = await client.post(f'/api/ledger/drafts/{draft_id}/continue', json={
'expected_revision': after1['version'], 'guidance': '尝试再换库', 'datasource_id': source1,
})
assert switch_rejected.status_code == 409
# Second subset continuation advances field_baseline cleanly onto existing workflow.
cont2 = await client.post(f'/api/ledger/drafts/{draft_id}/continue', json={
'expected_revision': after1['version'], 'guidance': '继续处理第二项', 'selected': [items[1]['id']],
})
assert cont2.status_code == 202, cont2.text
cand2 = cont2.json()['draft_id']
fake_model(monkeypatch, reply([items[1]], chunk))
await ledger._extract_in_background(cand2)
cand2_view = (await client.get(f'/api/ledger/drafts/{cand2}')).json()
merged2 = await client.post(f'/api/ledger/drafts/{cand2}/merge', json={
'expected_revision': after1['version'], 'candidate_revision': cand2_view['version'],
})
assert merged2.status_code == 200, merged2.text
after2 = merged2.json()
assert after2['metric_count'] == 3
assert after2['version'] == after1['version'] + 1
assert after2['implementation_progress']['completed'] == 2
assert after2['field_baseline']['status'] == 'known'
assert after2['field_baseline']['coverage']['status'] == 'complete'
assert len(after2['field_baseline']['metric_ids']) == 3