refactor: add CHECK constraints and drop redundant index in v3 tables

This commit is contained in:
2026-07-15 09:47:00 -04:00
parent 3cc8dc9105
commit 58a84dc72b
2 changed files with 101 additions and 12 deletions
+17 -6
View File
@@ -183,7 +183,7 @@ CREATE TABLE IF NOT EXISTS facts (
fact_type TEXT,
difficulty_tier INTEGER,
verifier_refs TEXT,
cross_agree INTEGER,
cross_agree INTEGER CHECK(cross_agree IN (0, 1)),
negative_at_target TEXT,
session_id TEXT,
ts TEXT
@@ -195,8 +195,8 @@ CREATE TABLE IF NOT EXISTS unit_verdict (
unit_id TEXT NOT NULL,
pair_id TEXT,
sub_pattern TEXT,
stage INTEGER NOT NULL,
verdict TEXT,
stage INTEGER NOT NULL CHECK(stage BETWEEN 1 AND 6),
verdict TEXT CHECK(verdict IN ('pass', 'fail', 'abstain')),
reason TEXT,
metric_value REAL,
model TEXT,
@@ -234,7 +234,7 @@ CREATE TABLE IF NOT EXISTS quarantine (
_DDL_RESUME_STATE = """
CREATE TABLE IF NOT EXISTS resume_state (
unit_id TEXT PRIMARY KEY,
status TEXT,
status TEXT CHECK(status IN ('pending', 'accepted', 'rejected')),
config_fingerprint TEXT,
seq_offset INTEGER
);
@@ -251,7 +251,7 @@ _DDL_V3_TABLES = [
_DDL_V3_INDEXES = [
"CREATE INDEX IF NOT EXISTS idx_facts_video ON facts(video_id);",
"CREATE INDEX IF NOT EXISTS idx_facts_session ON facts(session_id);",
"CREATE INDEX IF NOT EXISTS idx_uv_unit ON unit_verdict(unit_id);",
# unit_verdict(unit_id) 无需独立索引:主键 (unit_id, stage) 的左前缀已覆盖 unit_id 查询。
"CREATE INDEX IF NOT EXISTS idx_uv_sub_pattern ON unit_verdict(sub_pattern);",
"CREATE INDEX IF NOT EXISTS idx_quar_sub_pattern ON quarantine(sub_pattern);",
"CREATE INDEX IF NOT EXISTS idx_resume_status ON resume_state(status);",
@@ -848,6 +848,10 @@ class QuestionGenStore:
同一 (unit_id, stage) 重跑时以 upsert 覆盖旧行——每 unit 每层仅保留最新一条,
便于断点续跑重放而不残留过期裁决。
覆盖语义:ON CONFLICT DO UPDATE 用 ``excluded.*`` **整行覆盖**全部非键列
(含可空的 metric_value/model 及 ts)——只保留每层最新完整裁决,不支持部分字段
更新。调用方每次必须传完整行,否则会用 metric_value=None/model=None 误抹先前非空值。
Parameters
----------
unit_id : str
@@ -921,6 +925,9 @@ class QuestionGenStore:
同一 pair_id 重算时以 upsert 覆盖旧行——每 pair 仅保留最新一组度量。
覆盖语义:ON CONFLICT DO UPDATE 用 ``excluded.*`` **整行覆盖**全部非键列(含 ts),
不支持部分字段更新;调用方每次必须传完整度量行。
Parameters
----------
pair_id : str
@@ -980,7 +987,8 @@ class QuestionGenStore:
"""将失败题的内容指纹写入隔离区黑名单(quarantine 表,主键 content_fingerprint)。
同一 content_fingerprint 重复调用以 upsert 覆盖——保证语义相同题面只占一行,
补构造前查此表当黑名单去重。
补构造前查此表当黑名单去重。覆盖语义:ON CONFLICT DO UPDATE 用 ``excluded.*``
**整行覆盖**全部非键列(含 ts),不支持部分字段更新。
Parameters
----------
@@ -1025,6 +1033,9 @@ class QuestionGenStore:
由调用方据此判定旧进度作废并重跑。本表无 ts 列(schema 未定义),但方法体仍禁
进程内 now,保持 v3 观测一致的幂等可复现语义。
覆盖语义:ON CONFLICT DO UPDATE 用 ``excluded.*`` **整行覆盖**全部非键列,不支持
部分字段更新;调用方每次必须传完整状态行。
Parameters
----------
unit_id : str
+84 -6
View File
@@ -6,6 +6,7 @@
from __future__ import annotations
import sqlite3
from typing import TYPE_CHECKING
import pytest
@@ -173,19 +174,22 @@ def test_insert_unit_verdict_rerun_same_stage_overwrites(store: QuestionGenStore
ts="2026-07-15T11:10:00+00:00",
**common,
)
new_ts = "2026-07-15T11:20:00+00:00"
store.insert_unit_verdict(
verdict="pass",
reason="second",
metric_value=0.9,
model="m2",
ts="2026-07-15T11:20:00+00:00",
ts=new_ts,
**common,
)
rows = store._conn.execute(
"SELECT verdict, reason, metric_value FROM unit_verdict WHERE unit_id='unit-3' AND stage=2"
"SELECT verdict, reason, metric_value, ts "
"FROM unit_verdict WHERE unit_id='unit-3' AND stage=2"
).fetchall()
assert len(rows) == 1
assert rows[0] == ("pass", "second", 0.9)
# ts 随整行覆盖为第二次(新)ts,锁死 DO UPDATE SET ts=excluded.ts 语义。
assert rows[0] == ("pass", "second", 0.9, new_ts)
# ---------------------------------------------------------------------------
@@ -241,6 +245,7 @@ def test_insert_collapse_metrics_recompute_upserts(store: QuestionGenStore) -> N
subtitle_answerability=0.1,
ts="2026-07-15T12:00:00+00:00",
)
new_ts = "2026-07-15T12:30:00+00:00"
store.insert_collapse_metrics(
pair_id="pair-2",
text_only_acc=0.05,
@@ -251,13 +256,14 @@ def test_insert_collapse_metrics_recompute_upserts(store: QuestionGenStore) -> N
distractor_min_dist=0.9,
multiformat_consistency=0.95,
subtitle_answerability=0.0,
ts="2026-07-15T12:30:00+00:00",
ts=new_ts,
)
rows = store._conn.execute(
"SELECT text_only_acc, slot_chi2 FROM collapse_metrics WHERE pair_id='pair-2'"
"SELECT text_only_acc, slot_chi2, ts FROM collapse_metrics WHERE pair_id='pair-2'"
).fetchall()
assert len(rows) == 1
assert rows[0] == (0.05, 9.0)
# ts 随整行覆盖为第二次(新)ts,锁死 DO UPDATE SET ts=excluded.ts 语义。
assert rows[0] == (0.05, 9.0, new_ts)
# ---------------------------------------------------------------------------
@@ -346,6 +352,78 @@ def test_upsert_resume_state_same_unit_overwrites(store: QuestionGenStore) -> No
assert rows[0] == ("accepted", "cfg-2", 5)
# ---------------------------------------------------------------------------
# CHECK 约束 fail-loud(非法枚举写入直接报错,符合 §4.2/P5)
# ---------------------------------------------------------------------------
def test_unit_verdict_invalid_stage_raises(store: QuestionGenStore) -> None:
"""非法 stage(超出 1-6)触发 CHECK 约束,写入直接 IntegrityError。"""
with pytest.raises(sqlite3.IntegrityError):
store.insert_unit_verdict(
unit_id="unit-bad",
pair_id=None,
sub_pattern="order",
stage=7,
verdict="pass",
reason="r",
metric_value=None,
model=None,
session_id="sess-1",
ts="2026-07-15T11:00:00+00:00",
)
def test_unit_verdict_invalid_verdict_raises(store: QuestionGenStore) -> None:
"""非法 verdict(不在 pass/fail/abstain)触发 CHECK 约束,写入直接 IntegrityError。"""
with pytest.raises(sqlite3.IntegrityError):
store.insert_unit_verdict(
unit_id="unit-bad2",
pair_id=None,
sub_pattern="order",
stage=1,
verdict="maybe",
reason="r",
metric_value=None,
model=None,
session_id="sess-1",
ts="2026-07-15T11:00:00+00:00",
)
def test_facts_invalid_cross_agree_raises(store: QuestionGenStore) -> None:
"""非法 cross_agree(非 0/1)触发 CHECK 约束,写入直接 IntegrityError。"""
with pytest.raises(sqlite3.IntegrityError):
store.insert_fact(
fact_id="fact-bad",
video_id="v",
segment_id="s",
subject="a",
action="b",
object="c",
frame_ids="[]",
polarity="",
fact_type="state",
difficulty_tier=1,
verifier_refs="{}",
cross_agree=2,
negative_at_target="x",
session_id="sess",
ts="2026-07-15T11:00:00+00:00",
)
def test_resume_state_invalid_status_raises(store: QuestionGenStore) -> None:
"""非法 status(不在 pending/accepted/rejected)触发 CHECK 约束,写入直接 IntegrityError。"""
with pytest.raises(sqlite3.IntegrityError):
store.upsert_resume_state(
unit_id="unit-bad",
status="unknown",
config_fingerprint="cfg",
seq_offset=0,
)
# ---------------------------------------------------------------------------
# v2 旧表未受影响
# ---------------------------------------------------------------------------