feat(server): 工程骨架后端——上传 → 解析 → 清单最小链路

- core:配置(prod 拒绝默认密钥 / dev 短信 / 假模型)、租户事务 + 行级安全策略、
  problem+json 错误码、信封加密与签名下载、对象存储、事务内入队、Valkey stream 事件流、权限点
- identity:手机号 HMAC 存储、注册、登录(挤下线需确认,API3)、服务端会话
- billing:额度流水只追加,预扣 / 结算 / 退回,一笔预扣只有一个结局
- document:分片上传、按文件头识别格式、Gotenberg 转换、文字层与稳定句子编号
- checklist:标注规则识别、模型抽取、三道校验、标注类条款规则兜底、重跑对齐
- ai:提示词注册表(extract.substantive v1 来自 S4 v2)、假模型、OpenAI 兼容供应商、ai_call 流水
- 迁移:data-model 基线、套餐、Procrastinate、会话状态函数
- 测试 30 个:租户隔离、额度、失败退回、对象加密、契约一致性
- Compose(含 migrate)、镜像、CI

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
YANG JIANKUAN
2026-11-03 18:48:10 +08:00
parent 5747a523bb
commit ad7fba6021
106 changed files with 8094 additions and 0 deletions

View File

@@ -0,0 +1,23 @@
"""基线:data-model v0.1 的全部 40 张表、行级安全策略、角色授权。
Revision ID: 0001
"""
from __future__ import annotations
import sys
from pathlib import Path
sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
from sqlfiles import run_dir
revision = "0001"
down_revision = None
def upgrade() -> None:
run_dir("0001_baseline")
def downgrade() -> None:
raise NotImplementedError("基线不支持回退")

View File

@@ -0,0 +1,38 @@
"""套餐 v1:额度数字是 PRD 的建议值,D2 成本实测后改数字就是加一个版本(data-model §4.3)。
Revision ID: 0002
"""
from __future__ import annotations
import json
from alembic import op
from sqlalchemy import text
revision = "0002"
down_revision = "0001"
MB = 1024 * 1024
PLANS = [
("free", 1, "免费版", {"parse": 3}, {"max_file_bytes": 50 * MB, "max_pages": 300, "max_projects": 1}),
(
"pro",
1,
"专业版",
{"parse": 500, "generate": 300},
{"max_file_bytes": 200 * MB, "max_pages": 1000, "max_projects": None},
),
]
def upgrade() -> None:
for code, version, name, quotas, limits in PLANS:
op.get_bind().execute(
text("INSERT INTO plan (code, version, name, quotas, limits) VALUES (:c, :v, :n, :q, :l)"),
{"c": code, "v": version, "n": name, "q": json.dumps(quotas), "l": json.dumps(limits)},
)
def downgrade() -> None:
op.get_bind().execute(text("DELETE FROM plan WHERE version = 1"))

View File

@@ -0,0 +1,31 @@
"""任务队列 Procrastinate 的表(AR4)。放在 public schema:它的 SQL 函数里表名不带 schema。
任务状态的权威是业务表(parse_job 等);队列只负责派发(架构 §3.4)。这些表没有 tenant_id,不受行级安全策略约束。
升级 Procrastinate 时,按它的版本说明另加迁移(server/pyproject.toml 里锁定了大版本)。
Revision ID: 0003
"""
from __future__ import annotations
from alembic import op
from procrastinate.schema import SchemaManager
revision = "0003"
down_revision = "0002"
def upgrade() -> None:
bind = op.get_bind()
bind.exec_driver_sql("SET LOCAL search_path = public")
bind.exec_driver_sql(SchemaManager.get_schema().replace("%", "%%"))
bind.exec_driver_sql(
"GRANT SELECT, INSERT, UPDATE, DELETE ON ALL TABLES IN SCHEMA public TO bd_app;"
"GRANT USAGE, SELECT ON ALL SEQUENCES IN SCHEMA public TO bd_app;"
"GRANT EXECUTE ON ALL FUNCTIONS IN SCHEMA public TO bd_app;"
"SET LOCAL search_path = app, public"
)
def downgrade() -> None:
raise NotImplementedError

View File

@@ -0,0 +1,30 @@
"""登录前查令牌为什么失效:被吊销(含挤下线)还是过期(API3,api-design §2)。
和 auth_resolve_session 一样是 SECURITY DEFINER,所有者 bd_auth,只返回状态,不返回会话内容(data-model §2.3)。
Revision ID: 0004
"""
from __future__ import annotations
from alembic import op
revision = "0004"
down_revision = "0003"
def upgrade() -> None:
op.get_bind().exec_driver_sql("""
CREATE FUNCTION auth_session_state(p_token_hash bytea)
RETURNS TABLE (revoke_reason text, expired boolean)
LANGUAGE sql STABLE SECURITY DEFINER SET search_path = app AS $$
SELECT s.revoke_reason, s.expires_at <= now()
FROM user_session s WHERE s.token_hash = p_token_hash
$$;
ALTER FUNCTION auth_session_state(bytea) OWNER TO bd_auth;
GRANT EXECUTE ON FUNCTION auth_session_state(bytea) TO bd_app;
""")
def downgrade() -> None:
op.get_bind().exec_driver_sql("DROP FUNCTION auth_session_state(bytea)")