fix(platform): 配置发布激活时点名"将被丢掉的配置项"
config_release 是**整版本替换**语义:激活新版本后,旧版本的配置项全部不再生效。 所以新版本只要漏了某项,它就是**无声消失**的——agent_tools 里的工具白名单一少, 相关 Agent 的工具就被 fail-closed 拒掉,而现场表现只是"客服/风控什么都答不了", 没人会想到是发布配置少了一条。 本项目已经两次靠"发布前手工继承"规避(客服与风控的发布脚本里各写了一遍继承逻辑), 说明这个风险真实且反复出现。 改为在 activate 时先比对"被取代版本的配置项"与"新版本的配置项",把将被丢掉的逐条 写进 warning 日志。**不阻断激活**——有时确实是要主动撤下某项配置,拒绝会让正常运维 做不了事;这里要的是"事后能查到是谁把它弄没的"。 新增 tests/unit/service/test_config_release_dropped_items.py(3 条):丢项时点名、 完整继承时无噪音(否则运维会习惯性忽略这条日志)、首个版本不报丢项。 ruff / mypy(135 文件) / 610 unit+contract 全绿。
This commit is contained in:
@@ -1,3 +1,4 @@
|
||||
import logging
|
||||
from datetime import UTC, datetime
|
||||
from uuid import uuid4
|
||||
|
||||
@@ -8,6 +9,8 @@ from app.model.audit import InteractionAudit
|
||||
from app.model.configuration import ConfigRelease, PlatformConfigItem
|
||||
from app.model.platform import DomainEventOutbox
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
|
||||
class ConfigReleaseError(ValueError):
|
||||
pass
|
||||
@@ -45,6 +48,44 @@ class ConfigReleaseService:
|
||||
await self.session.flush()
|
||||
return release
|
||||
|
||||
async def _warn_dropped_items(
|
||||
self, release: ConfigRelease, previous_active: list[ConfigRelease]
|
||||
) -> None:
|
||||
"""列出"旧版本有、新版本没有"的配置项——它们在激活后会静默失效。
|
||||
|
||||
`config_release` 是**整版本替换**语义:激活新版本后,旧版本的所有配置项都不再生效。
|
||||
所以只要新版本漏了某一项,它就是**无声消失**的——工具白名单随之变成空集、相关
|
||||
Agent 的工具被 fail-closed 拒掉,而现场表现只是"客服/风控什么都答不了",
|
||||
没人会想到是发布配置少了一条。
|
||||
|
||||
本项目已经两次靠"发布前手工继承"规避(客服与风控的发布脚本里各写了一遍继承逻辑),
|
||||
这说明风险是真实且反复出现的。这里不阻断激活(有时确实是要撤下某项配置),
|
||||
但把它显式写进日志,让运维事后能查到"是谁把它弄没的"。
|
||||
"""
|
||||
if not previous_active:
|
||||
return
|
||||
rows = (await self.session.execute(
|
||||
select(
|
||||
PlatformConfigItem.namespace,
|
||||
PlatformConfigItem.config_key,
|
||||
PlatformConfigItem.release_id,
|
||||
).where(PlatformConfigItem.release_id.in_(
|
||||
[previous.id for previous in previous_active] + [release.id]
|
||||
))
|
||||
)).all()
|
||||
new_keys = {(namespace, key) for namespace, key, rid in rows if rid == release.id}
|
||||
for previous in previous_active:
|
||||
old_keys = {(namespace, key) for namespace, key, rid in rows if rid == previous.id}
|
||||
dropped = sorted(old_keys - new_keys)
|
||||
if dropped:
|
||||
logger.warning(
|
||||
"配置发布 %s 取代 %s:以下 %d 条配置项在新版本中不存在,激活后即失效 → %s",
|
||||
release.release_no,
|
||||
previous.release_no,
|
||||
len(dropped),
|
||||
[f"{namespace}/{key}" for namespace, key in dropped],
|
||||
)
|
||||
|
||||
async def activate(self, release_id: int, actor_id: int) -> ConfigRelease:
|
||||
release = await self._get(release_id)
|
||||
if release.status != "approved":
|
||||
@@ -54,6 +95,8 @@ class ConfigReleaseService:
|
||||
)
|
||||
now = self._now()
|
||||
previous_active = [previous for previous in active if previous.id != release.id]
|
||||
# 把"本次激活会让哪些配置项失效"显式记进日志,见 _warn_dropped_items 的说明。
|
||||
await self._warn_dropped_items(release, previous_active)
|
||||
for previous in previous_active:
|
||||
previous.status = "superseded"
|
||||
previous.updated_at = now
|
||||
|
||||
@@ -0,0 +1,91 @@
|
||||
"""配置发布激活时的"配置项被静默丢掉"警告。
|
||||
|
||||
**背景**:`config_release` 是**整版本替换**语义 —— 激活新版本后,旧版本的配置项全部不再
|
||||
生效。所以新版本只要漏了某项,它就是**无声消失**的:`agent_tools` 里的工具白名单一少,
|
||||
相关 Agent 的工具就被 fail-closed 拒掉,而现场表现只是"客服/风控什么都答不了",
|
||||
没人会想到是发布配置少了一条。
|
||||
|
||||
本项目已经两次靠"发布前手工继承"规避(客服与风控的发布脚本里各写了一遍继承逻辑),
|
||||
说明风险真实且反复出现。这里锁住:只要发生丢项,就必须在日志里点名。
|
||||
"""
|
||||
|
||||
import logging
|
||||
from typing import Any, cast
|
||||
|
||||
import pytest
|
||||
|
||||
from app.service.config_release_service import ConfigReleaseService
|
||||
|
||||
|
||||
class _FakeResult:
|
||||
def __init__(self, rows: list[tuple[Any, ...]]) -> None:
|
||||
self._rows = rows
|
||||
|
||||
def all(self) -> list[tuple[Any, ...]]:
|
||||
return self._rows
|
||||
|
||||
|
||||
class _FakeSession:
|
||||
"""只提供 `execute(...).all()`,因为被测方法只用到这一条路径。"""
|
||||
|
||||
def __init__(self, rows: list[tuple[Any, ...]]) -> None:
|
||||
self._rows = rows
|
||||
|
||||
async def execute(self, statement: Any) -> _FakeResult:
|
||||
return _FakeResult(self._rows)
|
||||
|
||||
|
||||
class _Release:
|
||||
def __init__(self, release_id: int, release_no: str) -> None:
|
||||
self.id = release_id
|
||||
self.release_no = release_no
|
||||
|
||||
|
||||
def _service(rows: list[tuple[Any, ...]]) -> ConfigReleaseService:
|
||||
return ConfigReleaseService(cast(Any, _FakeSession(rows)))
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_warns_when_new_release_drops_config_items(
|
||||
caplog: pytest.LogCaptureFixture,
|
||||
) -> None:
|
||||
"""新版本少了一条配置项时必须点名警告——它会静默失效。"""
|
||||
service = _service([
|
||||
("agent_tools", "customer_service:faq", 1), # 旧版本有
|
||||
("agent_tools", "risk:risk_overview", 2), # 新版本新增
|
||||
])
|
||||
|
||||
with caplog.at_level(logging.WARNING):
|
||||
await service._warn_dropped_items(_Release(2, "new"), [_Release(1, "old")])
|
||||
|
||||
assert "customer_service:faq" in caplog.text
|
||||
assert "agent_tools/customer_service:faq" in caplog.text
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_no_warning_when_everything_is_carried_over(
|
||||
caplog: pytest.LogCaptureFixture,
|
||||
) -> None:
|
||||
"""配置项被完整继承时不该有噪音——否则运维会习惯性忽略这条日志。"""
|
||||
service = _service([
|
||||
("agent_tools", "customer_service:faq", 1),
|
||||
("agent_tools", "customer_service:faq", 2),
|
||||
])
|
||||
|
||||
with caplog.at_level(logging.WARNING):
|
||||
await service._warn_dropped_items(_Release(2, "new"), [_Release(1, "old")])
|
||||
|
||||
assert caplog.text == ""
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_first_release_warns_about_nothing(
|
||||
caplog: pytest.LogCaptureFixture,
|
||||
) -> None:
|
||||
"""平台首个版本没有"前一个版本",不该报丢项。"""
|
||||
service = _service([("agent_tools", "customer_service:faq", 1)])
|
||||
|
||||
with caplog.at_level(logging.WARNING):
|
||||
await service._warn_dropped_items(_Release(1, "first"), [])
|
||||
|
||||
assert caplog.text == ""
|
||||
Reference in New Issue
Block a user