mirror of
https://github.com/ThearchyHelios/AgentLab.git
synced 2026-11-19 16:31:59 +00:00
0afffd4b9a
## 沙箱关不掉
get() 先看 prefer,而 _resolve_named 完全不看 settings.sandbox_backend。
配置里 off 只挡得住默认路径,图里任何 isolation=strict 的代码节点照样能跑。
另外 alias.get(prefer, prefer) 对未知档位原样透传,而 isolation 是图 JSON 里的
自由字符串,写 isolation="local" 就能点名那个明说了"没有任何访问控制"的兜底
后端——档位是用来提高隔离的,不该能降级。现在 off 优先判定,档位限白名单。
## formal 可以跑从未过闸的版本
"不可变"和"过过闸"是两件事:PATCH 存出来的草稿同样有不可变快照。原来只查
version 存不存在,于是工作流只要发布过任意一版,之后所有草稿都能以 formal
身份启动,发布期的 validate_graph + lint_for_publish 全部绕过。这和代码里
"正式运行的全部语义就这一条规则:必须引用一个不可变的已发布版本"直接矛盾。
现在要求 version == published_version,否则 409。
放大这个问题的是 update_workflow 改图不重置 status:发布过 governed 之后把
画布改成含 supervisor 的图,界面上仍显示 governed。status 说的是当前画布的
状态,改图就退回 draft;已发布的那版由 published_version 指着,formal 不受影响
(所以 formal 的前置检查也从看 status 改成只看 published_version)。
## governed lint 只查 contract 这个 key 在不在
契约内容一律不校验,于是 contract={"metrics_from": "随便一个字符串"} 就满足了
"受管模板必须声明出具契约"这条硬门禁——一个结构上保证每次都出 formal 的空壳
模板能合法发布成受管模板。门禁写成这样还不如没有:它给的是"这张图过过闸"的
错觉。现在实质校验:metrics_from 必须指向图里真实存在的口径卡节点、narrative
必须非空、governed 下 required 必须非空、非 strict 给警告。
端到端验证:空壳契约发布被拒(4 条 issue)、改图后 status 退回 draft 且
published_version 保留、formal 跑 v2 被 409 拒、跑已发布的 v1 正常。
195 lines
7.5 KiB
Python
195 lines
7.5 KiB
Python
from __future__ import annotations
|
||
|
||
import asyncio
|
||
import sys
|
||
from typing import Any
|
||
|
||
from app.core.config import settings
|
||
from app.sandbox.base import ExecResult, Sandbox, SandboxLimits
|
||
from app.sandbox.bubblewrap_sandbox import BubblewrapSandbox
|
||
from app.sandbox.local_sandbox import LocalSandbox
|
||
from app.sandbox.microvm_sandbox import MicroVMSandbox
|
||
from app.sandbox.seatbelt_sandbox import SeatbeltSandbox
|
||
|
||
|
||
class DisabledSandbox(Sandbox):
|
||
name = "off"
|
||
|
||
async def run(self, code: str, **kwargs: Any) -> ExecResult: # type: ignore[override]
|
||
return ExecResult(
|
||
ok=False,
|
||
backend=self.name,
|
||
error="沙箱已在设置中关闭(AGENTLAB_SANDBOX_BACKEND=off)",
|
||
)
|
||
|
||
async def health(self) -> dict[str, Any]:
|
||
return {"backend": self.name, "available": False, "isolated": False, "reason": "已关闭"}
|
||
|
||
|
||
# 按隔离强度从高到低排,auto 模式挑第一个能用的。
|
||
# microVM 排在最前:它是唯一能真正限住内存、隔离进程表的一档。
|
||
_BACKENDS: list[tuple[str, type[Sandbox]]] = [
|
||
("microvm", MicroVMSandbox),
|
||
("bubblewrap", BubblewrapSandbox),
|
||
("seatbelt", SeatbeltSandbox),
|
||
("local", LocalSandbox),
|
||
]
|
||
|
||
|
||
class SandboxManager:
|
||
"""挑选并持有代码执行后端。
|
||
|
||
隔离强度从高到低:microVM(独立内核,限额真实生效)→ bubblewrap / Seatbelt
|
||
(共享内核的访问控制)→ 裸子进程。
|
||
|
||
auto 模式挑第一个**已就绪**的,判断比 available() 更严:microVM 还要求
|
||
OCI 镜像已经拉到本地(实测首次拉取约 55s)。所以镜像没缓存的机器会安静地
|
||
用 Seatbelt,而不是让第一次跑代码的人干等一分钟;拉过一次之后 auto 就会
|
||
自动升到 microVM,不用改配置。
|
||
|
||
显式指定(配置写死 microvm,或节点选了 strict 档)不受这条限制——
|
||
那是使用者自己要的硬隔离,该等就等。
|
||
"""
|
||
|
||
def __init__(self) -> None:
|
||
self._backend: Sandbox | None = None
|
||
self._named: dict[str, Sandbox] = {} # 按名字缓存,供节点级档位使用
|
||
self._lock = asyncio.Lock()
|
||
self._resolved_from: str = ""
|
||
|
||
async def get(self, prefer: str | None = None) -> Sandbox:
|
||
"""取后端。prefer 指定档位时给那一档,拿不到就退回默认那台。
|
||
|
||
节点级的"隔离档位"走这条路:strict 要硬件隔离、fast 要低延迟,
|
||
而整机默认仍由 auto 决定。
|
||
"""
|
||
if prefer:
|
||
resolved = self._resolve_named(prefer)
|
||
if resolved is not None:
|
||
return resolved
|
||
if self._backend is not None:
|
||
return self._backend
|
||
async with self._lock:
|
||
if self._backend is not None:
|
||
return self._backend
|
||
self._backend = self._resolve()
|
||
self._named[self._backend.name] = self._backend
|
||
return self._backend
|
||
|
||
def _resolve_named(self, prefer: str) -> Sandbox | None:
|
||
"""按档位名取一个**已就绪**的后端;不可用时返回 None 让调用方退回默认。"""
|
||
# 关掉就是关掉。这条必须在最前面:否则配置里 off 只挡得住默认路径,
|
||
# 图里任何一个 isolation=strict 的代码节点照样能跑代码,
|
||
# DisabledSandbox 那句"沙箱已在设置中关闭"根本到不了。
|
||
if (settings.sandbox_backend or "auto").lower() == "off":
|
||
return None
|
||
|
||
# 只认这两个档位。以前是 alias.get(prefer, prefer) 原样透传,而 isolation
|
||
# 是图 JSON 里的自由字符串,于是图里写 isolation="local" 就能点名那个
|
||
# 明说了"没有任何访问控制"的兜底后端——档位是用来提高隔离的,不该能降级。
|
||
alias = {"strict": "microvm", "fast": "seatbelt" if sys.platform == "darwin" else "bubblewrap"}
|
||
name = alias.get(prefer)
|
||
if name is None:
|
||
return None
|
||
if name in self._named:
|
||
return self._named[name]
|
||
impl = dict(_BACKENDS).get(name)
|
||
if impl is None or not getattr(impl, "available", lambda: True)():
|
||
return None
|
||
backend = impl()
|
||
self._named[name] = backend
|
||
return backend
|
||
|
||
def _resolve(self) -> Sandbox:
|
||
choice = (settings.sandbox_backend or "auto").lower()
|
||
|
||
if choice == "off":
|
||
self._resolved_from = "配置为关闭"
|
||
return DisabledSandbox()
|
||
|
||
named = dict(_BACKENDS)
|
||
if choice in named:
|
||
impl = named[choice]
|
||
backend = impl()
|
||
available = getattr(impl, "available", lambda: True)()
|
||
self._resolved_from = (
|
||
f"配置指定 {choice}" if available
|
||
else f"配置指定 {choice},但当前环境不可用"
|
||
)
|
||
return backend
|
||
|
||
for name, impl in _BACKENDS:
|
||
if not getattr(impl, "available", lambda: True)():
|
||
continue
|
||
# auto 模式额外要求"开箱即用":microVM 镜像没拉过就先跳过
|
||
warm = getattr(impl, "image_ready", None)
|
||
if warm is not None and not warm():
|
||
self._resolved_from = f"{name} 可用但镜像未缓存,auto 先跳过"
|
||
continue
|
||
self._resolved_from = f"自动选择 {name}({_why(name)})"
|
||
return impl()
|
||
|
||
self._resolved_from = "没有可用的隔离后端,降级到本地子进程"
|
||
return LocalSandbox()
|
||
|
||
async def run(
|
||
self,
|
||
code: str,
|
||
*,
|
||
language: str = "python",
|
||
limits: SandboxLimits | None = None,
|
||
session_id: str | None = None,
|
||
env: dict[str, str] | None = None,
|
||
files: dict[str, str] | None = None,
|
||
backend: str | None = None,
|
||
) -> ExecResult:
|
||
impl = await self.get(backend)
|
||
return await impl.run(
|
||
code,
|
||
language=language,
|
||
limits=limits,
|
||
session_id=session_id,
|
||
env=env,
|
||
files=files,
|
||
)
|
||
|
||
async def health(self) -> dict[str, Any]:
|
||
backend = await self.get()
|
||
info = await backend.health()
|
||
info["selected_because"] = self._resolved_from
|
||
info["configured"] = settings.sandbox_backend
|
||
info["defaults"] = SandboxLimits().model_dump()
|
||
info["candidates"] = [
|
||
{
|
||
"name": name,
|
||
"available": getattr(impl, "available", lambda: True)(),
|
||
# 可用但没预热的(microVM 镜像未拉取)要单独标出来:
|
||
# 它不会被 auto 选中,但显式指定仍然能用
|
||
**({} if getattr(impl, "image_ready", None) is None
|
||
else {"warm": impl.image_ready()}),
|
||
}
|
||
for name, impl in _BACKENDS
|
||
]
|
||
return info
|
||
|
||
async def cleanup(self, session_id: str) -> None:
|
||
# 会话可能落在任意一档上(节点各自选了档位),全部清一遍
|
||
for impl in {id(b): b for b in [*self._named.values(), self._backend] if b}.values():
|
||
await impl.cleanup(session_id)
|
||
|
||
async def close(self) -> None:
|
||
for impl in {id(b): b for b in [*self._named.values(), self._backend] if b}.values():
|
||
await impl.close()
|
||
|
||
|
||
def _why(name: str) -> str:
|
||
return {
|
||
"microvm": "microsandbox 运行时已就绪,独立内核",
|
||
"seatbelt": "macOS 自带 sandbox-exec(microVM 未就绪)",
|
||
"bubblewrap": "检测到 bwrap",
|
||
"local": "无可用隔离原语",
|
||
}.get(name, "")
|
||
|
||
|
||
sandbox_manager = SandboxManager()
|