全栈安全加固、性能提速与架构演进:连接池与沙箱止血、双路混合检索、多版本归档及前端分包治理

- 第一批(安全加固与生命周期止血):隔离测试宿主机密钥与 .env 泄漏;修复 SQLite 取消查询挂死、:memory: 连接池报错与 EngineCache 世代竞态;补齐运行与评测删除时的 Checkpointer 线程、Artifact 磁盘文件与无主附件回收;加固三套沙箱路径越界、进程组清理与 Seatbelt 数据目录隔离,封堵 SSRF IPv4-mapped IPv6 元数据绕过与响应体 OOM;修复推理模型多轮 reasoning_content、结构化输出 Token 漏计与缓存折扣计费
- 第二批(数据库/集群/问数提速与首屏分包):补齐核心表索引与 SQLite WAL 调优,列表接口按需 defer 大列,收敛启动扫描并打通跨进程 slot 槽位唤醒;优化跨源全局选表、哈希桶数字绑定、流式 CSV 导入及多方言分析算子;剥离首屏静态引入的 @xyflow/react 与全局样式(首屏主包降至 461.59 kB),补充流式 rAF 攒批、Store 精细订阅与表单无障碍绑定
- 第三批(架构演进与工程自动化闭环):记录库超挂接阈值历史版本自动增量归档至单库多表(解除 SQLite 10 库 ATTACH 硬上限);重构知识库为词面 BM25 与全集合归一化向量矩阵真·双路独立召回并支持游标分批重建;升级多方言感知 SQL 模板词法分析(E'...' / $$...$$ / q'[...]' / # / 嵌套注释)与 Decimal / bool / NaN 参数校验,消除 ledger 层对 api 路由层的反向依赖;拆分 3101 行 CaliberDrafts 页面为 7 个领域子模块,消除检查脚本本机路径硬编码并新增 GitHub Actions CI 门禁与三批回归测试套件
This commit is contained in:
ThearchyHelios
2026-11-13 14:51:36 +08:00
parent 09949fc8ac
commit eddeded2a4
106 changed files with 7178 additions and 3913 deletions
+46 -31
View File
@@ -76,31 +76,38 @@ async def put_json(
幂等:同内容重复写只落一次盘;同一工件被多个 run 引用时,
数据库里各记一行引用(溯源需要知道"谁产出的"),文件只有一份。
"""
from sqlalchemy.exc import IntegrityError
from app.cluster.mutex import lock
from app.db.base import SessionLocal
from app.db.models import Artifact
text = canonical_json(obj)
artifact_id = content_hash(text)
path = _path_of(artifact_id)
if not path.exists():
# 原子落盘(core/atomic):写一半崩了不会留下半个工件;临时文件名带随机后缀,两个进程同时写同一份内容
# 各写各的,不会一个把另一个写到一半的截断(以前共用 <id>.tmp,后到的会把先到的半成品改名成正式文件)
atomic.write_text(path, text)
async with lock(f"artifact:{artifact_id[:2]}"):
if not path.exists():
# 原子落盘(core/atomic):写一半崩了不会留下半个工件;临时文件名带随机后缀,两个进程同时写同一份内容
# 各写各的,不会一个把另一个写到一半的截断;且与同分桶的 forget 互斥,不会刚写下就被并发回收删掉
atomic.write_text(path, text)
async with SessionLocal() as session:
existing = await session.get(Artifact, (artifact_id, run_id or ""))
if existing is None:
session.add(
Artifact(
id=artifact_id,
run_id=run_id or "",
kind=kind,
node_id=node_id,
size=len(text.encode()),
meta=meta or {},
async with SessionLocal() as session:
existing = await session.get(Artifact, (artifact_id, run_id or ""))
if existing is None:
session.add(
Artifact(
id=artifact_id,
run_id=run_id or "",
kind=kind,
node_id=node_id,
size=len(text.encode()),
meta=meta or {},
)
)
)
await session.commit()
try:
await session.commit()
except IntegrityError:
await session.rollback()
return artifact_id
@@ -148,27 +155,35 @@ async def forget(ids: Any) -> list[str]:
读到错的内容。"""
from sqlalchemy import select
from app.cluster.mutex import lock
from app.db.base import SessionLocal
from app.db.models import Artifact
wanted = sorted({i for i in ids or () if isinstance(i, str) and i and i.isalnum()})
if not wanted:
return []
async with SessionLocal() as session:
kept: set[str] = set()
for start in range(0, len(wanted), 500):
chunk = wanted[start:start + 500]
kept |= set((await session.execute(select(Artifact.id).where(Artifact.id.in_(chunk)))).scalars())
removed: list[str] = []
by_bucket: dict[str, list[str]] = {}
for artifact_id in wanted:
if artifact_id in kept:
continue
path = _path_of(artifact_id)
try:
path.unlink()
except FileNotFoundError:
continue
removed.append(artifact_id)
by_bucket.setdefault(artifact_id[:2], []).append(artifact_id)
removed: list[str] = []
for bucket in sorted(by_bucket):
bucket_ids = by_bucket[bucket]
async with lock(f"artifact:{bucket}"):
async with SessionLocal() as session:
kept: set[str] = set()
for start in range(0, len(bucket_ids), 500):
chunk = bucket_ids[start:start + 500]
kept |= set((await session.execute(select(Artifact.id).where(Artifact.id.in_(chunk)))).scalars())
for artifact_id in bucket_ids:
if artifact_id in kept:
continue
path = _path_of(artifact_id)
try:
path.unlink()
except FileNotFoundError:
continue
removed.append(artifact_id)
return removed
+20 -5
View File
@@ -2,7 +2,7 @@ from __future__ import annotations
import asyncio
import contextlib
from collections import defaultdict
from collections import OrderedDict, defaultdict
from collections.abc import Callable
from typing import AsyncIterator
@@ -11,6 +11,7 @@ from app.core.events import RunEventModel
# 单个订阅者的积压上限。慢消费者(比如卡住的浏览器标签)不应该拖垮整个 run,
# 所以队列满了就丢最旧的事件并打标记,而不是阻塞发布方。
_QUEUE_MAXSIZE = 2048
_MAX_SEQ_ENTRIES = 4096
class _Subscriber:
@@ -31,17 +32,30 @@ class EventBus:
def __init__(self) -> None:
self._subs: dict[str, set[_Subscriber]] = defaultdict(set)
self._seq: dict[str, int] = defaultdict(int)
self._seq: OrderedDict[str, int] = OrderedDict()
self._lock = asyncio.Lock()
self.tap: Callable[[RunEventModel], None] | None = None
def _trim_seq(self) -> None:
while len(self._seq) > _MAX_SEQ_ENTRIES:
oldest, _ = next(iter(self._seq.items()))
if self._subs.get(oldest):
self._seq.move_to_end(oldest)
break
self._seq.popitem(last=False)
def next_seq(self, run_id: str) -> int:
self._seq[run_id] += 1
return self._seq[run_id]
nxt = self._seq.get(run_id, 0) + 1
self._seq[run_id] = nxt
self._seq.move_to_end(run_id)
self._trim_seq()
return nxt
def set_seq(self, run_id: str, value: int) -> None:
"""恢复一个已存在的 run 时,把序号接上,保证前端排序不乱。"""
self._seq[run_id] = max(self._seq[run_id], value)
self._seq[run_id] = max(self._seq.get(run_id, 0), value)
self._seq.move_to_end(run_id)
self._trim_seq()
async def publish(self, event: RunEventModel, *, relayed: bool = False) -> None:
if not relayed and self.tap is not None:
@@ -61,6 +75,7 @@ class EventBus:
async def close(self, run_id: str) -> None:
"""给所有订阅者发结束哨兵,让 WebSocket 干净地收尾。"""
self._seq.pop(run_id, None)
for sub in list(self._subs.get(run_id, ())):
with contextlib.suppress(asyncio.QueueFull):
sub.queue.put_nowait(None)
+4 -1
View File
@@ -361,7 +361,10 @@ class Settings(BaseSettings):
@lru_cache
def get_settings() -> Settings:
s = Settings()
import os
no_dotenv = os.environ.get("AGENTLAB_DISABLE_DOTENV", "").strip().lower() in ("1", "true", "yes", "on")
s = Settings(_env_file=None) if no_dotenv else Settings()
s.ensure_dirs()
return s