zhaoxi/KiloStar
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: 新增工具插件、系统日志、workflow配置及前端优化

Browse Source 
1. 新增工具插件(edit_file, python_executor, search_file, shell_executor, write_file)
2. 新增系统事件日志模块和API
3. 新增workflow配置文件和详情API
4. 前端增加SSE、错误边界、设置引导等组件
5. 优化认证加密、速率限制、配置加载等工具模块
6. 删除废弃的cluster和health API
7. 补充单元测试和集成测试

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
zhaoxi
2026-06-03 07:34:43 +00:00
parent f04fef916f
commit a53ffebe0e
57 changed files with 2804 additions and 271 deletions
Download Patch File Download Diff File Expand all files Collapse all files
kilostar/core/global_state_machine/global_state_machine.py
+4
View File
@@ -203,6 +203,10 @@ class GlobalStateMachine:
"""返回某个 scope 下的"系统 + 自定义工具组"toolset 列表（不含 MCP）。"""
return self._global_tool_manager.get_toolsets_for_scope(scope)
def get_retrieval_toolsets_for_scope(self, scope: str) -> List[Any]:
"""仅返回 retrieval 工具集（system_node 专用，不包含 generation 工具）。"""
return self._global_tool_manager.get_retrieval_toolsets_for_scope(scope)
# ─── MCP Server Registry ───────────────────────────────────
async def add_mcp_server(self, server_id: str, config: Dict[str, Any]) -> bool:
kilostar/core/global_state_machine/tool_manager.py
+32
View File
@@ -34,7 +34,9 @@ class GlobalToolManager:
def __init__(self) -> None:
self.tool_metadata = {}
self._tool_funcs = defaultdict(dict)
self._retrieval_tool_funcs = defaultdict(dict)
self._system_toolsets = {}
self._retrieval_toolsets = {}
self._custom_toolsets = {}
self._third_party_funcs = {}
self.tool_mapper = defaultdict(dict)
@@ -75,11 +77,14 @@ class GlobalToolManager:
is_system = bool(tool_data_cls.model_fields.get("is_system").default)
category_field = tool_data_cls.model_fields.get("category")
category = (category_field.default if category_field else "other") or "other"
toolset_field = tool_data_cls.model_fields.get("toolset")
toolset_name = (toolset_field.default if toolset_field else "other") or "other"
self.tool_metadata[plugin_name] = {
"name": plugin_name,
"is_system": is_system,
"category": category,
"toolset": toolset_name,
"action_scope": list(action_scopes),
}
@@ -92,12 +97,15 @@ class GlobalToolManager:
for scope in scopes:
self._tool_funcs[scope][plugin_name] = tool_func
self.tool_mapper[scope][plugin_name] = tool_data_cls
if toolset_name == "retrieval":
self._retrieval_tool_funcs[scope][plugin_name] = tool_func
else:
self._third_party_funcs[plugin_name] = tool_func
for scope in scopes:
self.tool_mapper[scope][plugin_name] = tool_data_cls
self._build_system_toolsets()
self._build_retrieval_toolsets()
def _build_system_toolsets(self) -> None:
FunctionToolset = self._import_function_toolset()
@@ -114,6 +122,21 @@ class GlobalToolManager:
except Exception as e:
logger.error(f"Failed to build system toolset {scope}: {e}")
def _build_retrieval_toolsets(self) -> None:
FunctionToolset = self._import_function_toolset()
if FunctionToolset is None:
return
for scope, name_to_func in self._retrieval_tool_funcs.items():
if not name_to_func:
continue
try:
self._retrieval_toolsets[scope] = FunctionToolset(
tools=list(name_to_func.values()),
id=f"retrieval::{scope}",
)
except Exception as e:
logger.error(f"Failed to build retrieval toolset {scope}: {e}")
def rebuild_custom_toolsets(self, custom_defs: Dict[str, Dict[str, Any]]) -> None:
"""根据 DB 中的自定义工具组定义重建 custom FunctionToolset。"""
FunctionToolset = self._import_function_toolset()
@@ -170,6 +193,15 @@ class GlobalToolManager:
result.extend(self._custom_toolsets.values())
return result
def get_retrieval_toolsets_for_scope(self, scope: str) -> List[Any]:
"""仅返回 retrieval 工具集（system_node 专用）。"""
result: List[Any] = []
for s in ("default", scope):
ts = self._retrieval_toolsets.get(s)
if ts is not None:
result.append(ts)
return result
# ─── Metadata accessors ───
def is_third_party_tool(self, tool_name: str) -> bool:
kilostar/core/postgres_database/model/__init__.py
+2
View File
@@ -33,6 +33,7 @@ from kilostar.core.postgres_database.model.system_node import SystemNodeConfigMo
from kilostar.core.postgres_database.model.mcp_server import MCPServerModel
from kilostar.core.postgres_database.model.tool_config import ToolConfigModel
from kilostar.core.postgres_database.model.custom_toolset import CustomToolsetModel
from kilostar.core.postgres_database.model.system_event_log import SystemEventLog
# 兼容旧代码的别名
Provider = ProviderModel
@@ -61,5 +62,6 @@ __all__ = [
"MCPServerModel",
"ToolConfigModel",
"CustomToolsetModel",
"SystemEventLog",
"AgentType",
]
kilostar/core/postgres_database/model/system_event_log.py
+35
View File
@@ -0,0 +1,35 @@
from sqlalchemy import String, DateTime, Integer, func, Text
from sqlalchemy.orm import Mapped, mapped_column
from sqlalchemy.dialects.postgresql import JSONB
from .base import BaseDataModel
class SystemEventLog(BaseDataModel):
__tablename__ = "system_event_log"
id: Mapped[int] = mapped_column(
Integer, primary_key=True, autoincrement=True
)
trace_id: Mapped[str] = mapped_column(
String(64), index=True, comment="关联的工作流 trace_id"
)
event_type: Mapped[str] = mapped_column(
String(50), index=True,
comment="事件类型: workflow_start/step_enter/step_complete/step_error/workflow_complete/workflow_fail/system"
)
level: Mapped[str] = mapped_column(
String(10), index=True, default="info",
comment="日志级别: info/warn/error"
)
node_name: Mapped[str | None] = mapped_column(
String(100), nullable=True, comment="相关节点名称"
)
message: Mapped[str] = mapped_column(
Text, comment="日志消息正文"
)
extra_data: Mapped[dict | None] = mapped_column(
JSONB, nullable=True, comment="附加元数据（step_index/output 等）"
)
created_at: Mapped[str] = mapped_column(
DateTime(timezone=True), server_default=func.now(), index=True
)
kilostar/core/postgres_database/module/system_event_log.py
+72
View File
@@ -0,0 +1,72 @@
from __future__ import annotations
from typing import List, Optional
from sqlalchemy import select, desc
from sqlalchemy.ext.asyncio import async_sessionmaker, AsyncSession
from kilostar.core.postgres_database.model.system_event_log import SystemEventLog
from kilostar.core.postgres_database.database_exception import database_exception
class SystemEventLogDatabase:
def __init__(self, async_session_maker: async_sessionmaker[AsyncSession]):
self.async_session_maker = async_session_maker
@database_exception
async def insert_event(
self,
trace_id: str,
event_type: str,
level: str,
message: str,
node_name: Optional[str] = None,
metadata: Optional[dict] = None,
) -> None:
async with self.async_session_maker() as session:
log = SystemEventLog(
trace_id=trace_id,
event_type=event_type,
level=level,
message=message,
node_name=node_name,
extra_data=metadata,
)
session.add(log)
await session.commit()
@database_exception
async def query_events(
self,
trace_id: Optional[str] = None,
event_type: Optional[str] = None,
level: Optional[str] = None,
limit: int = 100,
offset: int = 0,
) -> List[dict]:
async with self.async_session_maker() as session:
stmt = select(SystemEventLog).order_by(desc(SystemEventLog.created_at))
if trace_id:
stmt = stmt.where(SystemEventLog.trace_id == trace_id)
if event_type:
stmt = stmt.where(SystemEventLog.event_type == event_type)
if level:
stmt = stmt.where(SystemEventLog.level == level)
stmt = stmt.offset(offset).limit(limit)
result = await session.execute(stmt)
rows = result.scalars().all()
return [
{
"id": r.id,
"trace_id": r.trace_id,
"event_type": r.event_type,
"level": r.level,
"node_name": r.node_name,
"message": r.message,
"metadata": r.extra_data,
"created_at": str(r.created_at) if r.created_at else None,
}
for r in rows
]
kilostar/core/postgres_database/postgres.py
+36 -2
View File
@@ -41,6 +41,7 @@ from kilostar.core.postgres_database.model.system_node import SystemNodeConfigMo
from kilostar.core.postgres_database.model.mcp_server import MCPServerModel
from kilostar.core.postgres_database.model.tool_config import ToolConfigModel
from kilostar.core.postgres_database.model.custom_toolset import CustomToolsetModel
from kilostar.core.postgres_database.model.system_event_log import SystemEventLog
from .module.individual import IndividualDatabase
from .module.user import AuthDatabase
@@ -51,6 +52,7 @@ from .module.chat_history import ChatHistoryDatabase
from .module.mcp_server import MCPServerDatabase
from .module.tool_config import ToolConfigDatabase
from .module.custom_toolset import CustomToolsetDatabase
from .module.system_event_log import SystemEventLogDatabase
@ray.remote
@@ -85,6 +87,7 @@ class PostgresDatabase:
self._mcp_server_database = MCPServerDatabase(self.async_session_maker)
self._tool_config_database = ToolConfigDatabase(self.async_session_maker)
self._custom_toolset_database = CustomToolsetDatabase(self.async_session_maker)
self._system_event_log_database = SystemEventLogDatabase(self.async_session_maker)
self.ready_event = asyncio.Event()
@@ -94,11 +97,10 @@ class PostgresDatabase:
async with self.async_engine.begin() as conn:
await conn.run_sync(BaseDataModel.metadata.create_all)
print("✅ 数据库表创建/验证完成")
self.ready_event.set()
except Exception as e:
print(f"❌ 数据库初始化失败: {e}")
raise
finally:
self.ready_event.set()
async def ping(self) -> bool:
"""轻量探活：等待 ready 后执行 ``SELECT 1``。"""
@@ -376,3 +378,35 @@ class PostgresDatabase:
"""删除一个自定义工具组。"""
await self.ready_event.wait()
return await self._custom_toolset_database.delete(toolset_id)
# System Event Log Methods
async def insert_event_log(
self,
trace_id: str,
event_type: str,
level: str,
message: str,
node_name=None,
metadata=None,
):
await self.ready_event.wait()
return await self._system_event_log_database.insert_event(
trace_id=trace_id,
event_type=event_type,
level=level,
message=message,
node_name=node_name,
metadata=metadata,
)
async def query_event_logs(
self, trace_id=None, event_type=None, level=None, limit=100, offset=0
):
await self.ready_event.wait()
return await self._system_event_log_database.query_events(
trace_id=trace_id,
event_type=event_type,
level=level,
limit=limit,
offset=offset,
)
kilostar/core/work/workflow/workflow_engine.py
+52 -6
View File
@@ -61,6 +61,7 @@ class WorkflowGraphState(BaseModel):
# 已发过 put_pending 的 HumanApproval step index 列表；resume 后避免重复推送。
# 用 list（不是 set）是为了 pydantic_graph 序列化 history 时 JSON 友好。
approvals_notified: List[int] = Field(default_factory=list)
jump_counts: Dict[str, int] = Field(default_factory=dict)
# 业务侧执行入口：把 step + state 喂进去，拿到 (output_text, success_bool)
@@ -277,8 +278,13 @@ async def _execute_step(
)
try:
output_text, success = await executor(step_data, state)
except Exception as e: # 执行器抛异常 → 走失败分支
step_timeout = step_data.get("timeout", 300)
output_text, success = await asyncio.wait_for(
executor(step_data, state), timeout=step_timeout
)
except asyncio.TimeoutError:
output_text, success = f"步骤执行超时（{step_data.get('timeout', 300)}s）", False
except Exception as e:
output_text, success = str(e), False
if success:
@@ -311,6 +317,25 @@ async def _execute_step(
logic_gate = step_data.get("logic_gate") or {}
fail_target = logic_gate.get("if_fail")
if fail_target and "jump_to_step_" in fail_target:
from kilostar.utils.config_loader import get_workflow_config
max_attempts = get_workflow_config().retry.max_attempts
jump_key = f"{state.current_step_index}->{fail_target}"
state.jump_counts[jump_key] = state.jump_counts.get(jump_key, 0) + 1
if state.jump_counts[jump_key] > max_attempts:
state.logs.append(
{
str(state.current_step_index): [
str(datetime.datetime.now()),
"failed",
f"环重试次数超过上限 ({max_attempts})，终止工作流",
]
}
)
await _persist_context(ctx, status=WorkflowStatus.FAILED.value)
return Finalize(status=WorkflowStatus.FAILED.value)
target_step = int(fail_target.split("_")[-1]) - 1
state.current_step_index = target_step
await _persist_context(ctx, status=WorkflowStatus.RUNNING.value)
@@ -495,14 +520,21 @@ async def resume_workflow_graph(
@ray.remote
def run_workflow_task(workflow_data: dict, trace_id: str):
def run_workflow_task(
workflow_data: dict, trace_id: str, resume_only: bool = False
):
"""workflow 的 ray task 入口：一次性执行，跑完即销毁。
生产路径下持久化交给 ``PostgresStatePersistence`` —— 即便进程崩溃，再 fire
一次相同 ``trace_id`` 的任务（或调 ``/workflow/{trace_id}/resume``）即可
续跑。同时为了支持 fresh start，先尝试 ``hydrate``：
续跑。入口先尝试 ``hydrate``：
- hydrate 拿到内容 → 走 resume 路径
- hydrate 没拿到 → 走全新路径
- hydrate 没拿到 → 走全新路径（用传入的 ``workflow_data``）
``resume_only``：由 ``/resume`` API 显式置 True。此模式下 hydrate 失败
（抛异常或没有持久化记录）必须 fail-fast，而不能 fall through 到全新路径——
否则会拿着空 ``workflow_data`` 空跑一个 ``work_link=[]`` 的 workflow 并误判
为 COMPLETED（静默 bug）。
ray task 是新进程，contextvars 不会从 caller 传过来，所以入口先 bind 一次
``trace_id``，让节点内的日志自动带上它。
@@ -511,6 +543,9 @@ def run_workflow_task(workflow_data: dict, trace_id: str):
from kilostar.core.work.workflow.graph_persistence import (
build_postgres_persistence,
)
from kilostar.utils.logger import get_logger
_logger = get_logger("workflow_task")
async def _entry() -> None:
with trace_id_scope(trace_id):
@@ -519,9 +554,20 @@ def run_workflow_task(workflow_data: dict, trace_id: str):
recovered = False
try:
recovered = await persistence.hydrate()
except Exception: # pragma: no cover - 防御
except Exception as e:
if resume_only:
_logger.error(f"resume 失败：无法 hydrate 图持久化记录: {e}")
raise
recovered = False
if resume_only and not recovered:
msg = (
f"resume 失败：trace {trace_id} 没有可恢复的图持久化记录，"
"拒绝以全新模式空跑"
)
_logger.error(msg)
raise RuntimeError(msg)
if recovered:
await resume_workflow_graph(trace_id, persistence=persistence)
else: