liumaolin
refactor(config): centralize configuration management in `project_config`
8f68d0a
"""
本地异步任务管理器
基于 asyncio.subprocess + SQLite 的本地任务队列实现。
适用于 macOS 本地训练和 Electron 集成场景。
"""
import asyncio
import json
import os
import sqlite3
import sys
import uuid
from datetime import datetime
from pathlib import Path
from typing import TYPE_CHECKING, Dict, Optional, AsyncGenerator, List
import aiosqlite
from project_config import settings, PROJECT_ROOT, get_pythonpath
from ..base import TaskQueueAdapter
if TYPE_CHECKING:
from ..base import DatabaseAdapter
# 进度消息标识符(与 run_pipeline.py 保持一致)
PROGRESS_PREFIX = "##PROGRESS##"
PROGRESS_SUFFIX = "##"
class AsyncTrainingManager(TaskQueueAdapter):
"""
基于 asyncio.subprocess 的异步任务管理器
特点:
1. 使用 asyncio.create_subprocess_exec() 异步启动训练子进程
2. 完全非阻塞,与 FastAPI 异步模型完美契合
3. SQLite 持久化任务状态,支持应用重启后恢复
4. 实时解析子进程输出获取进度
Example:
>>> manager = AsyncTrainingManager(db_path="./data/tasks.db")
>>> job_id = await manager.enqueue("task-123", {"exp_name": "test", ...})
>>>
>>> # 订阅进度
>>> async for progress in manager.subscribe_progress("task-123"):
... print(f"{progress['stage']}: {progress['progress']*100:.1f}%")
>>>
>>> # 取消任务
>>> await manager.cancel(job_id)
"""
def __init__(
self,
db_path: str = None,
max_concurrent: int = 1,
database_adapter: "DatabaseAdapter" = None
):
"""
初始化任务管理器
Args:
db_path: SQLite 数据库路径,默认使用 settings.SQLITE_PATH
max_concurrent: 最大并发任务数(本地通常为1)
database_adapter: 数据库适配器,用于同步更新 tasks 表
"""
self.db_path = db_path or str(settings.SQLITE_PATH)
self.max_concurrent = max_concurrent
self._database_adapter = database_adapter
# 运行时状态
self.running_processes: Dict[str, asyncio.subprocess.Process] = {} # task_id -> Process
self.progress_channels: Dict[str, asyncio.Queue] = {} # task_id -> Queue
self._running_count = 0
self._queue_lock = asyncio.Lock()
# task_id 到 job_id 的映射缓存
self._task_job_mapping: Dict[str, str] = {}
# 初始化数据库
self._init_db_sync()
def _init_db_sync(self) -> None:
"""同步初始化数据库(启动时调用)"""
Path(self.db_path).parent.mkdir(parents=True, exist_ok=True)
with sqlite3.connect(self.db_path) as conn:
conn.execute('''
CREATE TABLE IF NOT EXISTS task_queue (
job_id TEXT PRIMARY KEY,
task_id TEXT NOT NULL UNIQUE,
exp_name TEXT NOT NULL,
config TEXT NOT NULL,
status TEXT DEFAULT 'queued',
current_stage TEXT,
progress REAL DEFAULT 0,
overall_progress REAL DEFAULT 0,
message TEXT,
error_message TEXT,
created_at TEXT NOT NULL,
started_at TEXT,
completed_at TEXT
)
''')
conn.execute('CREATE INDEX IF NOT EXISTS idx_task_queue_status ON task_queue(status)')
conn.execute('CREATE INDEX IF NOT EXISTS idx_task_queue_task_id ON task_queue(task_id)')
conn.execute('CREATE INDEX IF NOT EXISTS idx_task_queue_created ON task_queue(created_at)')
conn.commit()
async def enqueue(self, task_id: str, config: Dict, priority: str = "normal") -> str:
"""
将任务加入队列并异步启动
Args:
task_id: 任务唯一标识
config: 任务配置,需包含:
- exp_name: 实验名称
- version: 模型版本
- stages: 阶段配置列表
priority: 优先级(当前实现忽略此参数)
Returns:
job_id: 作业ID
"""
job_id = str(uuid.uuid4())
exp_name = config.get("exp_name", "unknown")
# 持久化到 SQLite
async with aiosqlite.connect(self.db_path) as db:
await db.execute(
'''INSERT INTO task_queue
(job_id, task_id, exp_name, config, status, created_at)
VALUES (?, ?, ?, ?, 'queued', ?)''',
(job_id, task_id, exp_name, json.dumps(config, ensure_ascii=False),
datetime.utcnow().isoformat())
)
await db.commit()
# 缓存 task_id -> job_id 映射
self._task_job_mapping[task_id] = job_id
# 创建进度队列
self.progress_channels[task_id] = asyncio.Queue()
# 异步启动训练任务
asyncio.create_task(self._run_training_async(job_id, task_id, config))
return job_id
async def _run_training_async(self, job_id: str, task_id: str, config: Dict) -> None:
"""
异步执行训练 Pipeline
Args:
job_id: 作业ID
task_id: 任务ID
config: 任务配置
"""
config_path = None
try:
# 更新状态为 running
await self._update_status(
job_id,
status='running',
started_at=datetime.utcnow().isoformat()
)
await self._send_progress(task_id, {
"type": "progress",
"status": "running",
"message": "训练任务启动中...",
"progress": 0.0,
"overall_progress": 0.0,
})
# 写入临时配置文件
config_path = await self._write_config_file(task_id, config)
# 获取 run_pipeline.py 脚本路径
script_path = self._get_pipeline_script_path()
# 构建环境变量
env = os.environ.copy()
env['PYTHONPATH'] = get_pythonpath()
# 创建子进程
process = await asyncio.create_subprocess_exec(
sys.executable, script_path,
'--config', config_path,
'--task-id', task_id,
stdout=asyncio.subprocess.PIPE,
stderr=asyncio.subprocess.PIPE,
env=env,
cwd=str(PROJECT_ROOT),
)
self.running_processes[task_id] = process
self._running_count += 1
# 异步监控子进程输出
await self._monitor_process_output(task_id, job_id, process)
# 等待进程完成
returncode = await process.wait()
if returncode == 0:
await self._update_status(
job_id,
status='completed',
progress=1.0,
overall_progress=1.0,
message='训练完成',
completed_at=datetime.utcnow().isoformat()
)
await self._send_progress(task_id, {
"type": "progress",
"status": "completed",
"message": "训练完成",
"progress": 1.0,
"overall_progress": 1.0,
})
else:
# 尝试读取剩余的 stderr
stderr_data = await process.stderr.read()
error_msg = stderr_data.decode() if stderr_data else f"进程退出码: {returncode}"
await self._update_status(
job_id,
status='failed',
error_message=error_msg,
completed_at=datetime.utcnow().isoformat()
)
await self._send_progress(task_id, {
"type": "progress",
"status": "failed",
"message": f"训练失败: {error_msg[:200]}",
"error": error_msg,
})
except asyncio.CancelledError:
await self._update_status(
job_id,
status='cancelled',
message='任务已取消',
completed_at=datetime.utcnow().isoformat()
)
await self._send_progress(task_id, {
"type": "progress",
"status": "cancelled",
"message": "任务已取消",
})
except Exception as e:
error_msg = str(e)
await self._update_status(
job_id,
status='failed',
error_message=error_msg,
completed_at=datetime.utcnow().isoformat()
)
await self._send_progress(task_id, {
"type": "progress",
"status": "failed",
"message": f"任务执行出错: {error_msg}",
"error": error_msg,
})
finally:
# 清理
self.running_processes.pop(task_id, None)
self._running_count = max(0, self._running_count - 1)
# 清理临时配置文件
if config_path:
await self._cleanup_config_file(config_path)
async def _monitor_process_output(
self,
task_id: str,
job_id: str,
process: asyncio.subprocess.Process
) -> None:
"""
异步监控子进程输出并解析进度
Args:
task_id: 任务ID
job_id: 作业ID
process: 子进程对象
"""
async def read_stdout():
"""读取 stdout 并解析进度"""
while True:
line = await process.stdout.readline()
if not line:
break
text = line.decode('utf-8', errors='replace').strip()
if not text:
continue
# 检测进度标记
if text.startswith(PROGRESS_PREFIX) and text.endswith(PROGRESS_SUFFIX):
json_str = text[len(PROGRESS_PREFIX):-len(PROGRESS_SUFFIX)]
try:
progress_info = json.loads(json_str)
await self._handle_progress(task_id, job_id, progress_info)
except json.JSONDecodeError as e:
# 解析失败,作为普通日志处理
await self._send_progress(task_id, {
"type": "log",
"level": "warning",
"message": f"进度解析失败: {e}",
})
else:
# 普通输出,作为日志处理
await self._send_progress(task_id, {
"type": "log",
"level": "info",
"message": text,
})
async def read_stderr():
"""读取 stderr 作为错误日志"""
while True:
line = await process.stderr.readline()
if not line:
break
text = line.decode('utf-8', errors='replace').strip()
if text:
await self._send_progress(task_id, {
"type": "log",
"level": "error",
"message": text,
})
# 并发读取 stdout 和 stderr
await asyncio.gather(
read_stdout(),
read_stderr(),
return_exceptions=True
)
async def _handle_progress(
self,
task_id: str,
job_id: str,
progress_info: Dict
) -> None:
"""
处理进度信息
Args:
task_id: 任务ID
job_id: 作业ID
progress_info: 进度信息字典
"""
# 发送到订阅者
await self._send_progress(task_id, progress_info)
# 更新数据库中的进度
updates = {}
if 'stage' in progress_info:
updates['current_stage'] = progress_info['stage']
if 'progress' in progress_info:
updates['progress'] = progress_info['progress']
if 'overall_progress' in progress_info:
updates['overall_progress'] = progress_info['overall_progress']
if 'message' in progress_info:
updates['message'] = progress_info['message']
if 'status' in progress_info:
updates['status'] = progress_info['status']
if 'error' in progress_info:
updates['error_message'] = progress_info['error']
if updates:
await self._update_status(job_id, **updates)
async def _send_progress(self, task_id: str, progress_info: Dict) -> None:
"""
发送进度到订阅队列
Args:
task_id: 任务ID
progress_info: 进度信息
"""
if task_id in self.progress_channels:
# 添加时间戳
if 'timestamp' not in progress_info:
progress_info['timestamp'] = datetime.utcnow().isoformat()
await self.progress_channels[task_id].put(progress_info)
async def _update_status(self, job_id: str, **kwargs) -> None:
"""
更新任务状态
同时更新 task_queue 表和 tasks 表(通过 DatabaseAdapter)。
Args:
job_id: 作业ID
**kwargs: 要更新的字段
"""
if not kwargs:
return
# 1. 更新 task_queue 表
task_id = None
async with aiosqlite.connect(self.db_path) as db:
updates = []
values = []
for key, value in kwargs.items():
updates.append(f"{key} = ?")
values.append(value)
values.append(job_id)
await db.execute(
f"UPDATE task_queue SET {', '.join(updates)} WHERE job_id = ?",
values
)
await db.commit()
# 获取 task_id 用于同步更新 tasks 表
async with db.execute(
"SELECT task_id FROM task_queue WHERE job_id = ?", (job_id,)
) as cursor:
row = await cursor.fetchone()
if row:
task_id = row[0]
# 2. 同步更新 tasks 表(通过 DatabaseAdapter)
if self._database_adapter and task_id:
await self._sync_to_tasks_table(task_id, kwargs)
async def _sync_to_tasks_table(self, task_id: str, updates: Dict) -> None:
"""
同步状态更新到 tasks 表
字段映射:
- task_queue.progress -> tasks.stage_progress
- task_queue.overall_progress -> tasks.progress
- 其他字段直接映射
Args:
task_id: 任务ID
updates: 要更新的字段字典
"""
if not self._database_adapter:
return
# 字段映射
tasks_updates = {}
for key, value in updates.items():
if key == 'progress':
# task_queue.progress -> tasks.stage_progress
tasks_updates['stage_progress'] = value
elif key == 'overall_progress':
# task_queue.overall_progress -> tasks.progress
tasks_updates['progress'] = value
elif key in ('status', 'current_stage', 'message', 'error_message',
'started_at', 'completed_at'):
# 直接映射的字段
tasks_updates[key] = value
if tasks_updates:
try:
await self._database_adapter.update_task(task_id, tasks_updates)
except Exception as e:
# 记录错误但不中断主流程
import logging
logging.warning(f"Failed to sync task status to tasks table: {e}")
async def get_status(self, job_id: str) -> Dict:
"""
获取任务状态
Args:
job_id: 作业ID
Returns:
状态字典
"""
async with aiosqlite.connect(self.db_path) as db:
db.row_factory = aiosqlite.Row
async with db.execute(
"SELECT * FROM task_queue WHERE job_id = ?", (job_id,)
) as cursor:
row = await cursor.fetchone()
if row:
return dict(row)
return {"status": "not_found", "message": "任务不存在"}
async def get_status_by_task_id(self, task_id: str) -> Dict:
"""
通过 task_id 获取任务状态
Args:
task_id: 任务ID
Returns:
状态字典
"""
async with aiosqlite.connect(self.db_path) as db:
db.row_factory = aiosqlite.Row
async with db.execute(
"SELECT * FROM task_queue WHERE task_id = ?", (task_id,)
) as cursor:
row = await cursor.fetchone()
if row:
return dict(row)
return {"status": "not_found", "message": "任务不存在"}
async def cancel(self, job_id: str) -> bool:
"""
取消任务
Args:
job_id: 作业ID
Returns:
是否成功取消
"""
# 查找 task_id
async with aiosqlite.connect(self.db_path) as db:
async with db.execute(
"SELECT task_id, status FROM task_queue WHERE job_id = ?", (job_id,)
) as cursor:
row = await cursor.fetchone()
if not row:
return False
task_id, status = row
# 如果任务已经完成,无法取消
if status in ('completed', 'failed', 'cancelled'):
return False
# 终止进程
if task_id in self.running_processes:
process = self.running_processes[task_id]
# 先尝试优雅终止
process.terminate()
try:
# 等待进程终止
await asyncio.wait_for(process.wait(), timeout=5.0)
except asyncio.TimeoutError:
# 超时则强制终止
process.kill()
await process.wait()
return True
# 如果进程不在运行(可能还在队列中),直接更新状态
await self._update_status(
job_id,
status='cancelled',
message='任务已取消',
completed_at=datetime.utcnow().isoformat()
)
# 通知订阅者
if task_id in self.progress_channels:
await self._send_progress(task_id, {
"type": "progress",
"status": "cancelled",
"message": "任务已取消",
})
return True
async def subscribe_progress(self, task_id: str) -> AsyncGenerator[Dict, None]:
"""
订阅任务进度(用于 SSE 流)
Args:
task_id: 任务ID
Yields:
进度信息字典
"""
# 确保队列存在
if task_id not in self.progress_channels:
self.progress_channels[task_id] = asyncio.Queue()
queue = self.progress_channels[task_id]
# 首先发送当前状态
status = await self.get_status_by_task_id(task_id)
if status.get("status") != "not_found":
yield {
"type": "progress",
"status": status.get("status"),
"stage": status.get("current_stage"),
"progress": status.get("progress", 0.0),
"overall_progress": status.get("overall_progress", 0.0),
"message": status.get("message"),
"timestamp": datetime.utcnow().isoformat(),
}
# 持续接收进度更新
while True:
try:
# 30秒超时,发送心跳
progress = await asyncio.wait_for(queue.get(), timeout=30.0)
yield progress
# 检查是否为终态
if progress.get('status') in ('completed', 'failed', 'cancelled'):
break
except asyncio.TimeoutError:
# 发送心跳保持连接
yield {
"type": "heartbeat",
"timestamp": datetime.utcnow().isoformat(),
}
async def list_tasks(
self,
status: Optional[str] = None,
limit: int = 50,
offset: int = 0
) -> List[Dict]:
"""
列出任务
Args:
status: 按状态筛选
limit: 返回数量限制
offset: 偏移量
Returns:
任务列表
"""
async with aiosqlite.connect(self.db_path) as db:
db.row_factory = aiosqlite.Row
if status:
query = """
SELECT * FROM task_queue
WHERE status = ?
ORDER BY created_at DESC
LIMIT ? OFFSET ?
"""
params = (status, limit, offset)
else:
query = """
SELECT * FROM task_queue
ORDER BY created_at DESC
LIMIT ? OFFSET ?
"""
params = (limit, offset)
async with db.execute(query, params) as cursor:
rows = await cursor.fetchall()
return [dict(row) for row in rows]
async def recover_pending_tasks(self) -> int:
"""
应用重启后恢复未完成的任务
将 running 状态的任务标记为 interrupted,
可选择重新启动 queued 状态的任务。
Returns:
恢复的任务数量
"""
async with aiosqlite.connect(self.db_path) as db:
# 将 running 状态的任务标记为 interrupted
await db.execute(
"""UPDATE task_queue
SET status = 'interrupted',
message = '应用重启导致任务中断'
WHERE status = 'running'"""
)
await db.commit()
# 获取 queued 状态的任务
db.row_factory = aiosqlite.Row
async with db.execute(
"SELECT * FROM task_queue WHERE status = 'queued' ORDER BY created_at"
) as cursor:
queued_tasks = await cursor.fetchall()
# 重新启动 queued 状态的任务
recovered = 0
for task in queued_tasks:
task_id = task['task_id']
job_id = task['job_id']
config = json.loads(task['config'])
self.progress_channels[task_id] = asyncio.Queue()
asyncio.create_task(self._run_training_async(job_id, task_id, config))
recovered += 1
return recovered
async def cleanup_old_tasks(self, days: int = 7) -> int:
"""
清理旧任务记录
Args:
days: 保留天数
Returns:
删除的任务数量
"""
from datetime import timedelta
cutoff = (datetime.utcnow() - timedelta(days=days)).isoformat()
async with aiosqlite.connect(self.db_path) as db:
cursor = await db.execute(
"""DELETE FROM task_queue
WHERE status IN ('completed', 'failed', 'cancelled')
AND completed_at < ?""",
(cutoff,)
)
deleted = cursor.rowcount
await db.commit()
return deleted
def _get_pipeline_script_path(self) -> str:
"""获取 run_pipeline.py 脚本路径"""
return str(settings.PIPELINE_SCRIPT_PATH)
async def _write_config_file(self, task_id: str, config: Dict) -> str:
"""
写入临时配置文件
Args:
task_id: 任务ID
config: 配置字典
Returns:
配置文件路径
"""
config_path = settings.CONFIGS_DIR / f"{task_id}.json"
with open(config_path, 'w', encoding='utf-8') as f:
json.dump(config, f, ensure_ascii=False, indent=2)
return str(config_path)
async def _cleanup_config_file(self, config_path: str) -> None:
"""
清理临时配置文件
Args:
config_path: 配置文件路径
"""
try:
path = Path(config_path)
if path.exists():
path.unlink()
except Exception:
pass # 忽略清理错误