Skip to main content

HuggingFace Space Worker 分布式调度系统

Project description

HFS v2 - HuggingFace Space Worker 分布式调度系统

基于 Redis 的分布式 Worker 调度系统,用于管理 HuggingFace Space 资源池。

特性

  • 分布式调度 - 多 Worker 并发,自动负载均衡
  • 状态管理 - 原子操作(Lua 脚本),保证一致性
  • 健康检查 - 自动检测崩溃、超时、孤儿资源
  • 账号管理 - 多账号池,自动选择、cooldown、评分
  • Space 轮换 - 自动创建、绑定、轮换、复用
  • 场景配置 - 内置多种场景,支持自定义
  • Admin CLI - 命令行管理工具

快速开始

1. 安装

pip install mp-hfs

2. 配置环境(可选)

CLI 已内置默认 Redis URL,可直接使用。如需自定义:

export HFS_REDIS_URL="redis://:password@host:port/db"

3. 初始化账号池

# 添加账号
hfs-admin account create "hf_xxxxx" --max-spaces=6

# 查看账号
hfs-admin account list

参考 使用指南

4. 创建项目

创建 my-project.yaml

project:
  id: "my-project"
  scene: "production"
  required_nodes: 3
  
  start_script:
    type: inline
    inline: "python -m my_app"

nodes:
  ids: ["node-1", "node-2", "node-3"]
  # 或不指定,自动生成:my-project-1, my-project-2, my-project-3

初始化:

hfs-admin project init my-project.yaml
hfs-admin project bootstrap my-project

5. 监控

# 查看节点状态
hfs-admin node list --project my-project

# 查看 Space 状态
hfs-admin space list --project my-project

# 健康检查
hfs-admin health check

场景配置

场景 运行超时 部署间隔 适用场景
dev_test 5分钟 30秒 开发测试
short_task 30分钟 3分钟 短任务
long_task 1小时 5分钟 长任务
production 6-10小时随机 10分钟 生产环境(默认)

详见 场景配置文档

文档

用户文档

设计文档

开发文档

测试文档

架构

┌─────────────┐
│   Redis     │  ← 状态存储
└──────┬──────┘
       │
   ┌───┴────┐
   │        │
┌──▼──┐  ┌─▼───┐
│Worker│  │Worker│  ← 独立进程
└──┬──┘  └─┬───┘
   │       │
┌──▼───────▼──┐
│  Scheduler  │  ← 调度器
└─────────────┘

核心模块

  • state.py - 状态机 + 原子操作(Lua 脚本)
  • health.py - 健康检查(崩溃检测、一致性验证)
  • policy.py - 策略配置(场景、命名)
  • worker.py - Worker 主循环(心跳、进程管理)
  • scheduler.py - 调度器(分配、轮换、创建)
  • account.py - 账号管理(选择、cooldown、评分)
  • hf.py - HuggingFace API 封装

开发

本地开发

git clone <repo>
cd v2
pip install -e .
pytest tests/ -v

测试

# 运行所有测试
pytest tests/ -v

# 运行特定模块
pytest tests/test_state.py -v
pytest tests/test_worker.py -v
pytest tests/test_scheduler.py -v

许可

MIT License

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

mp_hfs-0.1.16.tar.gz (44.9 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

mp_hfs-0.1.16-py3-none-any.whl (53.8 kB view details)

Uploaded Python 3

File details

Details for the file mp_hfs-0.1.16.tar.gz.

File metadata

  • Download URL: mp_hfs-0.1.16.tar.gz
  • Upload date:
  • Size: 44.9 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.12.11

File hashes

Hashes for mp_hfs-0.1.16.tar.gz
Algorithm Hash digest
SHA256 9ebc35d4cf4a1679644c146301a7295830d404c1a4ba13b4cf599817cf5abd50
MD5 764e30c6787ab46f019595dd63259bab
BLAKE2b-256 03f3cc48b3a4e309524731bb353f4f86881b0860a6ef9d0b021ec1c564f29041

See more details on using hashes here.

File details

Details for the file mp_hfs-0.1.16-py3-none-any.whl.

File metadata

  • Download URL: mp_hfs-0.1.16-py3-none-any.whl
  • Upload date:
  • Size: 53.8 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.12.11

File hashes

Hashes for mp_hfs-0.1.16-py3-none-any.whl
Algorithm Hash digest
SHA256 2275375b8f5f35a8ba02036317b29a6f2eadc4aef235addce74d81cc9b622b21
MD5 00efaf48ee31ac714e1a64a75ed95080
BLAKE2b-256 e9d0a9da403b826ede8ebe57e96f85eab51931665e7de76370e70d52acb3f763

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page