zoo-framework 0.9.1b0

Python 声明式多任务编排框架，一站式支撑下一代 Agent 与工作流 Python declarative multi-task orchestration framework powering next-gen Agents and workflows

English | 中文

🇬🇧 English

What it is

Zoo Framework runs long-lived background tasks in Python. You declare task units ("Workers"), register them, and the framework keeps them running: it decides when each one is dispatched, keeps concurrent instances apart, observes how long they take, and shuts down cleanly with state on disk.

It is not a web framework and not a task queue — there is no HTTP layer, no broker, and no distributed scheduling. It is the in-process equivalent: a scheduler, an event pipeline, and a state store that you embed in a service.

The problem it solves

Wiring background work out of raw threading looks like twenty lines and turns into a few hundred: who owns the thread, what stops the same job from overlapping itself, how does a stuck job get noticed, what happens to in-flight work on shutdown, where does the state go. Each of those is a place to be subtly wrong, and the failure mode is usually silence — a job that stopped running, or two that ran at once.

Zoo Framework takes those decisions out of your code:

• Overlap — a Worker that is still in flight is skipped for the round, never dispatched twice concurrently.
• Stuck work — run_timeout observes and circuit-breaks. Note the framework does not force-kill a running Worker; it stops dispatching it.
• Shutdown — Master.shutdown() stops dispatch, cancels the schedule, stops the event pipeline and monitoring, then unregisters every Worker — which is where the state machine's final save happens.
• State — periodic and on-shutdown persistence, atomic replace, rolling backups.
• Errors — unsupported inputs raise rather than silently degrade.

How it compares

Zoo Framework is one point in a space that already has good tools. It is the right choice when the work is long-lived, in-process, and not worth a broker:

Read the table as a scope statement, not a scoreboard: Celery is the right answer the moment work must cross machines, and APScheduler is the right answer for cron. Zoo Framework deliberately declines both, and the row it wins on is "no external dependency, no broker, no cron daemon — but still scheduled, still observable, still persistent." The comparison reflects each project's general positioning; no cross-project benchmark was run.

Installation

pip install zoo-framework

Requires Python 3.13+.

Quick Start

A minimal runnable Worker. Save as main.py and run it:

from zoo_framework.core import Master
from zoo_framework.workers import BaseWorker


class MyWorker(BaseWorker):
    """A task unit that runs in a loop."""

    def __init__(self):
        super().__init__(
            {
                "is_loop": True,  # keep running across scheduling rounds
                "delay_time": 1.0,  # seconds to wait after each execution
                "name": "MyWorker",
                # "run_timeout": 30,  # optional: observe + circuit-break after 30s
            }
        )
        self.counter = 0

    def _execute(self):
        self.counter += 1
        print(f"Hello from MyWorker! Count: {self.counter}")


if __name__ == "__main__":
    master = Master()
    # Registered workers join the schedule. Master() alone only runs the two
    # built-in system workers (EventWorker / StateMachineWorker).
    master.register_worker("MyWorker", MyWorker)
    master.run()

There is no separate config file to write — Master() defaults to ./config.json relative to the working directory, and the example above runs without one. To scaffold a project with a config and directory layout, use the CLI instead:

zfc --create myapp      # -> myapp/src/{main.py,workers,conf,params,events}
cd myapp
zfc --worker my_task    # adds src/workers/my_task_worker.py and registers it
python src/main.py

A Worker must be registered as a class. WorkerRegistry validates with issubclass, so a function or an instance is rejected with TypeError: issubclass() arg 1 must be a class. This holds independently of @cage: any wrapper that replaces the class with a factory function fails the same way, and the rule still stands now that @cage is removed. Process-level sharing is declared through the container instead.

How it runs

Master.run()
   └─ scheduling round (every 1s)
        ├─ for each registered Worker:
        │    ├─ timed out?        → circuit-break, stop dispatching it
        │    ├─ still in flight?  → skip this round
        │    └─ otherwise         → dispatch to a thread / the pool
        └─ on completion: unregister in-flight, report result to the event pipeline

Workers execute concurrently within a round and are never dispatched twice at the same time. Master.shutdown() stops dispatch first, then cancels the scheduling task, stops the event loop, stops monitoring, and finally unregisters every Worker — which is where the state machine's last save happens.

How the pieces connect:

flowchart TB
    cfg["config.json"] --> pf["ParamsFactory<br/>resolved once, at first import"]
    m["Master"] --> pf
    m --> wr["WorkerRegistry<br/>module-level singleton"]
    m -->|"worker:mode"| w["Waiter<br/>ThreadPerTaskModel / ThreadPoolModel"]
    w -->|"dispatch, once per round"| wk["Worker._execute()"]
    wk --> set["WorkerDispatchCore.settle()<br/>single settlement point:<br/>unregister in-flight, report result"]
    set --> pipe["Event pipeline<br/>EventChannel → EventFIFO → Reactor"]
    m --> sm["StateMachineWorker<br/>periodic save, atomic replace + rolling backup"]

Features

Core concepts

The framework is named after a zoo, but the metaphor only affects naming, not semantics — if a name is unclear, read the right-hand column:

Event pipeline

from zoo_framework.core.aop import event


@event(topic="order.created", channel="business")
def on_order_created(req):
    print(req.topic, req.content)

Events are isolated per channel; an event can select its response mechanism (first responder only / by priority / all / a named responder). Reactor registration is idempotent — registering the same object twice does not append it twice.

State machine

Read and write by "scope + key path"; key paths support dotted nesting:

from zoo_framework.statemachine import StateMachineManager

sm = StateMachineManager()

sm.set_state("order", "status", "pending")  # first write creates the scope
sm.set_state("order", "status", "paid")  # overwrite
sm.get_state("order", "status")  # -> 'paid'

sm.set_state("order", "item.count", 2)  # nested key
sm.get_state("order", "item.count")  # -> 2

# Observe changes
sm.observe_state("order", "status", lambda payload: print(payload["value"]))
sm.unobserve_state("order", "status", observer)

State is persisted periodically by StateMachineWorker and once more on shutdown: written to a .tmp file then atomically replaced, keeping the last 5 backups in a sibling backups/ directory.

CLI

# Scaffold a project (produces <name>/src/{main.py,workers,conf,params,events})
zfc --create myapp

# Add a Worker to a project (writes src/workers/, and wires it into src/main.py)
cd myapp
zfc --worker my_task

# Run
python src/main.py

There are exactly two options, --create and --worker, and they may be combined in one call (zfc --create myapp --worker my_task puts the Worker into the project being created). Both validate their input before producing anything, and neither accepts a silently ineffective call:

• --create fails and exits if the target directory already exists — it does not merge and does not overwrite. Check for the directory yourself if you need idempotency in a script.
• --worker names must be valid Python identifiers: no leading digit, no hyphens, dots or spaces, no keywords. my-task and 123task are rejected; use underscores (my_task).

On failure the command exits non-zero with a reason, leaving no half-written output.

Performance

Measured on Windows, Python 3.13, with a representative workload (JSON encode/decode + string processing). bench/ has the scripts and raw data.

Framework overhead is roughly 100–220 µs per task and is dominated by thread dispatch and the wake-up back to the scheduler. It is therefore negligible for millisecond-scale work and dominant for sub-100 µs work — pick your task granularity accordingly. bench/DECISION.md has the breakdown and the cross-platform caveats.

On the numbers above. They are one-off evidence from the adopt-rust-core study, measured on Windows only — do not compare them against Linux runs. The maintained, version-by-version benchmark lives in the separate zoo-bench repo: live report. It runs on native Linux CI, compares against hand-written baselines and the standard-library concurrency models, publishes its raw data, and reports the tiers where the framework loses.

Built for AI-agent-generated code

The extension surface is deliberately narrow, so generated code is short, verifiable, and — when it is wrong — wrong loudly.

Three steps, no glue code

class OrderSyncWorker(BaseWorker):  # 1. subclass
    def __init__(self):
        super().__init__({"is_loop": True, "delay_time": 5, "name": "OrderSync"})

    def _execute(self):  # 2. write only the business logic
        sync_orders()


master.register_worker("OrderSync", OrderSyncWorker)  # 3. register

Thread management, concurrency limits, in-flight de-duplication, timeout circuit-breaking, graceful shutdown and state persistence are the framework's job. An agent does not have to generate that code — and therefore cannot get it wrong.

Configuration is separate from implementation

A Worker only depends on the props dict handed to __init__. It has no visibility into framework internals, so generating one does not require reading the source or understanding how Waiter / WorkerRegistry / EventReactor relate.

Failures are explicit, never silently swallowed

This matters more for generated code than for hand-written code: an agent cannot detect a silent downgrade, and a loud error is the signal it needs to self-correct.

Changes can be checked automatically

The framework carries a spec baseline and a 367-case regression suite covering the core contracts, so an agent's change can be judged by running pytest rather than by asking a human to read it.

Documentation

• Architecture — module layout and data flow
• API Reference
• Development Guide
• Debugging Guide
• Roadmap

Community & Feedback

• Questions, ideas, bug reports — GitHub Issues is the single channel; there is no chat server or mailing list.
• Security vulnerabilities — do not open a public issue. See SECURITY.md for the private reporting route.
• Code of Conduct — CODE_OF_CONDUCT.md.

The project is maintained by one person, so issues are answered on a best-effort basis. A report with a reproduction, your Python version and your OS gets a response far faster than one without.

Contributing

Contributions are welcome — see CONTRIBUTING.md for the full guide.

git clone https://github.com/YearsAlso/zoo-framework.git
cd zoo-framework
pip install -e ".[dev]"       # NOT `uv sync` — uv.lock is stale, see CONTRIBUTING.md
pre-commit install
pytest                        # 662 cases

Note: use an explicit Python 3.13 interpreter (.venv/Scripts/python.exe on Windows — if you prefer uv, uv run --no-sync; plain uv run re-locks from the stale uv.lock). ruff check, ruff format, pytest, mypy and bandit are all hard CI gates.

License

Apache License 2.0 © XiangMeng

🇨🇳 中文

这是什么

Zoo Framework 用来运行长期存活的后台任务。你声明任务单元（Worker）并注册它们， 框架负责让它们持续跑下去：决定每一个何时被派发、避免并发实例互相干扰、观测它们跑了 多久、并在停机时把状态干净地落盘。

它不是 Web 框架，也不是任务队列 —— 没有 HTTP 层、没有 broker、没有分布式调度。 它是进程内的等价物：一个可以嵌进服务里的调度器 + 事件管道 + 状态存储。

解决什么问题

用裸 threading 搭后台任务，写起来像是二十行的事，最后会长成几百行：线程归谁管、 怎么保证同一个任务不自我重叠、卡住的任务谁来发现、停机时正在跑的任务怎么办、状态放 哪里。每一条都是可能悄悄写错的地方，而且出错方式通常就是静默 —— 某个任务不再跑 了，或者两个实例同时跑了。

Zoo Framework 把这些决定从你的代码里拿走：

• 重叠 —— 仍在执行中的 Worker 本轮跳过，永远不会被并发派发两次。
• 卡住的任务 —— run_timeout 负责观测并熔断。注意框架不会强制终止正在执行的 Worker，它只是停止继续派发。
• 停机 —— Master.shutdown() 先停派发，再取消调度、停事件管道与监控，最后注销 全部 Worker —— 状态机的最后一次落盘就发生在这条链路的末尾。
• 状态 —— 周期落盘 + 停机落盘，原子替换，滚动备份。
• 错误 —— 不受支持的输入抛异常，而不是静默降级。

同类方案对比

Zoo Framework 处在一个已经有优秀工具的空间里。它合适的场景是：任务长期存活、在进程 内、且不值得为它引入 broker。

这张表是范围声明，不是打分表：任务一旦要跨机器，Celery 就是正确答案；需要 cron， APScheduler 就是正确答案。Zoo Framework 主动放弃了这两项，它赢的那一行是「无外部依赖、 无 broker、无 cron 守护进程 —— 但依然有调度、有观测、有持久化」。表中对比基于各项目公开 的通用定位，未做过跨项目压测。

安装

pip install zoo-framework

需要 Python 3.13+。

快速开始

一个最小可运行的 Worker。存成 main.py 直接运行：

from zoo_framework.core import Master
from zoo_framework.workers import BaseWorker


class MyWorker(BaseWorker):
    """一个循环执行的任务单元。"""

    def __init__(self):
        super().__init__(
            {
                "is_loop": True,  # 跨调度轮次持续执行
                "delay_time": 1.0,  # 单次执行结束后的等待秒数
                "name": "MyWorker",
                # "run_timeout": 30,  # 可选：超过 30 秒则熔断
            }
        )
        self.counter = 0

    def _execute(self):
        self.counter += 1
        print(f"Hello from MyWorker! 计数: {self.counter}")


if __name__ == "__main__":
    master = Master()
    # 注册后的 Worker 才会进入调度。只 Master() 的话只会跑内置的两个
    # 系统 Worker（EventWorker / StateMachineWorker）。
    master.register_worker("MyWorker", MyWorker)
    master.run()

不需要单独写配置文件 —— Master() 默认读工作目录下的 ./config.json，上面这个例子没有 配置文件也能跑。想要带配置和目录结构的脚手架，用 CLI：

zfc --create myapp      # -> myapp/src/{main.py,workers,conf,params,events}
cd myapp
zfc --worker my_task    # 写入 src/workers/my_task_worker.py 并自动注册
python src/main.py

Worker 必须以「类」的形式注册。 WorkerRegistry 用 issubclass 校验契约，传函数或 实例会被拒绝并抛 TypeError: issubclass() arg 1 must be a class。这条约束与 @cage 无关：任何把类换成工厂函数的包装都一样失败，@cage 删除后它依然成立。进程级共享 现在改由容器声明。

运行方式

Master.run()
   └─ 调度轮次（每秒一次）
        ├─ 遍历已注册的 Worker：
        │    ├─ 已超时？        → 熔断，不再派发
        │    ├─ 仍在执行？      → 本轮跳过
        │    └─ 否则            → 派发到线程 / 资源池
        └─ 执行结束时：注销在飞状态，把结果投递进事件管道

同一轮里多个 Worker 并发执行；同一个 Worker 永远不会被同时派发两次。 Master.shutdown() 先停止派发，再取消调度任务、停事件循环、停监控，最后注销全部 Worker —— 状态机的最后一次落盘就发生在这条链路的末尾。

各部件的连接关系：

flowchart TB
    cfg["config.json"] --> pf["ParamsFactory<br/>首次导入时解析一次"]
    m["Master"] --> pf
    m --> wr["WorkerRegistry<br/>模块级单例"]
    m -->|"worker:mode"| w["Waiter<br/>ThreadPerTaskModel / ThreadPoolModel"]
    w -->|"每轮派发一次"| wk["Worker._execute()"]
    wk --> set["WorkerDispatchCore.settle()<br/>单一结算收口：<br/>注销在飞 + 上报结果"]
    set --> pipe["事件管道<br/>EventChannel → EventFIFO → Reactor"]
    m --> sm["StateMachineWorker<br/>周期落盘：原子替换 + 滚动备份"]

核心特性

核心概念

框架用动物园隐喻命名，但隐喻只影响命名，不影响语义 —— 看不懂名字时看右列即可：

事件管道

from zoo_framework.core.aop import event


@event(topic="order.created", channel="business")
def on_order_created(req):
    print(req.topic, req.content)

事件按通道隔离；同一事件可选择响应机制（仅首个 / 按优先级 / 全部 / 指定响应器）。 响应器注册是幂等的 —— 同一对象重复注册不会重复追加。

状态机

按「作用域 + 键路径」读写，键路径支持点号嵌套：

from zoo_framework.statemachine import StateMachineManager

sm = StateMachineManager()

sm.set_state("order", "status", "pending")  # 首次写入自动创建作用域
sm.set_state("order", "status", "paid")  # 重复写入覆盖
sm.get_state("order", "status")  # -> 'paid'

sm.set_state("order", "item.count", 2)  # 嵌套键
sm.get_state("order", "item.count")  # -> 2

# 观察状态变化
sm.observe_state("order", "status", lambda payload: print(payload["value"]))
sm.unobserve_state("order", "status", observer)

状态由 StateMachineWorker 周期落盘，停机时再保存一次（写入 .tmp 后原子替换， 并在相邻的 backups/ 目录保留最近 5 份备份）。

CLI 工具

# 生成一个脚手架项目（产出 <name>/src/{main.py,workers,conf,params,events}）
zfc --create myapp

# 在项目里新增一个 Worker（写入 src/workers/，并接入 src/main.py 的注册表）
cd myapp
zfc --worker my_task

# 启动
python src/main.py

选项只有 --create 与 --worker 两个，可以在同一次调用里组合使用 （zfc --create myapp --worker my_task 会把 Worker 直接落进本次创建的项目）。 两者都会在产出之前校验输入，并且都不接受「静默无效」的调用：

• --create 的目标目录已存在时报错退出，不会合并、不会覆盖既有内容； 脚本中需要幂等时请自行先判断目录是否存在。
• --worker 的名称必须是合法 Python 标识符：不能以数字开头，不能含连字符、 点号或空格，不能是 Python 关键字。my-task / 123task 这类写法会被拒绝， 请改用下划线命名（my_task）。

失败时命令以非 0 退出码结束并说明原因，且不会留下半成品产物。

性能

Windows / Python 3.13 实测，负载为代表性任务（JSON 编解码 + 字符串处理）。 脚本与原始数据在 bench/。

框架自身开销约为每任务 100–220 µs，主要来自线程派发与调度线程的唤醒。因此它对 毫秒级任务是可忽略的，对 100 µs 以下的任务则占主导 —— 任务粒度请据此选择。 拆解与跨平台说明见 bench/DECISION.md。

关于上面这张表：它是 adopt-rust-core 那次研究留下的一次性证据，只在 Windows 上测过 —— 不要拿它跟 Linux 的数字对比。 逐版本维护的基准在独立仓库 zoo-bench： 线上报告。它跑在原生 Linux CI 上，与手写基线及 标准库并发模型横向对照，公开原始数据，并且如实列出本框架输掉的档位。

面向 AI Agent 的代码生成

框架的扩展面被刻意收窄，使生成的代码短、可校验，而且 —— 出错时出错得很大声。

三步接入，没有胶水代码

class OrderSyncWorker(BaseWorker):  # 1. 继承
    def __init__(self):
        super().__init__({"is_loop": True, "delay_time": 5, "name": "OrderSync"})

    def _execute(self):  # 2. 只写业务逻辑
        sync_orders()


master.register_worker("OrderSync", OrderSyncWorker)  # 3. 注册

线程管理、并发上限、在飞去重、超时熔断、优雅停机、状态落盘全部由框架承担。 Agent 不需要生成这些代码，也就不会把它们生成错。

配置与实现分离

Worker 只依赖传给 __init__ 的 props 字典，不感知框架内部结构。生成一个 Worker 不需要读框架源码，也不需要理解 Waiter / WorkerRegistry / EventReactor 之间的关系。

失败是显式的，不会被静默吞掉

这一点对生成式代码比对人工代码更重要：Agent 无法从「静默降级」中察觉自己写错了， 而明确的报错正是它自我修正所需的信号。

改动可被自动校验

框架带 spec 基线与 367 条回归用例，核心契约都有对应用例守护 —— Agent 生成的改动 可以靠 pytest 判断对错，而不必靠人逐行读。

文档

• 架构设计 — 模块划分与数据流
• API 参考
• 开发指南
• 调试指南
• 路线图

社区与反馈

• 提问、想法、Bug 上报 —— 统一走 GitHub Issues， 项目没有 IM 群或邮件列表。
• 安全漏洞 —— 不要开公开 issue，请见 SECURITY.md 的私密上报途径。
• 行为准则 —— CODE_OF_CONDUCT.md。

项目由一个人维护，issue 按尽力而为的原则响应。附带最小复现、Python 版本和操作系统的 报告，会比其他报告快得多。

贡献代码

欢迎贡献，完整流程见 CONTRIBUTING.md。

git clone https://github.com/YearsAlso/zoo-framework.git
cd zoo-framework
pip install -e ".[dev]"       # 请勿使用 `uv sync` —— uv.lock 陈旧，见 CONTRIBUTING.md
pre-commit install
pytest                        # 662 条用例

注意使用明确的 Python 3.13 解释器（Windows 上是 .venv/Scripts/python.exe； 偏好 uv 的话用 uv run --no-sync，裸 uv run 会从陈旧的 uv.lock 重新解析）。 CI 中 ruff check、ruff format、pytest、mypy 与 bandit 都是硬性门禁。

许可证

Apache License 2.0 © XiangMeng