Skip to main content

A powerful Python framework for building declarative, concurrent data processing workflows

Project description

Streamlet - 智能流式数据处理框架

Python License

声明式数据流处理框架:用方法链表达业务逻辑,框架自动处理异步/同步混合执行、并行调度和重试。

  • 🎯 声明式工作流.then() .fan_out_to() .fan_in() .branch_on() .repeat() 方法链构建数据流
  • 🤖 智能异步执行:自动检测 async/sync 函数并选择正确的执行策略,无需手动协调
  • 🔗 @node 装饰器:任意函数变为可组合节点,内置 pydantic 类型校验和依赖注入
  • 🛡️ 重试机制:基于异常分类的可配置指数退避重试

快速开始

pip install streamlet-py
from streamlet import node

@node
def double(x: int) -> int:
    return x * 2

@node
def add_ten(x: int) -> int:
    return x + 10

result = double.then(add_ten)(5)  # 20

核心 API

方法 功能 示例
.then(node) 顺序连接 a.then(b)(data)
.fan_out_to([nodes], executor="thread") 并行分发 a.fan_out_to([b, c])()
.fan_in(aggregator) 聚合并行结果 flow.fan_in(merge)()
.fan_out_in([nodes], agg) 扇出 + 聚合 a.fan_out_in([b, c], merge)()
.branch_on({key: node}) 条件分支 a.branch_on({True: b, False: c})()
.repeat(times, input_mode=RepeatInputMode.PREVIOUS_RESULT) 重复执行 a.repeat(3)(data)

fan_out_to / fan_out_inexecutor 用于选择并行调度策略: "thread" 使用线程池,适合同步或阻塞型 target;"async" 使用 asyncio.gather,只适合真实 async 且会主动 await 的 target。同步 target 在 "async" executor 下会直接运行在当前 event loop 中,不会自动放入 线程池,CPU 或阻塞 I/O 会阻塞 event loop。"auto" 会在全同步节点时选择 "thread";在包含异步节点时使用 hybrid 调度:async target 留在 event loop, 同步 target 放入线程池执行。

fan_out_to(...).then(...) 不会把成功结果列表传给下游;.then() 接收到的是 原始的 dict[str, ParallelResult]。fan-out 后如果要继续处理业务结果,应先用 .fan_in(aggregator) 显式聚合,或直接使用 .fan_out_in(...)

repeat 默认使用 RepeatInputMode.PREVIOUS_RESULT:第 1 轮执行 node(*args, **kwargs),之后把每轮成功返回值作为下一轮的单个位置参数。 如果需要为下一轮指定多个位置参数或关键字参数,返回 call_args(...)。 如果每轮都要使用最初传入的参数,使用 RepeatInputMode.SAME_INPUT

from streamlet import CallArgs, RepeatInputMode, call_args, node

@node
def inc(value: int) -> int:
    return value + 1

assert inc.repeat(3)(0) == 3

@node
def step(value: int, factor: int = 1) -> CallArgs:
    return call_args(value * factor, factor=factor)

assert step.repeat(3)(2, factor=10) == call_args(2000, factor=10)

@node
def collect(source: str, limit: int = 10) -> list[str]:
    return fetch_batch(source, limit=limit)

flow = collect.repeat(3, input_mode=RepeatInputMode.SAME_INPUT)
results = flow("orders", limit=50)

@node 内置 Pydantic 输入和返回值校验。返回值校验支持 from __future__ import annotations 下的延迟注解,并保留 Annotated[...] 元数据,例如 Field(gt=0);校验失败分别抛出 ValidationInputExceptionValidationOutputException。更多细节见 API 参考Pydantic TypeAdapter

示例

顺序流:ETL 管道

from streamlet import node
import asyncio

@node
async def fetch_data(source: str) -> dict:
    await asyncio.sleep(0.1)
    return {"value": 100, "source": source}

@node
def validate(data: dict) -> dict:
    if data["value"] <= 0:
        raise ValueError("invalid value")
    return data

@node
def enrich(data: dict) -> dict:
    return {**data, "doubled": data["value"] * 2}

pipeline = fetch_data.then(validate).then(enrich)

async def main():
    result = await pipeline("db")
    print(result)  # {"value": 100, "source": "db", "doubled": 200}

asyncio.run(main())

并行流:扇出 + 聚合

from streamlet import fan_out_args, node

@node
def source(x: int) -> dict:
    return {"value": x}

@node
def multiply(data: dict) -> int:
    return data["value"] * 2

@node
def add_ten(data: dict) -> int:
    return data["value"] + 10

@node
def aggregate(results: dict) -> dict:
    values = [r.result for r in results.values() if r.success]
    return {"total": sum(values), "results": values}

workflow = source.fan_out_to([multiply, add_ten], executor="thread").fan_in(aggregate)
result = workflow(5)
print(result)  # {"total": 25, "results": [10, 15]}

注意:fan_out_to([multiply, add_ten]).then(next_node) 会把 dict[str, ParallelResult] 原样传给 next_node,不会自动提取成功结果。 继续业务链时通常应写成 fan_out_to(...).fan_in(aggregate).then(next_node)

source 返回普通值时,所有 target 接收同一个单参数;需要为不同 target 传递不同 kwargs 时,返回显式的 fan_out_args

@node
def route(user_id: str):
    return fan_out_args(
        {"user_id": user_id, "limit": 10},
        {"user_id": user_id, "include_archived": False},
    )

flow = route.fan_out_to([fetch_orders, fetch_profile])

条件流:分支路由 + 依赖注入

from streamlet import BaseFlowContext, node
from dependency_injector.wiring import Provide

container = BaseFlowContext()

@node
def evaluate(data: dict) -> str:
    return "pass" if data["score"] >= 60 else "fail"

@node
def handle_pass(state: dict = Provide[BaseFlowContext.context]) -> dict:
    return {"result": "pass", "score": state["score"]}

@node
def handle_fail(state: dict = Provide[BaseFlowContext.context]) -> dict:
    return {"result": "fail", "score": state["score"]}

container.wire(modules=[__name__])
container.context()["score"] = 75

flow = evaluate.branch_on({"pass": handle_pass, "fail": handle_fail})
print(flow({"score": 75}))  # {"result": "pass", "score": 75}

branch_on 的语义是:条件节点接收调用输入并返回路由键,选中的分支节点 以零参数执行。框架不会把原始输入或条件返回值传给分支;分支需要业务数据时, 应通过 BaseFlowContext.context 等依赖注入方式读取。

fan-out 分支会复制父级 BaseFlowContext.context 后再执行,默认策略只浅拷贝 顶层 dict:新增或替换顶层 key 时分支互不影响;如果 value 本身是嵌套 list / dict / set 等可变对象,仍会共享同一个对象引用。需要提前拒绝这类 嵌套可变状态时,可以定义显式的 strict context:

from streamlet import BaseFlowContext, ContextVarProvider

class StrictFlowContext(BaseFlowContext):
    context = ContextVarProvider(dict, copy_policy="strict")

copy_policy="strict" 不会递归深拷贝 context;它会在 fan-out 隔离时检测并拒绝 会被浅拷贝共享的可变值,让共享状态风险尽早暴露。直接的 list / dict / set 会被拒绝,藏在 tuple / frozenset 等不可变容器里的可变值也会被拒绝, 例如 {"items": ("header", [])}

重试机制

from streamlet import node

@node(retry_count=3, retry_delay=0.5, backoff_factor=2.0, enable_retry=True)
def external_call(x: int) -> int:
    # 失败时自动重试,延迟按 0.5s → 1.0s → 2.0s 指数增长
    return call_external_api(x)

开发环境

git clone https://github.com/12306hujunjie/Streamlet.git
cd Streamlet

pdm install

pdm run pytest                                   # 运行测试
pdm run pytest --cov=src/streamlet              # 覆盖率
pdm run ruff check src/ tests/                   # 代码检查
pdm run mypy src/streamlet/                     # 类型检查

技术栈

  • Python 3.10+
  • dependency-injector — 依赖注入与 ContextVar 隔离状态管理
  • pydantic v2 — 类型校验

核心模块:asyncio | threading | concurrent.futures

文档

许可证

MIT — 详见 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

streamlet_py-1.0.0b1.tar.gz (39.9 kB view details)

Uploaded Source

Built Distribution

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

streamlet_py-1.0.0b1-py3-none-any.whl (21.6 kB view details)

Uploaded Python 3

File details

Details for the file streamlet_py-1.0.0b1.tar.gz.

File metadata

  • Download URL: streamlet_py-1.0.0b1.tar.gz
  • Upload date:
  • Size: 39.9 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: pdm/2.27.0 CPython/3.11.15 Linux/6.17.0-1018-azure

File hashes

Hashes for streamlet_py-1.0.0b1.tar.gz
Algorithm Hash digest
SHA256 ccdefde45a89e5e3080dfd9e885430f37927f99491c3ef01fedda93c96a456db
MD5 4004ed6a29e8d9ea4b2e069652c25dfc
BLAKE2b-256 e98f872438507d50c007e5ed10a92f716e5631c6dd21b042d28c80acc065d2e5

See more details on using hashes here.

File details

Details for the file streamlet_py-1.0.0b1-py3-none-any.whl.

File metadata

  • Download URL: streamlet_py-1.0.0b1-py3-none-any.whl
  • Upload date:
  • Size: 21.6 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: pdm/2.27.0 CPython/3.11.15 Linux/6.17.0-1018-azure

File hashes

Hashes for streamlet_py-1.0.0b1-py3-none-any.whl
Algorithm Hash digest
SHA256 82db7b87238cc37eb56b90c9e8f0d50a40a7ce3f7b2bebe1cbc40f9f6650ea27
MD5 1e19df7f347ade7ee05b38bcb720ebd5
BLAKE2b-256 c167127d5f3a43267a356b58f3e85f678d26c9d5514a83879fc28c508cbcd9bc

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