cloudoll 4.1.0__tar.gz → 4.2.1__tar.gz

cloudoll-4.2.1/AGENTS.md

@@ -0,0 +1,5 @@
+# Cloudoll library maintenance
+
+This repository implements Cloudoll; it is not a generated application. Read `AI.md` before changing AI assets. Public API behavior must be verified against source and tests, not inferred from other frameworks. The application-development skill is `cloudoll/ai/skills/cloudoll/SKILL.md`; read it when creating application examples.
+
+Official documentation is maintained in the sibling `cloudoll-py-docs` repository, under `src/views`. Bundled copies in `cloudoll/ai/docs` are generated: edit the original docs, then run the sync command in `AI.md`. Update docs and tests together when changing public behavior. Preserve Python 3.9 compatibility.

cloudoll-4.2.1/AI.md

@@ -0,0 +1,32 @@
+# AI-assisted Cloudoll development
+
+For application developers, `cloudoll create PROJECT` installs an `AGENTS.md` entry and `.agents/skills/cloudoll/SKILL.md`. Existing projects can run `python -m cloudoll.cli ai init .`. Existing user instructions are preserved; a different/customized skill is never silently replaced.
+
+No MCP server is required:
+
+```sh
+python -m cloudoll.cli ai list
+python -m cloudoll.cli ai read structure
+python -m cloudoll.cli ai read database
+python -m cloudoll.cli ai search Body
+```
+
+Use the project's Python environment. Ask your coding agent to read the project `AGENTS.md` and Cloudoll skill before coding. Tools that do not discover these files automatically can be given the paths explicitly. These instructions improve discoverability; they cannot guarantee that every model obeys them.
+
+## Maintaining the assets
+
+`cloudoll/ai/docs` contains verbatim snapshots of the official site's Markdown, not hand-maintained API summaries. `manifest.json` records the library version, source path, public URL, and SHA-256 for every topic. Relative site links in these snapshots refer to https://cloudoll.chuchur.com.
+
+From the library root, with the documentation repository available:
+
+```sh
+python scripts/sync_ai_docs.py --docs-root ../cloudoll-py-docs
+python scripts/sync_ai_docs.py --docs-root ../cloudoll-py-docs --check
+python -m pytest tests/test_ai.py
+python -m build
+python tests/check_wheel.py dist/cloudoll-VERSION-py3-none-any.whl
+```
+
+The sync also generates the site's `public/llms.txt` and `public/llms-full.txt`. Commit both repositories' changes. These files are discovery aids, not a replacement for version-matched local docs. When upgrading Cloudoll in an existing application, review its installed skill against the packaged skill and merge intentional changes; initialization refuses a differing skill to avoid overwriting custom instructions.
+
+Before release, run the cross-repository `--check` command. Tests validate the packaged snapshot integrity and version; they cannot detect uncommitted changes in an unavailable documentation repository.

{cloudoll-4.1.0 → cloudoll-4.2.1}/CHANGELOG.md

@@ -1,5 +1,28 @@
 # Changelog
 
+## 4.2.1
+
+- Correct the AI initialization regression test to compare the original bytes
+  rather than assume LF newlines on Windows. Keep preservation and idempotence
+  assertions enabled on every platform.
+- Normalize the generated full documentation's trailing newline.
+- Include all 4.2.0 features. The 4.2.0 tag was blocked by Windows CI and was
+  not published to PyPI; 4.2.1 is the replacement release.
+
+## 4.2.0
+
+Backward-compatible AI-assisted development release. Python 3.9+ remains supported.
+
+- Add offline `cloudoll ai list`, `read`, and `search` commands for bundled official
+  documentation, with source URLs, version metadata and content integrity checks.
+- Add `cloudoll ai init` for existing projects. New scaffolds automatically include
+  an AGENTS.md entry and a project-local Cloudoll skill covering documented project
+  structure, validation, ORM, resource lifecycle, security and verification.
+- Preserve existing project instructions and refuse to overwrite customized skills
+  or write through symbolic links. Initialization is repeatable.
+- Add documentation synchronization tooling and website LLM discovery assets.
+  Test both instruction initialization and AI resources in the built wheel.
+
 ## 4.1.0
 
 Backward-compatible feature release. Pydantic 2 (`>=2.0,<3`) is now a core

{cloudoll-4.1.0 → cloudoll-4.2.1}/MANIFEST.in

@@ -1,4 +1,6 @@
 recursive-include tests *.py
 include README.zh-CN.md
+include AI.md AGENTS.md
+include scripts/sync_ai_docs.py
 include CHANGELOG.md SECURITY.md
 recursive-include docs *.md

{cloudoll-4.1.0/cloudoll.egg-info → cloudoll-4.2.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: cloudoll
-Version: 4.1.0
+Version: 4.2.1
 Summary: Assist in quickly creating distributable microservices based on aiohttp, with a focus on simplicity and ease of use.
 Author-email: Qiu <smallerqiu@gmail.com>
 License-Expression: MIT
@@ -54,16 +54,25 @@ Dynamic: license-file
 
 English | [简体中文](README.zh-CN.md)
 
-Cloudoll 4.1.0 is an aiohttp-based Python library for routing, middleware,
+Cloudoll 4.2.1 is an aiohttp-based Python library for routing, middleware,
 templates, sessions, JWT, database access and application scaffolding.
 
+## AI-assisted development
+
+New scaffolds include `AGENTS.md` and a project-local Cloudoll skill. For existing
+projects, run `cloudoll ai init .` in the project's Python environment. Query the
+bundled official documentation offline with `cloudoll ai list`,
+`cloudoll ai read database`, or `cloudoll ai search Body`. No MCP is required.
+See [AI.md](AI.md) for version matching and maintenance. If your installed release
+does not have `cloudoll ai --help`, consult its matching README and source first.
+
 ## Install
 
 Python 3.9+ is required. Install only the optional drivers you use:
 
 ```sh
-python -m pip install 'cloudoll==4.1.0'
-python -m pip install 'cloudoll[mysql,postgres,cache]==4.1.0'
+python -m pip install 'cloudoll==4.2.1'
+python -m pip install 'cloudoll[mysql,postgres,cache]==4.2.1'
 ```
 
 The AWS extra requires Python 3.10+. Aurora remains untested; see
@@ -78,7 +87,7 @@ python -m pip install -r requirements.txt
 cloudoll start -n myapp
 ```
 
-The generated project targets 4.1.0 and receives a random JWT secret. Controllers
+The generated project targets 4.2.1 and receives a random JWT secret. Controllers
 and authentication middleware are examples: replace demo credentials and supply
 your own authorization and upload policies before deployment.
 
@@ -163,8 +172,8 @@ python -m ruff check cloudoll tests
 python -m mypy
 python -m pytest -q -m 'not integration' -k 'not aws'
 python -m build
-python tests/check_wheel.py dist/cloudoll-4.1.0-py3-none-any.whl
-python tests/check_release.py dist/cloudoll-4.1.0-py3-none-any.whl
+python tests/check_wheel.py dist/cloudoll-4.2.1-py3-none-any.whl
+python tests/check_release.py dist/cloudoll-4.2.1-py3-none-any.whl
 ```
 
 Native integration tests require explicitly configured disposable databases.

{cloudoll-4.1.0 → cloudoll-4.2.1}/README.md

@@ -2,16 +2,25 @@
 
 English | [简体中文](README.zh-CN.md)
 
-Cloudoll 4.1.0 is an aiohttp-based Python library for routing, middleware,
+Cloudoll 4.2.1 is an aiohttp-based Python library for routing, middleware,
 templates, sessions, JWT, database access and application scaffolding.
 
+## AI-assisted development
+
+New scaffolds include `AGENTS.md` and a project-local Cloudoll skill. For existing
+projects, run `cloudoll ai init .` in the project's Python environment. Query the
+bundled official documentation offline with `cloudoll ai list`,
+`cloudoll ai read database`, or `cloudoll ai search Body`. No MCP is required.
+See [AI.md](AI.md) for version matching and maintenance. If your installed release
+does not have `cloudoll ai --help`, consult its matching README and source first.
+
 ## Install
 
 Python 3.9+ is required. Install only the optional drivers you use:
 
 ```sh
-python -m pip install 'cloudoll==4.1.0'
-python -m pip install 'cloudoll[mysql,postgres,cache]==4.1.0'
+python -m pip install 'cloudoll==4.2.1'
+python -m pip install 'cloudoll[mysql,postgres,cache]==4.2.1'
 ```
 
 The AWS extra requires Python 3.10+. Aurora remains untested; see
@@ -26,7 +35,7 @@ python -m pip install -r requirements.txt
 cloudoll start -n myapp
 ```
 
-The generated project targets 4.1.0 and receives a random JWT secret. Controllers
+The generated project targets 4.2.1 and receives a random JWT secret. Controllers
 and authentication middleware are examples: replace demo credentials and supply
 your own authorization and upload policies before deployment.
 
@@ -111,8 +120,8 @@ python -m ruff check cloudoll tests
 python -m mypy
 python -m pytest -q -m 'not integration' -k 'not aws'
 python -m build
-python tests/check_wheel.py dist/cloudoll-4.1.0-py3-none-any.whl
-python tests/check_release.py dist/cloudoll-4.1.0-py3-none-any.whl
+python tests/check_wheel.py dist/cloudoll-4.2.1-py3-none-any.whl
+python tests/check_release.py dist/cloudoll-4.2.1-py3-none-any.whl
 ```
 
 Native integration tests require explicitly configured disposable databases.

{cloudoll-4.1.0 → cloudoll-4.2.1}/README.zh-CN.md

@@ -6,7 +6,16 @@
 
 Cloudoll 是一个基于 aiohttp 的 Python Web 开发库，提供路由自动注册、中间件、模板渲染、会话管理、JWT、数据库访问及命令行脚手架，帮助快速构建 Web 应用和微服务。
 
-本文适用于 Cloudoll 4.1.0。从 3.x 升级前请阅读下方迁移说明。
+本文适用于 Cloudoll 4.2.1。从 3.x 升级前请阅读下方迁移说明。
+
+## AI 辅助开发
+
+新建项目会自带 `AGENTS.md` 和项目级 Cloudoll Skill；已有项目可以在其 Python
+环境中运行 `cloudoll ai init .` 补装。通过 `cloudoll ai list`、
+`cloudoll ai read database` 和 `cloudoll ai search Body` 离线查询随包分发的官方文档，
+不需要 MCP。请让 AI 先读项目指令，再按文档设计实现。
+版本匹配及维护方式见 [AI.md](AI.md)。如果安装版本还不支持 `cloudoll ai --help`，
+请先查阅该版本的 README 和源码。
 
 ## 环境要求
 
@@ -16,10 +25,10 @@ Cloudoll 是一个基于 aiohttp 的 Python Web 开发库，提供路由自动
 
 ## 安装
 
-安装 4.1.0：
+安装 4.2.1：
 
 ```sh
-python -m pip install 'cloudoll==4.1.0'
+python -m pip install 'cloudoll==4.2.1'
 ```
 
 在当前仓库目录安装源码：
@@ -32,16 +41,16 @@ python -m pip install -e .
 
 ```sh
 # MySQL
-python -m pip install 'cloudoll[mysql]==4.1.0'
+python -m pip install 'cloudoll[mysql]==4.2.1'
 
 # PostgreSQL
-python -m pip install 'cloudoll[postgres]==4.1.0'
+python -m pip install 'cloudoll[postgres]==4.2.1'
 
 # AWS 数据库驱动
-python -m pip install 'cloudoll[aws]==4.1.0'
+python -m pip install 'cloudoll[aws]==4.2.1'
 
 # Redis / Memcached 会话或 Redis 数据库
-python -m pip install 'cloudoll[cache]==4.1.0'
+python -m pip install 'cloudoll[cache]==4.2.1'
 ```
 
 源码开发时可用 `python -m pip install -e '.[mysql,postgres,cache,dev]'` 安装。
@@ -537,8 +546,8 @@ cloudoll restart -n myapp
 python -m pip install -e '.[mysql,postgres,cache,dev]'
 python -m pytest -q -m 'not integration' -k 'not aws'
 python -m build
-python tests/check_wheel.py dist/cloudoll-4.1.0-py3-none-any.whl
-python tests/check_release.py dist/cloudoll-4.1.0-py3-none-any.whl
+python tests/check_wheel.py dist/cloudoll-4.2.1-py3-none-any.whl
+python tests/check_release.py dist/cloudoll-4.2.1-py3-none-any.whl
 ```
 
 最后一条命令用于验证构建出的 wheel，版本号变化后请调整文件名。测试使用模拟数据库连接及本机 HTTP 服务；真实数据库连接和 AWS 故障切换需要单独的集成环境。

cloudoll-4.2.1/cloudoll/__init__.py

@@ -0,0 +1 @@
+__version__ = "4.2.1"

cloudoll-4.2.1/cloudoll/ai/AGENTS.md

@@ -0,0 +1,7 @@
+## Cloudoll development
+
+Before building or changing a Cloudoll application, read `.agents/skills/cloudoll/SKILL.md` completely and follow its documentation workflow. Use the project's Python environment, not a globally installed version.
+
+Official documentation is bundled with the installed library: `python -m cloudoll.cli ai list`, `ai read <topic>`, and `ai search <query>`. These commands work offline without MCP. Read `structure`, `config`, `router`, and `validation` before designing an application; read `database` before database work. Consult the installed source when resolving a signature discrepancy; report the discrepancy rather than inventing an API.
+
+Prefer the documented scaffold, auto-discovered controllers/middlewares, YAML configuration, framework-owned database lifecycle, typed request validation, and ORM for ordinary CRUD. Low-level alternatives require a task-specific reason. Never silently replace Cloudoll's architecture with patterns from another framework.

cloudoll-4.2.1/cloudoll/ai/__init__.py

@@ -0,0 +1 @@
+"""Versioned documentation and project instructions for AI-assisted development."""

cloudoll-4.2.1/cloudoll/ai/commands.py

@@ -0,0 +1,106 @@
+"""Offline documentation queries and non-destructive project setup."""
+
+from importlib.resources import files
+from pathlib import Path
+
+import click
+
+BEGIN = "<!-- cloudoll-ai:start -->"
+END = "<!-- cloudoll-ai:end -->"
+
+
+def document_topics() -> list[str]:
+    return sorted(
+        p.name[:-3]
+        for p in files("cloudoll.ai").joinpath("docs").iterdir()
+        if p.name.endswith(".md")
+    )
+
+
+def read_document(topic: str) -> str:
+    if topic not in document_topics():
+        raise click.ClickException("Unknown topic. Run `cloudoll ai list`.")
+    return (
+        files("cloudoll.ai")
+        .joinpath("docs")
+        .joinpath(topic + ".md")
+        .read_text(encoding="utf-8")
+    )
+
+
+def initialize_project(directory: Path) -> None:
+    """Preserve user instructions; refuse to replace a customized skill."""
+    agents = directory / "AGENTS.md"
+    skill = directory / ".agents" / "skills" / "cloudoll" / "SKILL.md"
+    resource = files("cloudoll.ai")
+    skill_text = (
+        resource.joinpath("skills")
+        .joinpath("cloudoll")
+        .joinpath("SKILL.md")
+        .read_text(encoding="utf-8")
+    )
+    instructions = resource.joinpath("AGENTS.md").read_text(encoding="utf-8")
+    original = agents.read_text(encoding="utf-8") if agents.exists() else ""
+    if agents.is_symlink() or any(p.is_symlink() for p in (skill, *skill.parents)):
+        raise click.ClickException(
+            "Refusing to write AI instructions through a symbolic link."
+        )
+    if skill.exists() and skill.read_text(encoding="utf-8") != skill_text:
+        raise click.ClickException(
+            "Existing Cloudoll skill differs; review and merge it manually before initializing."
+        )
+    if original.count(BEGIN) != original.count(END) or original.count(BEGIN) > 1:
+        raise click.ClickException(
+            "Malformed Cloudoll block in AGENTS.md; repair it before initializing."
+        )
+    block = BEGIN + "\n" + instructions.rstrip() + "\n" + END
+    if BEGIN in original:
+        start, end = original.index(BEGIN), original.index(END)
+        if end < start:
+            raise click.ClickException("Malformed Cloudoll block in AGENTS.md.")
+        updated = original[:start] + block + original[end + len(END) :]
+    else:
+        updated = original + ("\n\n" if original else "") + block + "\n"
+    skill.parent.mkdir(parents=True, exist_ok=True)
+    skill.write_text(skill_text, encoding="utf-8")
+    agents.write_text(updated, encoding="utf-8")
+
+
+@click.group()
+def ai() -> None:
+    """Read bundled official docs and initialize project AI instructions (offline)."""
+
+
+@ai.command("list")
+def list_topics() -> None:
+    """List available documentation topics."""
+    click.echo("\n".join(document_topics()))
+
+
+@ai.command("read")
+@click.argument("topic")
+def read_topic(topic: str) -> None:
+    """Print an official documentation topic verbatim."""
+    click.echo(read_document(topic))
+
+
+@ai.command("search")
+@click.argument("query")
+def search(query: str) -> None:
+    """Find matching lines, with topic and line number."""
+    if not query.strip():
+        raise click.ClickException("Search query must not be empty.")
+    for topic in document_topics():
+        for number, line in enumerate(read_document(topic).splitlines(), 1):
+            if query.casefold() in line.casefold():
+                click.echo(f"{topic}:{number}: {line}")
+
+
+@ai.command("init")
+@click.argument(
+    "directory", type=click.Path(path_type=Path, file_okay=False), default="."
+)
+def init(directory: Path) -> None:
+    """Install AGENTS.md guidance and a project-local Cloudoll skill."""
+    initialize_project(directory)
+    click.echo(f"Cloudoll AI instructions installed in {directory.absolute()}")

cloudoll-4.2.1/cloudoll/ai/docs/ai.md

@@ -0,0 +1,61 @@
+---
+title: AI 辅助开发
+order: 15
+icon: structure
+---
+
+# AI 辅助开发
+
+使用 AI 编写 Cloudoll 项目时，应先让它阅读当前安装版本的文档和项目约定，再设计实现。不要仅凭其它 Python 框架的使用经验推测 Cloudoll API。
+
+> 以下 AI 命令从 Cloudoll 4.2.1 开始提供，使用前请运行 `cloudoll ai --help` 确认所安装版本是否支持。旧版本可先阅读对应版本的 README、源码和文档，不必为了获取指引直接升级业务依赖。
+
+## 新项目与已有项目
+
+```sh
+cloudoll create myapp
+```
+
+新建项目会包含 `AGENTS.md` 和 `.agents/skills/cloudoll/SKILL.md`。前者指向开发约定，后者说明应在何时查阅哪些文档、推荐架构和验证要求。
+
+已有项目可在项目的 Python 环境中运行：
+
+```sh
+python -m cloudoll.cli ai init .
+```
+
+初始化会保留 `AGENTS.md` 中其它内容，只维护标记内的 Cloudoll 说明。如果已有 Cloudoll Skill 与当前包不同，命令会停止，要求先人工审查和合并，不会覆盖自定义约定。
+
+## 离线查询文档
+
+不需要配置 MCP 服务，也不需要联网：
+
+```sh
+python -m cloudoll.cli ai list
+python -m cloudoll.cli ai read structure
+python -m cloudoll.cli ai read validation
+python -m cloudoll.cli ai read database
+python -m cloudoll.cli ai search transaction
+```
+
+查询返回随包分发的官方 Markdown 原文。每个主题记录来源地址和内容校验值，避免为 AI 单独手写一套与真实文档不同的 API。`read` 输出完整主题，`search` 输出匹配行及主题、行号；理解上下文时仍应阅读完整主题。
+
+## 如何给 AI 下达任务
+
+可以在任务前加入：
+
+```text
+使用当前项目安装的 Cloudoll。先阅读 AGENTS.md 和
+.agents/skills/cloudoll/SKILL.md，再通过 cloudoll ai read 查阅相关文档。
+新项目遵循自动发现的 controllers/middlewares、YAML 配置、框架管理的
+数据库生命周期、类型化参数校验和 ORM。若需要偏离默认方案，先说明理由。
+完成后说明查阅的主题、测试结果和未验证的部分。
+```
+
+不同 AI 工具发现项目指令的方式不同；如果它没有自动读取文件，请明确提供路径。Skill 和指令不是强制执行机制，仍需检查代码与测试结果。
+
+推荐先阅读[项目结构](/structure)、[配置](/config)、[路由](/router)、[参数校验](/validation)；涉及持久化时阅读[数据库](/database)。集中注册路由、显式创建连接和原始 SQL 并非无效 API，但不应无理由地取代普通应用的推荐结构。
+
+## 在线发现入口
+
+网站提供 [llms.txt](https://cloudoll.chuchur.com/llms.txt) 和 [llms-full.txt](https://cloudoll.chuchur.com/llms-full.txt)，分别用于查找文档入口和读取聚合内容。网站反映当前文档，旧项目应优先使用已安装版本随包提供的文档。最终 API 兼容性以安装版本的源码与测试为依据。

cloudoll-4.2.1/cloudoll/ai/docs/app-start.md

@@ -0,0 +1,81 @@
+---
+title: 生命周期
+order: 150
+icon: lifecycle
+---
+
+# 应用生命周期
+
+默认入口模块为项目根目录的 `app.py`。可用 `cloudoll start -n myapp -e hooks` 指定 `hooks.py`；参数是模块名，不带 .py。
+
+## 钩子
+
+```python
+# app.py
+from cloudoll import logging
+
+def on_create(application):
+    # 同步钩子，参数为 Cloudoll Application。
+    application.config.setdefault("business", {})
+
+async def on_startup(app):
+    # 参数为底层 aiohttp Application；配置的数据库已初始化。
+    logging.info("应用启动完成")
+
+async def on_shutdown(app):
+    logging.info("应用正在停止")
+
+async def on_cleanup(app):
+    logging.info("应用资源清理阶段")
+```
+
+on_startup 被 await，耗时初始化会延迟服务就绪；不要在其中直接运行永不结束的循环。不要在模块导入阶段创建连接或事件循环任务。
+
+配置声明的数据库和 Session 存储由框架启动、关闭。自己创建的 HTTP 客户端、后台任务等仍由业务负责生命周期管理。不要依赖普通 on_cleanup 钩子还能使用已由资源上下文关闭的数据库。
+
+## 后台任务：on_task
+
+on_task 必须是**只 yield 一次的异步生成器**。yield 前初始化，yield 后清理：
+
+```python
+# app.py
+import asyncio
+from contextlib import suppress
+from cloudoll import logging
+
+async def worker():
+    while True:
+        await asyncio.sleep(30)
+        logging.info("后台任务心跳")
+
+async def on_task(app):
+    task = asyncio.create_task(worker())
+    try:
+        yield
+    finally:
+        task.cancel()
+        with suppress(asyncio.CancelledError):
+            await task
+```
+
+不要把上例写成返回 cleanup 函数的普通协程，也不要额外给 on_task 加 `@asynccontextmanager`：这个钩子加载器期待的是异步生成器，不是上下文管理器对象。`contextlib.asynccontextmanager` 本身并没有被弃用。
+
+业务应另行设计后台任务失败告警、重试和停止策略，Cloudoll 不会自动监督所有 create_task。长时间阻塞 I/O 应使用异步库或合适的线程隔离，不能用无 await 的 while 循环阻塞事件循环。
+
+## HTTP 客户端与清理
+
+可在 on_task 中创建并用 async with 管理长期复用的 HTTP Session，见 [HTTP 客户端](/http-client)。
+
+生产使用 SIGTERM 等正常退出流程，给任务和连接池留出足够清理时间；强制杀进程不保证 finally 得以执行。框架会尝试关闭配置创建的资源，但业务自身的第三方关闭函数若一直挂起，仍需要进程管理器的退出超时兜底。
+
+## 资源清理预算
+
+server.resource_close_timeout 默认 10 秒，限制单个配置资源关闭；server.resource_shutdown_timeout 默认 30 秒，限制一轮资源清理。两项都必须是有限正数，不接受 None。原生数据库池另有 close_timeout（默认 10 秒）。配置示例见[配置文件](/config)。
+
+资源按逆序串行清理。单项失败后，在剩余预算内继续清理其他资源；预算耗尽会明确报错，未处理资源可在之后 release() 重试。成功关闭的资源不会重复关闭，仍在执行的旧关闭任务不会重复启动。调用方取消等待时，已开始的清理在预算内继续，然后传播取消。
+
+不配合取消的第三方协程会被停止等待，但不能被 Python 强制终止；同步阻塞 close()、用户钩子及后台任务不在这项预算的保证范围内。资源清理预算不等于 HTTP 请求排空或整个进程退出期限，需要进程管理器最终兜底。
+
+## 开发子进程与热重载
+
+每个开发子进程按显式项目根目录新建 Application，模块加载和工作目录保持一致；不使用 fork 继承的父进程应用缓存或上下文。on_startup 完成且监听端口成功后才发送就绪信号；启动错误会返回父进程，热重载先停止旧实例再启动新实例。