实现下单任务网关与本地 worker

按 docs/order-gateway.md 落地:第三个部署单元 app.gateway(:31109)承担任务队列
+ 状态镜像;本地 worker 在 app.trading.worker 内,按 RAKUTEN_ORDER_GATEWAY_URL
决定是否启动。规格 §5 最关键约束已守:租约过期绝不自动重投,恢复只能 reclaim,
worker 收到 lease_count>1 时先核对站点订单。

站点交互(加购/下单/付款/订单列表反查)按规格 §10 留接口缝,site_interact.py
全部 NotImplementedError,verify.py 恒返回 unknown——等真实账号实测后再填,
不写猜测的提交逻辑。

310 个测试全绿,覆盖规格 §9 验收清单 12 条;架构测试守住三方互不 import。

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
2026-07-27 16:25:20 +08:00
co-authored by Claude Opus 4.6
parent 91bfab7196
commit 07107094a7
35 changed files with 4342 additions and 62 deletions
+39 -9
View File
@@ -1,6 +1,7 @@
# 本仓库出个服务,共用这一份配置:
# 本仓库出个服务,共用这一份配置:
# 抓取服务 python -m app.scraping.main —— 匿名、无状态、可多开实例
# 交易服务 python -m app.trading.main —— 带账号登录态、有状态,只能单实例
# 下单任务网关 python -m app.gateway.main —— 任务队列 + 状态镜像,只能单实例
# 各自只读自己那部分,下面按用途分组标注。
# ---- 抓取服务监听地址,通常本地用 127.0.0.1,容器/服务器用 0.0.0.0 ----
@@ -11,6 +12,10 @@ RAKUTEN_APP_PORT=31107
RAKUTEN_TRADING_HOST=0.0.0.0
RAKUTEN_TRADING_PORT=31108
# ---- 下单任务网关监听地址(部署在服务器侧;本地 worker 出站长轮询取任务)----
RAKUTEN_GATEWAY_HOST=0.0.0.0
RAKUTEN_GATEWAY_PORT=31109
# 运行环境:dev / prod / test
RAKUTEN_APP_ENV=dev
@@ -56,14 +61,6 @@ RAKUTEN_PROXY_USERNAME=
# 代理密码(如代理需要认证则填写)
RAKUTEN_PROXY_PASSWORD=
# ---- 以下仅交易服务使用 ----
# 人工登录后落盘的 cookie 目录(相对项目根目录)。
# 里面是可直接冒充账号的凭据,已在 .gitignore 排除,不要提交、不要外传。
RAKUTEN_AUTH_STATE_DIR=.auth
# 下单金额上限(日元):实际应付超过该值直接拒绝提交,防止解析出错或页面改版
# 导致买到远超预期的订单。设为 0 表示不设上限(不建议)。
RAKUTEN_ORDER_MAX_TOTAL_YEN=30000
# ---- OpenTelemetry traces(可选;默认关闭)----
# 启用后把抓取-解析链路以 span 导出到 OTLP/HTTP endpoint,
# 用于排查"抓到的内容为什么解析不出预期字段"。
@@ -76,3 +73,36 @@ RAKUTEN_OTEL_SERVICE_NAME=
RAKUTEN_OTEL_HEADERS=
# 失败 HTML 快照上限字节(默认 2MB,超出截断并标注 truncated=true)
RAKUTEN_OTEL_SNAPSHOT_MAX_BYTES=2000000
# ---- 以下仅交易服务使用 ----
# 人工登录后落盘的 cookie 目录(相对项目根目录)。
# 里面是可直接冒充账号的凭据,已在 .gitignore 排除,不要提交、不要外传。
RAKUTEN_AUTH_STATE_DIR=.auth
# 下单金额上限(日元):实际应付超过该值直接拒绝提交,防止解析出错或页面改版
# 导致买到远超预期的订单。设为 0 表示不设上限(不建议)。
RAKUTEN_ORDER_MAX_TOTAL_YEN=30000
# ---- 以下仅下单任务网关使用 ----
# 任务队列 SQLite 文件路径(相对项目根目录)。务必放在持久化卷上,丢了等于
# 丢了一批下单任务。详见 docs/order-gateway.md。
RAKUTEN_GATEWAY_DB_PATH=data/gateway.db
# 任务租约 TTL(秒)。worker 领取后必须在此时间内首次 report 或 renew,否则
# 任务被置为 stale(**绝不自动重投**,需要人工 reclaim)。
RAKUTEN_LEASE_TTL_SECONDS=300
# 长轮询单次最长挂起秒数。worker 端 wait 参数会被夹到这个上限。
RAKUTEN_LEASE_MAX_WAIT_SECONDS=60
# worker 心跳超时阈值(秒)。超过即视为失联,/health 报 degraded。
RAKUTEN_WORKER_OFFLINE_ALERT_SECONDS=300
# ---- 以下仅交易服务内的下单 worker 使用 ----
# 网关 URL。**留空则不启动 worker**,交易服务只跑登录态接口。
# 部署形态:本地机(NAT 后无公网入口)通过出站长轮询从这里领任务。
RAKUTEN_ORDER_GATEWAY_URL=
# worker 标识。同一时间只能有一个 worker 持有 lease,留空时取主机名。
RAKUTEN_WORKER_ID=
# 本地订单 SQLite 文件路径(执行事实的权威记录)。
RAKUTEN_TRADING_DB_PATH=data/trading.db
# 页面证据目录(HTML 快照 + 截图 + meta.json,按 task_id 分子目录)。
RAKUTEN_EVIDENCE_DIR=data/evidence
# 抓取服务基地址。worker 需要商品数据(加购用 purchase 块)时出站请求这里。
RAKUTEN_SCRAPER_BASE_URL=
+9 -4
View File
@@ -118,16 +118,21 @@ USER rakuten
RUN /app/.venv/bin/playwright install chromium
# 健康检查走应用自带的 /health(轻量、不会触发真实抓取)。
# 端口取 RAKUTEN_HEALTH_PORT,未设时用抓取服务的端口——交易容器启动时把它
# 设成 31108 即可,个服务共用这一个镜像。
# 端口取 RAKUTEN_HEALTH_PORT,未设时用抓取服务的端口——交易/网关容器启动时把它
# 设成对应端口即可,个服务共用这一个镜像。
HEALTHCHECK --interval=30s --timeout=5s --start-period=20s --retries=3 \
CMD curl -fsS http://127.0.0.1:${RAKUTEN_HEALTH_PORT:-${RAKUTEN_APP_PORT}}/health || exit 1
# 31107 抓取服务,31108 交易服务(同一镜像,用 command 区分跑哪个)
EXPOSE 31107 31108
# 31107 抓取服务,31108 交易服务,31109 下单任务网关(同一镜像,用 command 区分跑哪个)
EXPOSE 31107 31108 31109
# 默认起抓取服务。交易服务覆盖 command 即可:
# docker run -e RAKUTEN_HEALTH_PORT=31108 -v ./.auth:/app/.auth \
# <image> python -m app.trading.main
# 注意交易服务持有登录态与订单监控,**只能起一个实例**,不要挂多副本。
#
# 网关部署在服务器侧(与抓取服务同机或邻近),任务队列 SQLite 需要持久化卷:
# docker run -e RAKUTEN_HEALTH_PORT=31109 -v ./data:/app/data \
# <image> python -m app.gateway.main
# 网关也**只能起一个实例**:SQLite 单连接 + 全局并发度 1,多实例会抢同一个 DB。
CMD ["python", "-m", "app.scraping.main"]
+58 -26
View File
@@ -9,36 +9,42 @@
两站均支持**搜索**、**商品详情**、**商家信息**与**商家名下商品**。
## 个部署单元
## 个部署单元
同一个仓库出**个服务**,分进程运行:
同一个仓库出**个服务**,分进程运行:
| | 抓取服务 `app.scraping` | 交易服务 `app.trading` |
| --- | --- | --- |
| 启动 | `python -m app.scraping.main`(:31107) | `python -m app.trading.main`(:31108) |
| 账号 | 全程匿名 | 必须带登录态 cookie |
| 状态 | 无状态,请求-响应 | 有状态:订单、页面证据、付款进度 |
| 失败重试 | 幂等,重试无代价 | **不可逆**,重复提交即重复下单 |
| 实例数 | 想开几个开几个 | **只能一个**(或按账号分片) |
| 出口 IP | 被限速换掉即可 | 频繁漂移会触发风控 |
| | 抓取服务 `app.scraping` | 交易服务 `app.trading` | 下单任务网关 `app.gateway` |
| --- | --- | --- | --- |
| 启动 | `python -m app.scraping.main`(:31107) | `python -m app.trading.main`(:31108) | `python -m app.gateway.main`(:31109) |
| 部署位置 | 服务器 | 本地(NAT 后) | 服务器 |
| 账号 | 全程匿名 | 必须带登录态 cookie | 不接触账号 |
| 状态 | 无状态,请求-响应 | 有状态:订单、页面证据、付款进度 | 有状态:任务队列 + 状态镜像 |
| 失败重试 | 幂等,重试无代价 | **不可逆**,重复提交即重复下单 | 任务编排侧,**不自动重投** |
| 实例数 | 想开几个开几个 | **只能一个**(或按账号分片) | **只能一个**(SQLite + 全局并发度 1) |
| 出口 IP | 被限速换掉即可 | 频繁漂移会触发风控 | 不出站到站点 |
拆开的决定性理由是「实例数」那一行,而不是「要不要登录」:登录态 cookie 全局唯一、
订单监控是常驻轮询,一旦与抓取同进程,抓取横向扩容就会把登录态和轮询任务复制 N 份,
让同一个账号被多个进程并发操作。
抓取与交易拆开的决定性理由是「实例数」那一行,而不是「要不要登录」:登录态 cookie
全局唯一、订单监控是常驻轮询,一旦与抓取同进程,抓取横向扩容就会把登录态和轮询任务
复制 N 份,让同一个账号被多个进程并发操作。
依赖方向固定为 `scraping → shared``trading → shared`,两侧互不 import
`tests/test_architecture.py` 会守着)。交易侧需要商品信息时,走抓取服务的 HTTP 接口——
下单要用的 `purchase` 块本来就是抓取服务的对外契约
网关单独成第三个部署单元,是因为它**有状态**(任务队列),不能塞进可多开的抓取服务,
也不能塞进在本地、零入站端口的交易服务——本地通过出站长轮询从这里领任务。
完整规格见 [docs/order-gateway.md](docs/order-gateway.md)
依赖方向固定为 `scraping → shared``trading → shared``gateway → shared`,三方互不 import
`tests/test_architecture.py` 会守着)。交易侧需要商品信息时走抓取服务的 HTTP 接口,
需要任务调度时走网关的 HTTP 接口——下单要用的 `purchase` 块本来就是抓取服务的对外契约。
抓取服务部署在服务器,交易服务部署在本地(便于管理账号、排查支付问题),本地在 NAT 后
没有公网入口,因此下单请求不是推进来的,而是由本地长轮询主动领取。
任务网关与本地 worker 的规格见 [docs/order-gateway.md](docs/order-gateway.md)(待实现)
任务网关与本地 worker 的规格见 [docs/order-gateway.md](docs/order-gateway.md)。
```
app/
shared/ 配置、错误码、日志、响应信封与鉴权(两侧共用,不认识侧)
shared/ 配置、错误码、日志、响应信封与鉴权、任务/订单状态枚举(三方共用,不认识任何一侧)
scraping/ 站点常量 / 会话 / 解析器 / 抓取路由(本 README 的绝大部分)
trading/ 登录态、(在建)加购下单付款与订单监控
trading/ 登录态、加购下单付款(在建)与本地下单 worker(领任务、留证据)
gateway/ 下单任务队列 + 状态镜像 + 长轮询领取接口
```
## 抓取原理
@@ -123,8 +129,23 @@ PC UA 在搜索页、详情页、店铺页上都能拿到完整模板。因此
| `POST /api/auth/status` | 查询登录态,默认真实探测一次 |
| `POST /api/auth/reload` | 人工重新登录后免重启换上新 cookie |
两个服务共用同一个 Bearer Token,错误码表也是同一份。
启动后分别在 `http://127.0.0.1:31107/docs``:31108/docs` 查看 OpenAPI 文档。
下单任务网关(:31109):
| 接口 | 说明 |
| --- | --- |
| `GET /health` | 健康检查,含 worker 心跳、长时间无人领任务告警 |
| `POST /api/orders` | 上游提交下单意图(幂等) |
| `GET /api/orders/lease` | 本地 worker 长轮询领取(全局并发度 1) |
| `POST /api/orders/{id}/renew` | 续租(worker 在长任务里每 60s 调一次) |
| `POST /api/orders/{id}/report` | 本地回报订单状态(同 state 重复上报幂等) |
| `POST /api/orders/{id}/reclaim` | 把 stale 任务重新租给 worker(**绝不自动重投**) |
| `GET /api/orders/{id}` | 任务详情 + 完整状态历史 |
| `GET /api/orders` | 任务列表(运维与上游对账用) |
网关的契约与状态机详见 [docs/order-gateway.md](docs/order-gateway.md)。
三个服务共用同一个 Bearer Token,错误码表也是同一份。
启动后分别在 `http://127.0.0.1:31107/docs``:31108/docs``:31109/docs` 查看 OpenAPI 文档。
## 安装
@@ -138,7 +159,7 @@ uv sync --extra dev --extra browser
## 启动
启动前建议先按 `.env.example` 配置 `.env`个服务共用这一份。
启动前建议先按 `.env.example` 配置 `.env`个服务共用这一份。
```bash
# 抓取服务,默认 0.0.0.0:31107;可多开实例
@@ -146,9 +167,12 @@ uv sync --extra dev --extra browser
# 交易服务,默认 0.0.0.0:31108;只能起一个实例
.venv/Scripts/python.exe -m app.trading.main
# 下单任务网关,默认 0.0.0.0:31109;只能起一个实例(SQLite + 全局并发度 1)
.venv/Scripts/python.exe -m app.gateway.main
```
只需要抓取时不必起交易服务。交易服务启动前要先人工登录一次:
只需要抓取时不必起交易/网关服务。交易服务启动前要先人工登录一次:
```bash
.venv/Scripts/python.exe scripts/login.py --site all
@@ -557,20 +581,28 @@ cookie 落在 `.auth/`(已 gitignore,内含可直接冒充账号的凭据,
| 5002 | 加购失败(交易服务) | 400 |
| 5003 | 下单失败(交易服务) | 400 |
| 5004 | 下单安全闸门未通过:未显式确认或金额超上限(交易服务) | 400 |
| 6001 | 任务不存在(网关) | 404 |
| 6002 | 租约无效:不是持有者、已过期或任务已终结(网关) | 409 |
| 6003 | 任务状态不允许该操作(如对已终结任务 reclaim)(网关) | 409 |
| 6004 | 已有任务在执行中,本次不发放(正常返回空,仅诊断用) | 200 |
错误码在两站、个服务之间通用。ラクマ 链路不会出现 `3002`(无反爬拦截行为)
错误码在两站、个服务之间通用。ラクマ 链路不会出现 `3002`(无反爬拦截行为)
`4002`(无子站跳转);`5xxx` 只会来自交易服务——抓取服务全程匿名,不会有登录态问题。
`5001``5004` 都标记为不可重试:前者要人工重新登录,后者要调用方改入参。
`6xxx` 只会来自网关,全部标记为不可重试——任务编排侧重试无意义,部分场景
(如租约过期)重试可能变成重复下单。
## 常用环境变量
完整列表见 [.env.example](.env.example)。配置项前缀统一为 `RAKUTEN_`个服务共用同一份
完整列表见 [.env.example](.env.example)。配置项前缀统一为 `RAKUTEN_`个服务共用同一份
配置文件、各读各的那部分;两站共用同一套抓取参数(并发数、超时、重试次数),
`SESSION_TTL` 与浏览器兜底只对乐天链路生效。
- 抓取服务:`RAKUTEN_APP_HOST``RAKUTEN_APP_PORT`(默认 `31107`)、`RAKUTEN_APP_ENV`
- 交易服务:`RAKUTEN_TRADING_HOST``RAKUTEN_TRADING_PORT`(默认 `31108`)、`RAKUTEN_AUTH_STATE_DIR`(默认 `.auth`)、`RAKUTEN_ORDER_MAX_TOTAL_YEN`(默认 `30000`
- 鉴权`RAKUTEN_BEARER_TOKEN`(两个服务共用
- 下单任务网关`RAKUTEN_GATEWAY_HOST``RAKUTEN_GATEWAY_PORT`(默认 `31109`)、`RAKUTEN_GATEWAY_DB_PATH`(默认 `data/gateway.db`)、`RAKUTEN_LEASE_TTL_SECONDS`(默认 `300`)、`RAKUTEN_WORKER_OFFLINE_ALERT_SECONDS`(默认 `300`
- 本地下单 worker(在交易服务内,按 `RAKUTEN_ORDER_GATEWAY_URL` 是否配置决定是否启动):`RAKUTEN_ORDER_GATEWAY_URL``RAKUTEN_WORKER_ID``RAKUTEN_TRADING_DB_PATH`(默认 `data/trading.db`)、`RAKUTEN_EVIDENCE_DIR`(默认 `data/evidence`)、`RAKUTEN_SCRAPER_BASE_URL`
- 鉴权:`RAKUTEN_BEARER_TOKEN`(三个服务共用)
- 抓取:`RAKUTEN_MAX_SITE_CONCURRENCY`(默认 `8`,两站各自独立计数)、`RAKUTEN_HTTP_MAX_ATTEMPTS`(默认 `3`)、`RAKUTEN_SESSION_TTL_SECONDS`(默认 `1800`,仅乐天)
- 浏览器兜底(仅乐天):`RAKUTEN_BROWSER_FALLBACK_ENABLED``RAKUTEN_BROWSER_HEADLESS``RAKUTEN_BROWSER_CHANNEL`
- 代理(需日本 IP 时):`RAKUTEN_PROXY_SERVER``RAKUTEN_PROXY_USERNAME``RAKUTEN_PROXY_PASSWORD`
+12
View File
@@ -0,0 +1,12 @@
"""下单任务网关:第三个部署单元
抓取服务(app.scraping)匿名、无状态、可多开;交易服务(app.trading)带账号、
有状态、单实例;本网关是任务队列与状态镜像,部署在服务器侧,**独立于两者**:
- 与抓取服务分进程的原因:抓取实例可以多开,而任务队列有状态,多实例会抢同一批
任务(同一账号的写操作必须串行)。
- 与交易服务分进程的原因:交易在本地(NAT 后无公网入口),网关在服务器,本地
通过出站长轮询从这里领任务。
完整规格见 docs/order-gateway.md。
"""
+1
View File
@@ -0,0 +1 @@
"""网关 API 包"""
+1
View File
@@ -0,0 +1 @@
"""网关路由集合"""
+23
View File
@@ -0,0 +1,23 @@
"""网关健康检查路由"""
from fastapi import APIRouter, Depends
from app.gateway.container import GatewayContainer
from app.gateway.models import GatewayHealthData
from app.shared.api import ApiResponse, get_container
router = APIRouter(tags=["health"])
@router.get("/health", response_model=ApiResponse[GatewayHealthData])
async def health(
container: GatewayContainer = Depends(get_container),
) -> ApiResponse[GatewayHealthData]:
"""网关健康状态
只做规格 §4.7 的两条兜底告警:worker 失联、任务长时间无人领。
付款期限监控不在网关——本地机 7×24 在线,那套逻辑放本地。
"""
data = await container.task_queue.health_snapshot()
return ApiResponse[GatewayHealthData](
success=True, msg="success", data=data, code=0
)
+181
View File
@@ -0,0 +1,181 @@
"""下单任务网关路由
承担规格 §4 的全部接口:
- POST /api/orders 上游提交下单意图(幂等)
- GET /api/orders/lease 本地长轮询领取
- POST /api/orders/{id}/renew 续租
- POST /api/orders/{id}/report 本地回报状态
- POST /api/orders/{id}/reclaim 把 stale 任务重新租给 worker(绝不自动重投)
- GET /api/orders/{id} 任务详情 + 完整状态历史
- GET /api/orders 任务列表(运维与上游对账)
handler 风格与 app/scraping/api/routes/scrape.py 保持一致:不在 handler 里写
try/except,业务异常通过 AppError 自动套进 register_exception_handlers。
"""
from __future__ import annotations
from fastapi import APIRouter, Depends, Query
from app.gateway.container import GatewayContainer
from app.gateway.models import (
LeaseData,
RenewData,
ReclaimData,
ReportData,
SubmitOrderData,
SubmitOrderRequest,
TaskDetail,
TaskListData,
RenewRequest,
ReclaimRequest,
ReportRequest,
)
from app.shared.api import ApiResponse, get_container, require_bearer_token
router = APIRouter(prefix="/api/orders", tags=["orders"])
@router.post(
"",
response_model=ApiResponse[SubmitOrderData],
dependencies=[Depends(require_bearer_token)],
)
async def submit_order(
payload: SubmitOrderRequest,
container: GatewayContainer = Depends(get_container),
) -> ApiResponse[SubmitOrderData]:
"""上游提交下单意图
重复提交同一 task_id 不新建任务,返回既有任务且 created=false——上游重发
不会变成两单。
"""
data = await container.task_queue.submit(
task_id=payload.task_id, site=payload.site, intent=payload.intent
)
return ApiResponse[SubmitOrderData](
success=True, msg="success", data=data, code=0
)
@router.get(
"/lease",
response_model=ApiResponse[LeaseData | None],
dependencies=[Depends(require_bearer_token)],
)
async def lease_order(
worker_id: str = Query(...),
wait: int = Query(default=30, ge=0, le=300),
site: str | None = Query(default=None),
container: GatewayContainer = Depends(get_container),
) -> ApiResponse[LeaseData | None]:
"""本地长轮询领取
无可领任务时挂起到 wait 秒后返回 data: null,HTTP 仍 200。已有 leased/running
任务时立即返回空(全局并发度 1)。
"""
data = await container.task_queue.lease(
worker_id=worker_id,
wait=wait,
site=site,
max_wait=container.settings.lease_max_wait_seconds,
)
return ApiResponse[LeaseData | None](
success=True, msg="success", data=data, code=0
)
@router.post(
"/{task_id}/renew",
response_model=ApiResponse[RenewData],
dependencies=[Depends(require_bearer_token)],
)
async def renew_order(
task_id: str,
payload: RenewRequest,
container: GatewayContainer = Depends(get_container),
) -> ApiResponse[RenewData]:
"""续租。worker 在长任务中每 60 秒调一次,避免租约 TTL 误判"""
data = await container.task_queue.renew(task_id, payload.worker_id)
return ApiResponse[RenewData](success=True, msg="success", data=data, code=0)
@router.post(
"/{task_id}/report",
response_model=ApiResponse[ReportData],
dependencies=[Depends(require_bearer_token)],
)
async def report_order(
task_id: str,
payload: ReportRequest,
container: GatewayContainer = Depends(get_container),
) -> ApiResponse[ReportData]:
"""本地回报订单状态
同一 (task_id, state) 重复上报幂等。terminal=true 时释放租约并推到终态。
"""
data = await container.task_queue.report(
task_id,
worker_id=payload.worker_id,
state=payload.state,
payable_yen=payload.payable_yen,
pay_deadline=payload.pay_deadline,
site_order_id=payload.site_order_id,
evidence_ref=payload.evidence_ref,
detail=payload.detail,
terminal=payload.terminal,
terminal_status=payload.terminal_status.value if payload.terminal_status else None,
)
return ApiResponse[ReportData](success=True, msg="success", data=data, code=0)
@router.post(
"/{task_id}/reclaim",
response_model=ApiResponse[ReclaimData],
dependencies=[Depends(require_bearer_token)],
)
async def reclaim_order(
task_id: str,
payload: ReclaimRequest,
container: GatewayContainer = Depends(get_container),
) -> ApiResponse[ReclaimData]:
"""把 stale 任务重新租给 worker(恢复领取)
返回体的 lease_count 必然 > 1,worker 必须先核对站点订单列表再决定是否
继续执行——核对不出结论时报 needs_human,绝不重新提交。
"""
data = await container.task_queue.reclaim(task_id, payload.worker_id)
return ApiResponse[ReclaimData](success=True, msg="success", data=data, code=0)
@router.get(
"/{task_id}",
response_model=ApiResponse[TaskDetail],
dependencies=[Depends(require_bearer_token)],
)
async def get_order(
task_id: str,
container: GatewayContainer = Depends(get_container),
) -> ApiResponse[TaskDetail]:
"""任务详情 + 完整状态历史"""
data = await container.task_queue.get_task_detail(task_id)
return ApiResponse[TaskDetail](success=True, msg="success", data=data, code=0)
@router.get(
"",
response_model=ApiResponse[TaskListData],
dependencies=[Depends(require_bearer_token)],
)
async def list_orders(
status: str | None = Query(default=None),
site: str | None = Query(default=None),
limit: int = Query(default=50, ge=1, le=500),
offset: int = Query(default=0, ge=0),
container: GatewayContainer = Depends(get_container),
) -> ApiResponse[TaskListData]:
"""任务列表(运维与上游对账用)"""
data = await container.task_queue.list_tasks(
status=status, site=site, limit=limit, offset=offset
)
return ApiResponse[TaskListData](success=True, msg="success", data=data, code=0)
+29
View File
@@ -0,0 +1,29 @@
"""网关容器:集中管理网关侧服务实例,用于依赖注入"""
from __future__ import annotations
import asyncio
import logging
from dataclasses import dataclass
from app.gateway.db import GatewayDB
from app.gateway.task_queue import TaskQueue
from app.shared.config import Settings
logger = logging.getLogger(__name__)
@dataclass(slots=True)
class GatewayContainer:
"""网关容器
与抓取/交易容器同样的依赖注入风格,但持有的是任务队列与 SQLite 连接,
生命周期由 main.lifespan 管理:start 时打开 DB,close 时关闭。
`sweep_task` 是常驻后台扫描,把过期的 leased/running 推到 stale。
即使没有 lease 请求,过期的任务也会被及时发现(规格 §5 的关键约束)。
"""
settings: Settings
db: GatewayDB
task_queue: TaskQueue
sweep_task: asyncio.Task | None = None
+356
View File
@@ -0,0 +1,356 @@
"""网关 SQLite 访问层:三张表 + 原生 SQL,无业务逻辑
业务规则(状态机、并发度、幂等)放在 `task_queue.py`,本模块只负责持久化与查询,
返回 dataclass 行。所有时间戳以 ISO8601 UTC 字符串存储(以 `Z` 结尾),便于
跨进程对账;时间戳运算在 task_queue 层完成。
写操作由 task_queue 的 asyncio.Lock 串行化(详见该模块),本层不重复加锁,
因此**调用方必须确保写操作在外层锁的保护下进行**。
"""
from __future__ import annotations
import json
import logging
from dataclasses import dataclass
from pathlib import Path
from typing import Any
import aiosqlite
logger = logging.getLogger(__name__)
SCHEMA = """
CREATE TABLE IF NOT EXISTS tasks (
task_id TEXT PRIMARY KEY,
site TEXT NOT NULL,
intent_json TEXT NOT NULL,
status TEXT NOT NULL,
lease_owner TEXT,
lease_expires_at TEXT,
lease_count INTEGER NOT NULL DEFAULT 0,
created_at TEXT NOT NULL,
updated_at TEXT NOT NULL
);
CREATE INDEX IF NOT EXISTS idx_tasks_pending ON tasks(status, created_at);
CREATE TABLE IF NOT EXISTS task_reports (
task_id TEXT NOT NULL,
state TEXT NOT NULL,
payable_yen INTEGER,
pay_deadline TEXT,
site_order_id TEXT,
evidence_ref TEXT,
detail TEXT,
reported_at TEXT NOT NULL,
PRIMARY KEY (task_id, state)
);
CREATE TABLE IF NOT EXISTS workers (
worker_id TEXT PRIMARY KEY,
last_seen_at TEXT NOT NULL
);
"""
@dataclass(slots=True)
class TaskRow:
"""tasks 表一行的强类型视图"""
task_id: str
site: str
intent_json: str
status: str
lease_owner: str | None
lease_expires_at: str | None
lease_count: int
created_at: str
updated_at: str
@property
def intent(self) -> dict[str, Any]:
return json.loads(self.intent_json)
@dataclass(slots=True)
class ReportRow:
"""task_reports 表一行的强类型视图"""
task_id: str
state: str
payable_yen: int | None
pay_deadline: str | None
site_order_id: str | None
evidence_ref: str | None
detail: str
reported_at: str
@dataclass(slots=True)
class WorkerRow:
"""workers 表一行的强类型视图"""
worker_id: str
last_seen_at: str
def _row_to_task(row: aiosqlite.Row) -> TaskRow:
return TaskRow(
task_id=row["task_id"],
site=row["site"],
intent_json=row["intent_json"],
status=row["status"],
lease_owner=row["lease_owner"],
lease_expires_at=row["lease_expires_at"],
lease_count=row["lease_count"],
created_at=row["created_at"],
updated_at=row["updated_at"],
)
def _row_to_report(row: aiosqlite.Row) -> ReportRow:
return ReportRow(
task_id=row["task_id"],
state=row["state"],
payable_yen=row["payable_yen"],
pay_deadline=row["pay_deadline"],
site_order_id=row["site_order_id"],
evidence_ref=row["evidence_ref"],
detail=row["detail"],
reported_at=row["reported_at"],
)
def _row_to_worker(row: aiosqlite.Row) -> WorkerRow:
return WorkerRow(worker_id=row["worker_id"], last_seen_at=row["last_seen_at"])
class GatewayDB:
"""网关 SQLite 访问对象
单连接 + aiosqlite 的内部线程。生命周期由 `GatewayContainer` 管理:
`start()` 在 lifespan 启动时调用,`close()` 在关闭时调用。
"""
def __init__(self, db_path: Path):
self._db_path = db_path
self._conn: aiosqlite.Connection | None = None
async def start(self) -> None:
"""打开连接并初始化 schema(IF NOT EXISTS,可重复执行)"""
self._db_path.parent.mkdir(parents=True, exist_ok=True)
self._conn = await aiosqlite.connect(str(self._db_path))
self._conn.row_factory = aiosqlite.Row
await self._conn.executescript(SCHEMA)
await self._conn.commit()
logger.info("网关 DB 已就绪:%s", self._db_path)
async def close(self) -> None:
if self._conn is not None:
await self._conn.close()
self._conn = None
@property
def conn(self) -> aiosqlite.Connection:
if self._conn is None:
raise RuntimeError("GatewayDB 未启动:先调用 start()")
return self._conn
# ---- tasks ----
async def get_task(self, task_id: str) -> TaskRow | None:
async with self.conn.execute("SELECT * FROM tasks WHERE task_id = ?", (task_id,)) as cur:
row = await cur.fetchone()
return _row_to_task(row) if row else None
async def insert_task(self, row: TaskRow) -> bool:
"""插入新任务。返回 True=新建,False=task_id 已存在(幂等命中)"""
try:
await self.conn.execute(
"INSERT INTO tasks (task_id, site, intent_json, status, "
"lease_owner, lease_expires_at, lease_count, created_at, updated_at) "
"VALUES (?, ?, ?, ?, NULL, NULL, 0, ?, ?)",
(
row.task_id,
row.site,
row.intent_json,
row.status,
row.created_at,
row.updated_at,
),
)
await self.conn.commit()
return True
except aiosqlite.IntegrityError:
return False
async def update_task(
self,
task_id: str,
*,
status: str | None = None,
lease_owner: str | None = None,
lease_expires_at: str | None = None,
lease_count: int | None = None,
updated_at: str | None = None,
clear_lease: bool = False,
) -> None:
"""更新任务字段。clear_lease=True 时把 lease_owner/expires_at 置 NULL"""
sets: list[str] = []
params: list[Any] = []
if status is not None:
sets.append("status = ?")
params.append(status)
if lease_owner is not None:
sets.append("lease_owner = ?")
params.append(lease_owner)
if lease_expires_at is not None:
sets.append("lease_expires_at = ?")
params.append(lease_expires_at)
if lease_count is not None:
sets.append("lease_count = ?")
params.append(lease_count)
if updated_at is not None:
sets.append("updated_at = ?")
params.append(updated_at)
if clear_lease:
sets.append("lease_owner = NULL")
sets.append("lease_expires_at = NULL")
if not sets:
return
params.append(task_id)
await self.conn.execute(
f"UPDATE tasks SET {', '.join(sets)} WHERE task_id = ?",
params,
)
await self.conn.commit()
async def list_tasks(
self,
*,
status: str | None = None,
site: str | None = None,
limit: int = 50,
offset: int = 0,
) -> tuple[list[TaskRow], int]:
"""分页列出任务,按 created_at 升序"""
where = []
params: list[Any] = []
if status:
where.append("status = ?")
params.append(status)
if site:
where.append("site = ?")
params.append(site)
clause = f"WHERE {' AND '.join(where)}" if where else ""
async with self.conn.execute(
f"SELECT COUNT(*) FROM tasks {clause}",
params,
) as cur:
total = (await cur.fetchone())[0]
sql = f"SELECT * FROM tasks {clause} ORDER BY created_at ASC LIMIT ? OFFSET ?"
async with self.conn.execute(sql, [*params, limit, offset]) as cur:
rows = await cur.fetchall()
return [_row_to_task(r) for r in rows], total
async def list_tasks_in_statuses(self, statuses: tuple[str, ...]) -> list[TaskRow]:
"""取处于给定状态集合的全部任务(不分页,用于健康检查的 active 列表)"""
if not statuses:
return []
placeholders = ", ".join("?" for _ in statuses)
sql = (
f"SELECT * FROM tasks WHERE status IN ({placeholders}) "
"ORDER BY created_at ASC"
)
async with self.conn.execute(sql, statuses) as cur:
rows = await cur.fetchall()
return [_row_to_task(r) for r in rows]
async def pick_queued_task(self, site: str | None) -> TaskRow | None:
"""取最早的 queued 任务,可选按站点过滤"""
if site:
sql = "SELECT * FROM tasks WHERE status = ? AND site = ? ORDER BY created_at ASC LIMIT 1"
params: tuple[Any, ...] = ("queued", site)
else:
sql = "SELECT * FROM tasks WHERE status = ? ORDER BY created_at ASC LIMIT 1"
params = ("queued",)
async with self.conn.execute(sql, params) as cur:
row = await cur.fetchone()
return _row_to_task(row) if row else None
# ---- task_reports ----
async def upsert_report(self, row: ReportRow) -> bool:
"""写入一条 report。同一 (task_id, state) 已存在则覆盖(幂等)。
返回 True=本次新写入,False=覆盖了既有行。
先 SELECT 再 INSERT OR REPLACE,避免依赖 SQLite rowcount——后者在
INSERT OR REPLACE 上即使命中既有行也报 1,无法区分。
"""
async with self.conn.execute(
"SELECT 1 FROM task_reports WHERE task_id = ? AND state = ?",
(row.task_id, row.state),
) as cur:
existed = await cur.fetchone() is not None
await self.conn.execute(
"INSERT OR REPLACE INTO task_reports "
"(task_id, state, payable_yen, pay_deadline, site_order_id, "
"evidence_ref, detail, reported_at) VALUES (?, ?, ?, ?, ?, ?, ?, ?)",
(
row.task_id,
row.state,
row.payable_yen,
row.pay_deadline,
row.site_order_id,
row.evidence_ref,
row.detail,
row.reported_at,
),
)
await self.conn.commit()
return not existed
async def get_reports(self, task_id: str) -> list[ReportRow]:
async with self.conn.execute(
"SELECT * FROM task_reports WHERE task_id = ? ORDER BY reported_at ASC",
(task_id,),
) as cur:
rows = await cur.fetchall()
return [_row_to_report(r) for r in rows]
async def get_latest_state(self, task_id: str) -> str | None:
"""取该任务最近一次 report 的 state,无 report 返回 None"""
async with self.conn.execute(
"SELECT state FROM task_reports WHERE task_id = ? "
"ORDER BY reported_at DESC LIMIT 1",
(task_id,),
) as cur:
row = await cur.fetchone()
return row["state"] if row else None
# ---- workers ----
async def upsert_worker(self, worker_id: str, last_seen_at: str) -> None:
await self.conn.execute(
"INSERT INTO workers (worker_id, last_seen_at) VALUES (?, ?) "
"ON CONFLICT(worker_id) DO UPDATE SET last_seen_at = excluded.last_seen_at",
(worker_id, last_seen_at),
)
await self.conn.commit()
async def list_workers(self) -> list[WorkerRow]:
async with self.conn.execute("SELECT * FROM workers ORDER BY last_seen_at DESC") as cur:
rows = await cur.fetchall()
return [_row_to_worker(r) for r in rows]
# ---- 计数 ----
async def count_by_status(self, status: str) -> int:
async with self.conn.execute(
"SELECT COUNT(*) FROM tasks WHERE status = ?",
(status,),
) as cur:
return (await cur.fetchone())[0]
+124
View File
@@ -0,0 +1,124 @@
"""下单任务网关入口:FastAPI 应用创建与生命周期管理
部署在服务器侧,与抓取服务(:31107)、交易服务(:31108)并列。本地 worker 通过
出站长轮询从这里领任务;上游业务系统通过 POST /api/orders 提交下单意图。
为什么必须独立部署单元(不能塞进抓取服务):抓取无状态可多开,任务队列有状态,
多实例会抢同一批任务,同一账号的写操作必须串行(详见 docs/order-gateway.md §2)。
"""
from __future__ import annotations
import asyncio
import logging
from contextlib import asynccontextmanager
from fastapi import FastAPI
from app.gateway.api.routes.health import router as health_router
from app.gateway.api.routes.orders import router as orders_router
from app.gateway.container import GatewayContainer
from app.gateway.db import GatewayDB
from app.gateway.task_queue import TaskQueue
from app.shared.api import register_exception_handlers
from app.shared.config import get_settings
from app.shared.logging_setup import configure_logging
logger = logging.getLogger(__name__)
SWEEP_INTERVAL_SECONDS = 60
def build_container() -> GatewayContainer:
"""构建网关容器:DB + 任务队列"""
settings = get_settings()
db = GatewayDB(settings.gateway_db_path_resolved)
task_queue = TaskQueue(
db,
lease_ttl_seconds=settings.lease_ttl_seconds,
worker_offline_alert_seconds=settings.worker_offline_alert_seconds,
)
return GatewayContainer(settings=settings, db=db, task_queue=task_queue)
async def _sweep_loop(container: GatewayContainer) -> None:
"""常驻后台任务:每 60 秒扫一次过期租约,把 leased/running 推到 stale
lease 请求本身也会顺带扫一次,但 worker 退场后没人来 lease,必须靠这个
兜底——否则过期的任务永远停在 active 状态,健康检查看不到,stale 任务也
取不出来 reclaim。
"""
while True:
try:
await asyncio.sleep(SWEEP_INTERVAL_SECONDS)
swept = await container.task_queue.sweep()
if swept:
logger.info("sweep 把 %s 个过期任务置为 stale", swept)
except asyncio.CancelledError:
raise
except Exception: # noqa: BLE001
# 后台任务不能因为偶发错误退出,否则过期任务再也无人清理
logger.exception("sweep 后台任务出错,将继续重试")
@asynccontextmanager
async def lifespan(app: FastAPI):
"""应用生命周期:启动 DB、起 sweep 后台任务、关闭时反序释放"""
container = build_container()
app.state.container = container
configure_logging(container.settings)
logger.info(
"网关启动:%s:%s", container.settings.gateway_host, container.settings.gateway_port
)
logger.info("当前环境:%s", container.settings.app_env)
logger.info("DB 路径:%s", container.settings.gateway_db_path_resolved)
await container.db.start()
# 启动时先扫一次:上次进程退出时可能留有 leased/running 的过期任务
startup_swept = await container.task_queue.sweep()
if startup_swept:
logger.warning("启动时把 %s 个遗留过期任务置为 stale", startup_swept)
container.sweep_task = asyncio.create_task(
_sweep_loop(container), name="gateway-sweep"
)
try:
yield
finally:
if container.sweep_task is not None:
container.sweep_task.cancel()
try:
await container.sweep_task
except asyncio.CancelledError:
pass
container.sweep_task = None
await container.db.close()
def create_app() -> FastAPI:
"""创建 FastAPI 应用实例,注册路由和异常处理器"""
app = FastAPI(title="Rakuten Order Gateway", lifespan=lifespan)
app.include_router(health_router)
app.include_router(orders_router)
register_exception_handlers(app)
return app
app = create_app()
if __name__ == "__main__":
import uvicorn
settings = get_settings()
configure_logging(settings)
uvicorn.run(
"app.gateway.main:app",
host=settings.gateway_host,
port=settings.gateway_port,
log_config=None,
timeout_keep_alive=120,
# 单进程:SQLite 单连接 + 全局并发度 1,多进程会抢同一个 DB
workers=1,
)
+194
View File
@@ -0,0 +1,194 @@
"""网关 API 数据模型:请求体与响应体
intent 字段刻意保留成 `dict[str, Any]`——网关不解释下单意图,结构由 trading 侧
定义。网关只负责把它存下来、原样吐给 worker,避免业务规则悄悄渗进任务队列。
"""
from __future__ import annotations
from typing import Any
from pydantic import BaseModel, Field
from app.shared.task_state import OrderState, TaskStatus
# ---- POST /api/orders ----
class SubmitOrderRequest(BaseModel):
"""上游提交下单意图
task_id 可选:上游自带的幂等键。不传则服务端生成。同一个 task_id 重复提交
不新建任务,返回既有任务且 created=false。
"""
task_id: str | None = None
site: str
intent: dict[str, Any]
class SubmitOrderData(BaseModel):
"""提交响应"""
task_id: str
status: TaskStatus
created: bool # True=本次新建,False=命中既有任务(幂等)
# ---- GET /api/orders/lease ----
class LeaseData(BaseModel):
"""lease 响应
无任务可领时整个 data 为 null(HTTP 仍 200)。lease_count > 1 表示这是恢复
领取(stale → reclaim),worker 必须先核对站点订单列表,见 runner 主循环。
"""
task_id: str
site: str
intent: dict[str, Any]
lease_expires_at: str # ISO8601 UTC
lease_count: int
known_state: OrderState | None # 之前上报过的最新订单状态;首次领取为 null
# ---- POST /api/orders/{id}/renew ----
class RenewRequest(BaseModel):
"""续租请求"""
worker_id: str
class RenewData(BaseModel):
"""续租响应"""
task_id: str
lease_expires_at: str
lease_count: int
# ---- POST /api/orders/{id}/report ----
class ReportRequest(BaseModel):
"""本地回报订单状态
同一 (task_id, state) 重复上报是幂等的——网络抖动导致 worker 重发时覆盖同一行,
不产生第二条记录。terminal=true 时释放租约并把任务推到终态。
"""
worker_id: str
state: OrderState
payable_yen: int | None = None
pay_deadline: str | None = None # ISO8601
site_order_id: str | None = None
evidence_ref: str | None = None # 本地相对路径,不含扩展名
detail: str = ""
terminal: bool = False
terminal_status: TaskStatus | None = None # terminal=true 时指定终态,缺省按 state 推断
class ReportData(BaseModel):
"""回报响应"""
task_id: str
status: TaskStatus
recorded: bool # False=同 (task_id, state) 已存在,本次为幂等覆盖;True=新写入
# ---- POST /api/orders/{id}/reclaim ----
class ReclaimRequest(BaseModel):
"""把 stale 任务重新租给 worker"""
worker_id: str
class ReclaimData(BaseModel):
"""reclaim 响应,结构与 LeaseData 一致,但 lease_count 必然 > 1"""
task_id: str
site: str
intent: dict[str, Any]
lease_expires_at: str
lease_count: int
known_state: OrderState | None
# ---- GET /api/orders/{id} 与 GET /api/orders ----
class ReportEntry(BaseModel):
"""task_reports 单行"""
state: OrderState
payable_yen: int | None = None
pay_deadline: str | None = None
site_order_id: str | None = None
evidence_ref: str | None = None
detail: str = ""
reported_at: str
class TaskDetail(BaseModel):
"""单个任务详情"""
task_id: str
site: str
intent: dict[str, Any]
status: TaskStatus
lease_owner: str | None = None
lease_expires_at: str | None = None
lease_count: int = 0
created_at: str
updated_at: str
latest_state: OrderState | None = None
reports: list[ReportEntry] = Field(default_factory=list)
class TaskListData(BaseModel):
"""任务列表"""
items: list[TaskDetail]
total: int
limit: int
offset: int
# ---- GET /health ----
class WorkerHealthEntry(BaseModel):
"""单个 worker 的健康指标"""
worker_id: str
last_seen_at: str
last_seen_seconds: int
class QueuedAlertEntry(BaseModel):
"""长时间无人领的任务告警项"""
task_id: str
site: str
created_at: str
age_seconds: int
class GatewayHealthData(BaseModel):
"""网关健康状态
只做规格 §4.7 的两条兜底告警:worker 失联、任务长时间无人领。付款期限监控
不在网关——本地机 7×24 在线,那套逻辑放本地。
"""
status: str # "ok" 或 "degraded"
queued_count: int
active_tasks: list[TaskDetail] = Field(default_factory=list)
workers: list[WorkerHealthEntry] = Field(default_factory=list)
offline_workers: list[WorkerHealthEntry] = Field(default_factory=list)
stale_queued_tasks: list[QueuedAlertEntry] = Field(default_factory=list)
+505
View File
@@ -0,0 +1,505 @@
"""任务队列业务逻辑:状态机、租约、并发度、幂等
所有写操作通过 `self._lock` 串行化。该锁同时保护「检查 + 写入」的原子序列
(如 lease 时「检查是否已有 active 任务 → 取 queued → 改为 leased」),避免两个
lease 请求并发取到同一个任务。
长轮询通过 `self._cond`(基于同一把锁的 Condition)唤醒:submit / report(terminal)
/ reclaim 失败等任何可能让「队列前进」的事件都 notify_all。lease 在无可领任务时
wait_for,超时返回 None。
绝不自动重投(docs/order-gateway.md §5):sweep 把过期任务置 stale 后**不**回到
queued,普通 lease 取不到它;恢复只能走 reclaim。
"""
from __future__ import annotations
import asyncio
import logging
import secrets
from datetime import datetime, timedelta, timezone
from app.shared.task_state import (
ACTIVE_STATUSES,
LEASABLE_STATUSES,
RECLAIMABLE_STATUSES,
TERMINAL_STATUSES,
TaskStatus,
)
from app.gateway.db import GatewayDB, ReportRow, TaskRow
from app.gateway.models import (
GatewayHealthData,
LeaseData,
QueuedAlertEntry,
ReportEntry,
RenewData,
ReclaimData,
ReportData,
SubmitOrderData,
TaskDetail,
TaskListData,
WorkerHealthEntry,
)
from app.shared.errors import (
InvalidTaskStateError,
LeaseInvalidError,
TaskNotFoundError,
)
logger = logging.getLogger(__name__)
def utcnow() -> datetime:
return datetime.now(timezone.utc)
def to_iso(dt: datetime) -> str:
"""统一时间戳格式:ISO8601 UTC,秒精度,Z 结尾"""
return dt.astimezone(timezone.utc).replace(microsecond=0).isoformat().replace("+00:00", "Z")
def parse_iso(s: str) -> datetime:
"""解析 to_iso 产出的字符串(也兼容带 +00:00 的变体)"""
if s.endswith("Z"):
s = s[:-1] + "+00:00"
return datetime.fromisoformat(s)
def _generate_task_id() -> str:
"""服务端生成的 task_id:日期 + 8 字节随机 hex,避免与上游自带的冲突"""
return f"po-{to_iso(utcnow())[:10].replace('-', '')}-{secrets.token_hex(4)}"
def _task_to_detail(task: TaskRow, latest_state: str | None, reports: list[ReportEntry]) -> TaskDetail:
return TaskDetail(
task_id=task.task_id,
site=task.site,
intent=task.intent,
status=TaskStatus(task.status),
lease_owner=task.lease_owner,
lease_expires_at=task.lease_expires_at,
lease_count=task.lease_count,
created_at=task.created_at,
updated_at=task.updated_at,
latest_state=latest_state,
reports=reports,
)
class TaskQueue:
"""任务队列业务逻辑
持有一个 GatewayDB 实例。所有写操作与「检查 + 写入」原子序列都在 `self._lock`
保护下;长轮询等待通过 `self._cond` 唤醒。
"""
def __init__(self, db: GatewayDB, *, lease_ttl_seconds: int, worker_offline_alert_seconds: int):
self._db = db
self._lease_ttl = lease_ttl_seconds
self._worker_offline_alert_seconds = worker_offline_alert_seconds
self._lock = asyncio.Lock()
self._cond = asyncio.Condition(self._lock)
# ---- 提交 ----
async def submit(self, *, task_id: str | None, site: str, intent: dict) -> SubmitOrderData:
"""上游提交下单意图。task_id 缺省时服务端生成;重复提交幂等"""
import json
tid = task_id or _generate_task_id()
now = to_iso(utcnow())
row = TaskRow(
task_id=tid,
site=site,
intent_json=json.dumps(intent, ensure_ascii=False),
status=TaskStatus.QUEUED.value,
lease_owner=None,
lease_expires_at=None,
lease_count=0,
created_at=now,
updated_at=now,
)
async with self._lock:
created = await self._db.insert_task(row)
if created:
self._cond.notify_all()
else:
existing = await self._db.get_task(tid)
# 既不可能创建失败又拿不到既有行:除非被并发删除,按罕见错误处理
if existing is None:
raise RuntimeError(f"任务 {tid} 既未新建也无法读取,状态异常")
row = existing
return SubmitOrderData(task_id=row.task_id, status=TaskStatus(row.status), created=created)
# ---- 领取 ----
async def lease(
self,
*,
worker_id: str,
wait: int,
site: str | None,
max_wait: int,
) -> LeaseData | None:
"""本地长轮询领取
- 全局并发度 1:已有 leased/running 任务时立即返回 None
- 无可领任务时挂起最多 `min(wait, max_wait)` 秒,超时返回 None
- 拿到任务时 queued → leased,写入 lease_owner / lease_expires_at / lease_count
- 顺带刷新 worker 心跳(last_seen_at),这就是 lease 兼任心跳的设计
"""
effective_wait = max(0, min(wait, max_wait))
deadline = utcnow() + timedelta(seconds=effective_wait)
async with self._lock:
# 先把过期租约扫到 stale,避免「卡死的任务」堵住新任务
await self._sweep_locked()
await self._db.upsert_worker(worker_id, to_iso(utcnow()))
while True:
result = await self._try_pick_locked(worker_id, site)
if result is not None:
return result
remaining = (deadline - utcnow()).total_seconds()
if remaining <= 0:
return None
try:
await asyncio.wait_for(self._cond.wait(), timeout=remaining)
except asyncio.TimeoutError:
return None
async def _try_pick_locked(self, worker_id: str, site: str | None) -> LeaseData | None:
"""锁内尝试领取一次。返回 None 表示此刻无可领任务"""
# 全局并发度 1:已有 leased/running 立即返回空
active = await self._db.list_tasks_in_statuses(tuple(s.value for s in ACTIVE_STATUSES))
if active:
return None
task = await self._db.pick_queued_task(site)
if task is None:
return None
return await self._lease_to_locked(task, worker_id, known_state=None)
async def _lease_to_locked(
self,
task: TaskRow,
worker_id: str,
*,
known_state: str | None,
) -> LeaseData:
"""把一个 queued 或 stale 任务租给 worker。调用方必须持有 self._lock"""
expires = to_iso(utcnow() + timedelta(seconds=self._lease_ttl))
new_count = task.lease_count + 1
await self._db.update_task(
task.task_id,
status=TaskStatus.LEASED.value,
lease_owner=worker_id,
lease_expires_at=expires,
lease_count=new_count,
updated_at=to_iso(utcnow()),
)
# 已知状态:取最近一次 report 的 state(首次领取为 None)
if known_state is None:
known_state = await self._db.get_latest_state(task.task_id)
return LeaseData(
task_id=task.task_id,
site=task.site,
intent=task.intent,
lease_expires_at=expires,
lease_count=new_count,
known_state=known_state, # type: ignore[arg-type]
)
# ---- 续租 ----
async def renew(self, task_id: str, worker_id: str) -> RenewData:
async with self._lock:
task = await self._db.get_task(task_id)
if task is None:
raise TaskNotFoundError(task_id)
if task.lease_owner != worker_id:
raise LeaseInvalidError(f"worker {worker_id} 不是任务 {task_id} 的租约持有者")
if TaskStatus(task.status) in TERMINAL_STATUSES:
raise LeaseInvalidError(f"任务 {task_id} 已终结,无法续租")
if TaskStatus(task.status) not in ACTIVE_STATUSES:
raise LeaseInvalidError(f"任务 {task_id} 状态 {task.status},无可续租约")
expires = to_iso(utcnow() + timedelta(seconds=self._lease_ttl))
await self._db.update_task(
task_id,
lease_expires_at=expires,
updated_at=to_iso(utcnow()),
)
return RenewData(task_id=task_id, lease_expires_at=expires, lease_count=task.lease_count)
# ---- 回报 ----
async def report(
self,
task_id: str,
*,
worker_id: str,
state: str,
payable_yen: int | None,
pay_deadline: str | None,
site_order_id: str | None,
evidence_ref: str | None,
detail: str,
terminal: bool,
terminal_status: str | None,
) -> ReportData:
async with self._lock:
task = await self._db.get_task(task_id)
if task is None:
raise TaskNotFoundError(task_id)
if task.lease_owner != worker_id:
raise LeaseInvalidError(f"worker {worker_id} 不是任务 {task_id} 的租约持有者")
status = TaskStatus(task.status)
if status in TERMINAL_STATUSES:
# 任务已终结:仍然追加 report(规格 §6 末尾「任务已 terminal 的仍可上报,
# gateway 追加到 task_reports」),但不允许改任务状态
if terminal and terminal_status is not None and terminal_status != status.value:
raise InvalidTaskStateError(
f"任务 {task_id} 已是终态 {status.value},不能改为 {terminal_status}"
)
elif status == TaskStatus.LEASED:
# 首次 report:leased → running
await self._db.update_task(
task_id, status=TaskStatus.RUNNING.value, updated_at=to_iso(utcnow())
)
now = to_iso(utcnow())
inserted = await self._db.upsert_report(
ReportRow(
task_id=task_id,
state=state,
payable_yen=payable_yen,
pay_deadline=pay_deadline,
site_order_id=site_order_id,
evidence_ref=evidence_ref,
detail=detail,
reported_at=now,
)
)
final_status = status.value
if terminal:
final_status = self._resolve_terminal_status(
current=status, explicit=terminal_status, state=state
)
await self._db.update_task(
task_id,
status=final_status,
clear_lease=True,
updated_at=now,
)
# 任务终结可能让并发度 1 的闸门放开,唤醒等着的 lease
self._cond.notify_all()
return ReportData(task_id=task_id, status=TaskStatus(final_status), recorded=inserted)
@staticmethod
def _resolve_terminal_status(
*, current: TaskStatus, explicit: str | None, state: str
) -> str:
"""terminal=true 时决定任务的终态。
显式指定优先;否则按订单 state 推断:paid → succeeded,cancelled → failed,
其余(如 awaiting_payment 永久搁置)→ needs_human。
注意 cancelled 既可能来自明确失败,也可能是站点侧自动取消(如付款期限过期),
统一记 failed 让上游看到需要处理。
"""
if explicit is not None:
if explicit not in {s.value for s in TERMINAL_STATUSES}:
raise InvalidTaskStateError(f"terminal_status={explicit} 不是终态")
return explicit
if state == "paid":
return TaskStatus.SUCCEEDED.value
if state == "cancelled":
return TaskStatus.FAILED.value
# awaiting_payment 等无明确结局的状态被标 terminal 时,多半是 worker 检测到
# 3DS 之类人工环节,按 needs_human 处理
return TaskStatus.NEEDS_HUMAN.value
# ---- 恢复(绝不自动重投,见规格 §5)----
async def reclaim(self, task_id: str, worker_id: str) -> ReclaimData:
"""把 stale 任务重新租给 worker。lease_count 必然 > 1"""
async with self._lock:
task = await self._db.get_task(task_id)
if task is None:
raise TaskNotFoundError(task_id)
if TaskStatus(task.status) not in RECLAIMABLE_STATUSES:
raise InvalidTaskStateError(
f"任务 {task_id} 状态 {task.status},不能 reclaim(只能 reclaim stale 任务)"
)
known_state = await self._db.get_latest_state(task_id)
lease = await self._lease_to_locked(task, worker_id, known_state=known_state)
# 复用 LeaseData 的字段构造 ReclaimData(结构一致)
return ReclaimData(
task_id=lease.task_id,
site=lease.site,
intent=lease.intent,
lease_expires_at=lease.lease_expires_at,
lease_count=lease.lease_count,
known_state=lease.known_state,
)
# ---- 过期清扫 ----
async def sweep(self) -> int:
"""把过期的 leased/running 任务置 stale。返回清扫条数"""
async with self._lock:
return await self._sweep_locked()
async def _sweep_locked(self) -> int:
"""锁内执行 sweep。调用方必须持有 self._lock"""
active = await self._db.list_tasks_in_statuses(tuple(s.value for s in ACTIVE_STATUSES))
now = utcnow()
swept = 0
for task in active:
if task.lease_expires_at and parse_iso(task.lease_expires_at) < now:
await self._db.update_task(
task.task_id,
status=TaskStatus.STALE.value,
clear_lease=True,
updated_at=to_iso(now),
)
swept += 1
logger.warning(
"租约过期,任务置 stale(不自动重投):task_id=%s site=%s",
task.task_id,
task.site,
)
if swept:
# stale 后腾出了「全局并发度 1」的槽位,唤醒等待者
# (虽然 stale 任务不会被普通 lease 取到,但队列里可能还有 queued)
self._cond.notify_all()
return swept
# ---- 查询 ----
async def get_task_detail(self, task_id: str) -> TaskDetail:
task = await self._db.get_task(task_id)
if task is None:
raise TaskNotFoundError(task_id)
reports = await self._db.get_reports(task_id)
latest = reports[-1].state if reports else None
return _task_to_detail(
task,
latest_state=latest,
reports=[
ReportEntry(
state=r.state,
payable_yen=r.payable_yen,
pay_deadline=r.pay_deadline,
site_order_id=r.site_order_id,
evidence_ref=r.evidence_ref,
detail=r.detail,
reported_at=r.reported_at,
)
for r in reports
],
)
async def list_tasks(
self,
*,
status: str | None,
site: str | None,
limit: int,
offset: int,
) -> TaskListData:
rows, total = await self._db.list_tasks(
status=status, site=site, limit=limit, offset=offset
)
items: list[TaskDetail] = []
for row in rows:
reports = await self._db.get_reports(row.task_id)
latest = reports[-1].state if reports else None
items.append(
_task_to_detail(
row,
latest_state=latest,
reports=[
ReportEntry(
state=r.state,
payable_yen=r.payable_yen,
pay_deadline=r.pay_deadline,
site_order_id=r.site_order_id,
evidence_ref=r.evidence_ref,
detail=r.detail,
reported_at=r.reported_at,
)
for r in reports
],
)
)
return TaskListData(items=items, total=total, limit=limit, offset=offset)
# ---- 健康检查 ----
async def health_snapshot(self) -> GatewayHealthData:
"""规格 §4.7:只做 worker 失联 + 任务长时间无人领两条兜底告警"""
async with self._lock:
queued_count = await self._db.count_by_status(TaskStatus.QUEUED.value)
active_rows = await self._db.list_tasks_in_statuses(
tuple(s.value for s in ACTIVE_STATUSES)
)
queued_rows = await self._db.list_tasks_in_statuses(
(TaskStatus.QUEUED.value,)
)
workers = await self._db.list_workers()
now = utcnow()
offline_workers: list[WorkerHealthEntry] = []
worker_entries: list[WorkerHealthEntry] = []
for w in workers:
last_seen = parse_iso(w.last_seen_at)
age = int((now - last_seen).total_seconds())
entry = WorkerHealthEntry(
worker_id=w.worker_id,
last_seen_at=w.last_seen_at,
last_seen_seconds=age,
)
worker_entries.append(entry)
if age > self._worker_offline_alert_seconds:
offline_workers.append(entry)
stale_queued_threshold = utcnow() - timedelta(seconds=self._worker_offline_alert_seconds)
stale_queued: list[QueuedAlertEntry] = []
for t in queued_rows:
created = parse_iso(t.created_at)
if created < stale_queued_threshold:
stale_queued.append(
QueuedAlertEntry(
task_id=t.task_id,
site=t.site,
created_at=t.created_at,
age_seconds=int((now - created).total_seconds()),
)
)
active_details = [
_task_to_detail(
t,
latest_state=await self._db.get_latest_state(t.task_id),
reports=[],
)
for t in active_rows
]
degraded = bool(offline_workers or stale_queued)
return GatewayHealthData(
status="degraded" if degraded else "ok",
queued_count=queued_count,
active_tasks=active_details,
workers=worker_entries,
offline_workers=offline_workers,
stale_queued_tasks=stale_queued,
)
+72 -3
View File
@@ -3,14 +3,16 @@
配置项统一使用 RAKUTEN_ 前缀,例如 RAKUTEN_APP_PORT=31107。
支持 .env 文件自动加载。
抓取服务与交易服务是两个进程,但共用这一个 Settings 类:两边都要日志、代理、
超时与同一个 Bearer token,拆成份配置只会让部署时多维护一套。下面按
「通用 / 仅抓取 / 仅交易」分区标注,各进程只读自己那部分。
抓取、交易与下单任务网关是三个进程,但共用这一个 Settings 类:三方都要日志、
代理、超时与同一个 Bearer token,拆成份配置只会让部署时多维护一套。下面按
「通用 / 仅抓取 / 仅交易 / 仅网关 / 仅交易 worker」分区标注,各进程只读自己那部分。
"""
import socket
from functools import lru_cache
from pathlib import Path
from typing import Literal
from pydantic import Field
from pydantic_settings import BaseSettings, SettingsConfigDict
BASE_DIR = Path(__file__).resolve().parent.parent.parent
@@ -47,6 +49,11 @@ class Settings(BaseSettings):
trading_host: str = "0.0.0.0"
trading_port: int = 31108
# 下单任务网关监听地址(第三个部署单元)。网关部署在服务器侧,本地 worker
# 通过出站长轮询从这里领任务。详见 docs/order-gateway.md。
gateway_host: str = "0.0.0.0"
gateway_port: int = 31109
# ---- 日志配置 ----
log_level: str = "INFO"
log_to_file: bool | None = None # None 表示根据环境自动决定
@@ -102,6 +109,36 @@ class Settings(BaseSettings):
# 页面改版导致买到远超预期的订单。设为 0 表示不设上限(不建议)。
order_max_total_yen: int = 30000
# ---- 下单任务网关(仅网关进程 app.gateway.main 使用)----
# 网关的 SQLite 文件路径(相对项目根目录)。任务队列与状态镜像都在这里,
# 部署时务必放在持久化卷上,丢了等于丢了一批下单任务。
gateway_db_path: str = "data/gateway.db"
# 任务租约 TTL(秒)。worker 领取后必须在此时间内首次 report 或 renew,
# 否则网关把任务标记为 stale。**绝不自动重投**(见 docs/order-gateway.md §5)。
lease_ttl_seconds: int = 300
# 长轮询单次最长挂起秒数。worker 端的 wait 参数会被夹到这个上限。
lease_max_wait_seconds: int = Field(default=60, ge=1, le=300)
# worker 心跳超时阈值(秒)。网关 /health 据此判断 worker 是否失联:
# 正常 worker 每 30 秒来一次 lease,超过该阈值未来 lease 即视为异常。
worker_offline_alert_seconds: int = 300
# ---- 本地下单 worker(仅交易服务内的 worker 子模块使用)----
# 网关 URL。**留空则不启动 worker**,交易服务只跑登录态接口。
# 部署形态:本地机(NAT 后无公网入口)通过出站长轮询领任务,详见
# docs/order-gateway.md。
order_gateway_url: str = ""
# worker 标识。同一时间只能有一个 worker 持有 lease,这个值用来区分不同
# 本地机;留空时取主机名。
worker_id: str | None = None
# 本地订单 SQLite 文件路径。订单主表 + 状态事件 + 证据索引都在这里,
# 是执行事实的权威记录;网关上只是镜像。
trading_db_path: str = "data/trading.db"
# 页面证据落盘目录(HTML 快照 + 截图 + meta.json),按 task_id 分子目录。
evidence_dir: str = "data/evidence"
# 抓取服务基地址。worker 需要商品数据(加购要用 purchase 块)时出站请求这里,
# 不直接 import 解析器——抓取与交易是两个进程,详见 README「两个部署单元」。
scraper_base_url: str = ""
# ---- 目标站点 ----
home_url: str = DEFAULT_HOME_URL
@@ -155,6 +192,38 @@ class Settings(BaseSettings):
path.mkdir(parents=True, exist_ok=True)
return path
@property
def gateway_db_path_resolved(self) -> Path:
"""网关 SQLite 文件的绝对路径,父目录不存在时创建"""
path = Path(self.gateway_db_path)
if not path.is_absolute():
path = BASE_DIR / path
path.parent.mkdir(parents=True, exist_ok=True)
return path
@property
def trading_db_path_resolved(self) -> Path:
"""本地订单 SQLite 文件的绝对路径,父目录不存在时创建"""
path = Path(self.trading_db_path)
if not path.is_absolute():
path = BASE_DIR / path
path.parent.mkdir(parents=True, exist_ok=True)
return path
@property
def evidence_path(self) -> Path:
"""证据目录的绝对路径,不存在时创建"""
path = Path(self.evidence_dir)
if not path.is_absolute():
path = BASE_DIR / path
path.mkdir(parents=True, exist_ok=True)
return path
@property
def worker_id_effective(self) -> str:
"""worker 标识:显式配置优先,否则取主机名"""
return self.worker_id or socket.gethostname()
@lru_cache(maxsize=1)
def get_settings() -> Settings:
+55
View File
@@ -6,6 +6,7 @@
- 3xxx: 反爬/上游阻断相关错误
- 4xxx: 页面解析错误
- 5xxx: 加购/下单错误(需要账号登录态的写操作)
- 6xxx: 下单任务编排错误(网关侧的租约与状态机)
"""
@@ -169,3 +170,57 @@ class OrderGuardError(AppError):
def __init__(self, message: str):
super().__init__(message=message, code="ORDER_GUARD", err_code=5004, retryable=False)
# ---- 下单任务编排(仅网关进程 app.gateway.main 使用)----
# 下单不可逆,因此任务队列侧的错误一律标记 retryable=False——重复入队/重投
# 都可能变成重复下单。详见 docs/order-gateway.md §8。
class TaskNotFoundError(AppError):
"""任务不存在"""
def __init__(self, task_id: str):
super().__init__(
message=f"任务不存在:{task_id}",
code="TASK_NOT_FOUND",
err_code=6001,
retryable=False,
status_code=404,
)
self.task_id = task_id
class LeaseInvalidError(AppError):
"""租约无效:不是持有者、已过期或任务已终结
worker 在 renew / report / reclaim 时必须校验自己是当前租约的持有者,且任务
尚未终结。任意一条不满足都报 6002,让 worker 停下来而不是猜测当前状态。
"""
def __init__(self, message: str = "租约无效"):
super().__init__(
message=message,
code="LEASE_INVALID",
err_code=6002,
retryable=False,
status_code=409,
)
class InvalidTaskStateError(AppError):
"""任务状态不允许该操作
例如对已终结(succeeded/failed/needs_human)的任务再 reclaim。与 6002 的区别:
6002 是租约本身的问题(不是持有者 / 已过期),6003 是状态机层面的问题
(当前状态不接受这个动作)。
"""
def __init__(self, message: str = "任务状态不允许该操作"):
super().__init__(
message=message,
code="INVALID_TASK_STATE",
err_code=6003,
retryable=False,
status_code=409,
)
+65
View File
@@ -0,0 +1,65 @@
"""任务与订单状态枚举:网关与本地 worker 共享的词汇表
放在 shared 层是因为这是网关与 worker 之间的 HTTP 契约——两侧都需要知道合法
取值,不能由任何一方私有持有。shared 本身不引入对 scraping/trading 的依赖,
这个模块也一样:纯枚举与状态集合,不引用任何业务模块。
两层状态不要混:task.status 描述「这个任务被谁领了、做完没有」,order state
描述「这笔订单在站点上走到哪一步」。前者权威方是网关,后者权威方是本地 trading,
网关只存镜像。详见 docs/order-gateway.md §3。
迁移图(task.status):
queued ──lease──> leased ──首次 report──> running ──terminal report──> succeeded
▲ │ │ └─> failed
│ └──── 租约过期 ──────────┴──> stale └─> needs_human
└── 只有人工介入才能从 stale 回到 leased(reclaim)
注意 stale **不**自动回 queued——下单不可逆,自动重投等于再买一次。
"""
from __future__ import annotations
from enum import StrEnum
class TaskStatus(StrEnum):
"""任务状态(网关权威)"""
QUEUED = "queued" # 已入队,等待 worker 领取
LEASED = "leased" # 已被 worker 领走,尚未首次 report
RUNNING = "running" # worker 已首次 report,正在进行
SUCCEEDED = "succeeded" # 终态:成功
FAILED = "failed" # 终态:worker 明确失败
NEEDS_HUMAN = "needs_human" # 终态:需要人介入(如 3DS、核对不出结论)
STALE = "stale" # 租约过期,等人工 reclaim,不自动重投
# 终态集合:到达后任何 renew / report / reclaim 都报 6003
TERMINAL_STATUSES: frozenset[TaskStatus] = frozenset(
{TaskStatus.SUCCEEDED, TaskStatus.FAILED, TaskStatus.NEEDS_HUMAN}
)
# 处于「执行中」的状态:这些状态存在时,lease 一律返回空(全局并发度 1)
ACTIVE_STATUSES: frozenset[TaskStatus] = frozenset({TaskStatus.LEASED, TaskStatus.RUNNING})
# 可以被普通 lease 领走的状态:只有 queued
LEASABLE_STATUSES: frozenset[TaskStatus] = frozenset({TaskStatus.QUEUED})
# 可以被 reclaim 领走的状态:只有 stale
RECLAIMABLE_STATUSES: frozenset[TaskStatus] = frozenset({TaskStatus.STALE})
class OrderState(StrEnum):
"""订单状态(本地权威,网关只存镜像)
沿用 docs/order-gateway.md §3 给的取值。网关本身不解释这些值,只是字符串透传。
"""
CREATED = "created"
IN_CART = "in_cart"
ORDERED = "ordered"
AWAITING_PAYMENT = "awaiting_payment"
PAID = "paid"
SHIPPED = "shipped"
DELIVERED = "delivered"
CANCELLED = "cancelled"
+17 -3
View File
@@ -1,9 +1,15 @@
"""交易服务容器:集中管理交易侧服务实例,用于依赖注入
目前只有登录态会话。加购、下单、付款、订单监控与页面证据留痕的服务实例
后续挂在这里,它们共享同一份 auth_session——同一个账号的写操作必须同一
cookie 通道,且必须串行,不能各建各的客户端。
持有两类东西:
- 登录态会话(auth_session——加购与下单必须同一份 cookie,全程共享
- 本地下单 worker(可选)——RAKUTEN_ORDER_GATEWAY_URL 留空时整套不构造
worker 的子组件(gateway_client / local_db / evidence_store / runner)都在
app.trading.worker 包内,与抓取/网关互不 import;worker 需要商品数据时走抓取
服务的 HTTP 接口,需要任务调度时走网关的 HTTP 接口。
"""
from __future__ import annotations
from dataclasses import dataclass
from app.shared.config import Settings
@@ -17,7 +23,15 @@ class TradingContainer:
与抓取容器最本质的差别不是字段多少,而是**这个进程有状态**:登录态、
订单、页面证据都属于某个具体账号,因此交易服务只能单实例运行
(或按账号分片),不能像抓取服务那样随意横向扩容。
worker_* 字段在 RAKUTEN_ORDER_GATEWAY_URL 留空时为 None——此时交易服务
只跑登录态接口,不起后台 worker。
"""
settings: Settings
auth_session: AuthSession
# 以下四项仅在 order_gateway_url 配置时构造;否则为 None,worker 不启动
worker_client: object | None = None # app.trading.worker.client.GatewayClient
worker_local_db: object | None = None # app.trading.worker.local_db.LocalDB
worker_evidence: object | None = None # app.trading.worker.evidence.EvidenceStore
worker_runner: object | None = None # app.trading.worker.runner.WorkerRunner
+65 -4
View File
@@ -10,10 +10,12 @@
因此本服务只能单实例运行(或按账号分片),扩容靠抓取服务那一侧。
当前只提供登录态的查询与重载;加购、下单、付款与订单监控在此基础上叠加。
worker 启动条件:RAKUTEN_ORDER_GATEWAY_URL 非空。留空时本服务只跑登录态接口,
不构造任何 worker 子组件(本地 DB / 证据目录 / 出站客户端都不创建)。
"""
from __future__ import annotations
import asyncio
import logging
from contextlib import asynccontextmanager
@@ -32,14 +34,41 @@ logger = logging.getLogger(__name__)
def build_container() -> TradingContainer:
"""构建交易服务容器,组装所有依赖"""
"""构建交易服务容器:登录态会话始终构造,worker 仅在配置了网关 URL 时构造"""
settings = get_settings()
return TradingContainer(settings=settings, auth_session=AuthSession(settings))
container = TradingContainer(settings=settings, auth_session=AuthSession(settings))
if settings.order_gateway_url:
# 延迟 import:未配置网关 URL 时不加载 worker 模块(也就不会拉起 aiosqlite 等)
from app.trading.worker.client import GatewayClient
from app.trading.worker.evidence import EvidenceStore
from app.trading.worker.local_db import LocalDB
from app.trading.worker.runner import WorkerRunner
client = GatewayClient(
settings.order_gateway_url,
settings.bearer_token,
timeout=max(60.0, settings.lease_max_wait_seconds + 10),
)
local_db = LocalDB(settings.trading_db_path_resolved)
evidence = EvidenceStore(settings.evidence_path)
runner = WorkerRunner(
settings=settings,
gateway_client=client,
local_db=local_db,
evidence=evidence,
)
container.worker_client = client
container.worker_local_db = local_db
container.worker_evidence = evidence
container.worker_runner = runner
return container
@asynccontextmanager
async def lifespan(app: FastAPI):
"""应用生命周期管理:启动时加载登录态,关闭时释放 HTTP 客户端"""
"""应用生命周期管理:登录态会话 + 可选 worker"""
container = build_container()
app.state.container = container
@@ -52,10 +81,42 @@ async def lifespan(app: FastAPI):
container.settings.trading_port,
)
logger.info("当前环境:%s", container.settings.app_env)
await container.auth_session.start()
worker_task: asyncio.Task | None = None
if container.worker_runner is not None:
# 顺序:先开本地 DB(worker 写证据前要先能写库),再起后台任务
assert container.worker_local_db is not None
assert container.worker_client is not None
await container.worker_local_db.start()
worker_task = asyncio.create_task(
container.worker_runner.run(), name="trading-worker"
)
logger.info(
"下单 worker 已启动:worker_id=%s gateway=%s",
container.settings.worker_id_effective,
container.settings.order_gateway_url,
)
else:
logger.info(
"未配置 RAKUTEN_ORDER_GATEWAY_URL,下单 worker 不启动(仅登录态接口)"
)
try:
yield
finally:
if worker_task is not None:
container.worker_runner.stop() # type: ignore[union-attr]
worker_task.cancel()
try:
await worker_task
except asyncio.CancelledError:
pass
if container.worker_local_db is not None:
await container.worker_local_db.close()
if container.worker_client is not None:
await container.worker_client.aclose()
await container.auth_session.close()
shutdown_telemetry()
+9
View File
@@ -0,0 +1,9 @@
"""本地下单 worker:从网关长轮询领任务,本地执行下单与付款
部署在本地交易服务进程内NAT 后无公网入口通过出站长轮询从服务器侧的
order-gateway 取任务**零入站端口**所有跨机通信都由本地发起
主循环见 runner.WorkerRunner.runlease 本地幂等闸门 恢复核对 执行 监控
站点交互加购 / 下单 / 付款当前未实测docs/order-gateway.md §10留作
NotImplementedError 缺口runner 捕获后上报 needs_human
"""
+145
View File
@@ -0,0 +1,145 @@
"""网关 HTTP 客户端:出站长轮询领任务、回报状态、续租、恢复领取
worker 与网关之间是出站单向通信所有方法都包装成成功 data失败
AppError的形式runner 拿到 AppError 直接打日志或转报 needs_human
请求与响应严格使用 shared.api ApiResponse 信封网关侧错误码6xxx原样上抛
不在这里翻译鉴权头从 settings.bearer_token 与抓取/交易服务共用同一份
** import app.gateway**worker 只看 HTTP 响应 JSON本地用 LeaseTask 表达领到的
任务与网关侧的 Pydantic 模型解耦
"""
from __future__ import annotations
import logging
from typing import Any
import httpx
from app.shared.errors import AppError
from app.shared.task_state import OrderState, TaskStatus
from app.trading.worker.models import LeaseTask
logger = logging.getLogger(__name__)
class GatewayClient:
"""网关 HTTP 客户端
一份 AsyncClient 实例贯穿 worker 整个生命周期连接池由 httpx 管理
"""
def __init__(self, base_url: str, bearer_token: str, *, timeout: float = 60.0):
# 末尾去斜杠,避免 base + "/api/..." 拼出双斜杠
self._base_url = base_url.rstrip("/")
self._client = httpx.AsyncClient(
base_url=self._base_url,
headers={"Authorization": f"Bearer {bearer_token}"},
timeout=timeout,
)
async def aclose(self) -> None:
await self._client.aclose()
# ---- 基础封装 ----
async def _request(self, method: str, path: str, **kwargs: Any) -> dict[str, Any]:
"""发起请求并解信封。失败(success=False)抛 AppError"""
response = await self._client.request(method, path, **kwargs)
try:
body = response.json()
except ValueError as exc:
raise AppError(
message=f"网关响应不是合法 JSON:HTTP {response.status_code}",
code="GATEWAY_BAD_BODY",
err_code=3001,
retryable=True,
) from exc
if not body.get("success"):
raise AppError(
message=body.get("msg", "网关返回失败"),
code="GATEWAY_ERROR",
err_code=int(body.get("code", 1500)),
retryable=False,
status_code=response.status_code,
)
return body
# ---- 接口 ----
async def lease(
self, worker_id: str, *, wait: int = 30, site: str | None = None
) -> LeaseTask | None:
"""长轮询领取。无可领任务时返回 None"""
params: dict[str, Any] = {"worker_id": worker_id, "wait": wait}
if site:
params["site"] = site
body = await self._request("GET", "/api/orders/lease", params=params)
data = body.get("data")
if not data:
return None
return LeaseTask(
task_id=data["task_id"],
site=data["site"],
intent=data.get("intent") or {},
lease_expires_at=data.get("lease_expires_at", ""),
lease_count=data.get("lease_count", 1),
known_state=data.get("known_state"),
)
async def renew(self, task_id: str, worker_id: str) -> dict[str, Any]:
"""续租,返回网关响应里的 data 字段"""
body = await self._request(
"POST",
f"/api/orders/{task_id}/renew",
json={"worker_id": worker_id},
)
return body["data"]
async def report(
self,
task_id: str,
worker_id: str,
*,
state: OrderState,
payable_yen: int | None = None,
pay_deadline: str | None = None,
site_order_id: str | None = None,
evidence_ref: str | None = None,
detail: str = "",
terminal: bool = False,
terminal_status: TaskStatus | None = None,
) -> dict[str, Any]:
"""回报状态,返回网关响应里的 data 字段"""
payload: dict[str, Any] = {
"worker_id": worker_id,
"state": state.value,
"payable_yen": payable_yen,
"pay_deadline": pay_deadline,
"site_order_id": site_order_id,
"evidence_ref": evidence_ref,
"detail": detail,
"terminal": terminal,
}
if terminal_status is not None:
payload["terminal_status"] = terminal_status.value
body = await self._request("POST", f"/api/orders/{task_id}/report", json=payload)
return body["data"]
async def reclaim(self, task_id: str, worker_id: str) -> LeaseTask:
"""恢复领取。返回的 LeaseTask 必然 lease_count > 1"""
body = await self._request(
"POST",
f"/api/orders/{task_id}/reclaim",
json={"worker_id": worker_id},
)
data = body["data"]
return LeaseTask(
task_id=data["task_id"],
site=data["site"],
intent=data.get("intent") or {},
lease_expires_at=data.get("lease_expires_at", ""),
lease_count=data.get("lease_count", 1),
known_state=data.get("known_state"),
)
+56
View File
@@ -0,0 +1,56 @@
"""证据落盘:HTML 快照、截图与 meta.json
规格 §6.1每个任务一个子目录按步骤序号命名文件report 里的 evidence_ref
`{task_id}/{序号}-{步骤名}`不含扩展名
证据**只留本地**约束 4网关只存 evidence_ref 这个相对路径字符串
"""
from __future__ import annotations
import json
import logging
from pathlib import Path
from typing import Any
logger = logging.getLogger(__name__)
class EvidenceStore:
"""把每一步的页面快照、截图与 meta 落到 {evidence_dir}/{task_id}/ 下"""
def __init__(self, evidence_dir: Path):
self._root = evidence_dir
def step_dir(self, task_id: str) -> Path:
path = self._root / task_id
path.mkdir(parents=True, exist_ok=True)
return path
def write_step(
self,
task_id: str,
step_no: int,
step_name: str,
*,
html: str | None = None,
png: bytes | None = None,
meta: dict[str, Any] | None = None,
) -> str:
"""写入一步的证据文件
返回相对 evidence_dir 的引用路径不含扩展名用于 report evidence_ref
字段文件命名`{序号:02d}-{步骤名}.{html|png|meta.json}`
"""
step_dir = self.step_dir(task_id)
stem = f"{step_no:02d}-{step_name}"
rel_stem = f"{task_id}/{stem}"
if html is not None:
(step_dir / f"{stem}.html").write_text(html, encoding="utf-8")
if png is not None:
(step_dir / f"{stem}.png").write_bytes(png)
if meta is not None:
(step_dir / f"{stem}.meta.json").write_text(
json.dumps(meta, ensure_ascii=False, indent=2), encoding="utf-8"
)
return rel_stem
+181
View File
@@ -0,0 +1,181 @@
"""本地订单 SQLite:执行事实的权威记录
与网关的 tasks/task_reports 表对偶但本地这张表才是这单到底下没下付没付
的权威网关失联或数据丢失时本地这张表用于补报与对账
三张表
- orders每笔任务一行记录开始/结束时间与最终状态
- order_eventsappend-only 的状态迁移事件与网关 task_reports 对账
- evidence_index每步证据文件的相对路径索引便于事后翻查
"""
from __future__ import annotations
import json
import logging
from dataclasses import dataclass
from datetime import datetime, timezone
from pathlib import Path
from typing import Any
import aiosqlite
logger = logging.getLogger(__name__)
SCHEMA = """
CREATE TABLE IF NOT EXISTS orders (
task_id TEXT PRIMARY KEY,
site TEXT NOT NULL,
intent_json TEXT NOT NULL,
final_state TEXT,
started_at TEXT NOT NULL,
finished_at TEXT
);
CREATE TABLE IF NOT EXISTS order_events (
task_id TEXT NOT NULL,
state TEXT NOT NULL,
detail TEXT,
evidence_ref TEXT,
recorded_at TEXT NOT NULL,
PRIMARY KEY (task_id, state)
);
CREATE TABLE IF NOT EXISTS evidence_index (
task_id TEXT NOT NULL,
step_no INTEGER NOT NULL,
step_name TEXT NOT NULL,
rel_path TEXT NOT NULL,
saved_at TEXT NOT NULL,
PRIMARY KEY (task_id, step_no)
);
"""
def _utcnow_iso() -> str:
return datetime.now(timezone.utc).replace(microsecond=0).isoformat().replace("+00:00", "Z")
@dataclass(slots=True)
class OrderRow:
task_id: str
site: str
intent_json: str
final_state: str | None
started_at: str
finished_at: str | None
@property
def intent(self) -> dict[str, Any]:
return json.loads(self.intent_json)
class LocalDB:
"""本地订单 SQLite 访问对象"""
def __init__(self, db_path: Path):
self._db_path = db_path
self._conn: aiosqlite.Connection | None = None
async def start(self) -> None:
self._db_path.parent.mkdir(parents=True, exist_ok=True)
self._conn = await aiosqlite.connect(str(self._db_path))
self._conn.row_factory = aiosqlite.Row
await self._conn.executescript(SCHEMA)
await self._conn.commit()
logger.info("本地订单 DB 已就绪:%s", self._db_path)
async def close(self) -> None:
if self._conn is not None:
await self._conn.close()
self._conn = None
@property
def conn(self) -> aiosqlite.Connection:
if self._conn is None:
raise RuntimeError("LocalDB 未启动:先调用 start()")
return self._conn
# ---- orders ----
async def ensure_started(self, task_id: str, site: str, intent: dict[str, Any]) -> None:
"""记录任务开始执行。已存在则忽略(同任务被重新领回时)"""
await self.conn.execute(
"INSERT OR IGNORE INTO orders (task_id, site, intent_json, started_at) "
"VALUES (?, ?, ?, ?)",
(task_id, site, json.dumps(intent, ensure_ascii=False), _utcnow_iso()),
)
await self.conn.commit()
async def get_order(self, task_id: str) -> OrderRow | None:
async with self.conn.execute(
"SELECT * FROM orders WHERE task_id = ?",
(task_id,),
) as cur:
row = await cur.fetchone()
if row is None:
return None
return OrderRow(
task_id=row["task_id"],
site=row["site"],
intent_json=row["intent_json"],
final_state=row["final_state"],
started_at=row["started_at"],
finished_at=row["finished_at"],
)
async def has_finished(self, task_id: str) -> bool:
"""是否已有终态记录。用于主循环里的本地幂等闸门"""
async with self.conn.execute(
"SELECT 1 FROM orders WHERE task_id = ? AND finished_at IS NOT NULL",
(task_id,),
) as cur:
return await cur.fetchone() is not None
async def mark_finished(self, task_id: str, final_state: str) -> None:
await self.conn.execute(
"UPDATE orders SET final_state = ?, finished_at = ? WHERE task_id = ?",
(final_state, _utcnow_iso(), task_id),
)
await self.conn.commit()
async def final_state(self, task_id: str) -> str | None:
async with self.conn.execute(
"SELECT final_state FROM orders WHERE task_id = ?",
(task_id,),
) as cur:
row = await cur.fetchone()
return row["final_state"] if row else None
# ---- events ----
async def record_event(
self,
task_id: str,
state: str,
*,
detail: str = "",
evidence_ref: str | None = None,
) -> None:
"""记录一次状态迁移事件(与网关 task_reports 对账)
同一 (task_id, state) 重复记录视为同一次INSERT OR IGNORE与网关侧的
INSERT OR REPLACE 不同本地这张表是事件日志不覆盖既有记录
"""
await self.conn.execute(
"INSERT OR IGNORE INTO order_events "
"(task_id, state, detail, evidence_ref, recorded_at) VALUES (?, ?, ?, ?, ?)",
(task_id, state, detail, evidence_ref, _utcnow_iso()),
)
await self.conn.commit()
# ---- evidence index ----
async def index_evidence(
self, task_id: str, step_no: int, step_name: str, rel_path: str
) -> None:
await self.conn.execute(
"INSERT OR REPLACE INTO evidence_index "
"(task_id, step_no, step_name, rel_path, saved_at) VALUES (?, ?, ?, ?, ?)",
(task_id, step_no, step_name, rel_path, _utcnow_iso()),
)
await self.conn.commit()
+28
View File
@@ -0,0 +1,28 @@
"""worker 侧数据模型:不引用网关的 Pydantic 模型
worker 通过 HTTP 与网关交互看到的只有 JSON这里定义 worker 自己用来表达
领到的任务核对结论的数据类结构上对应网关响应** import**
app.gateway 的任何东西保证架构边界worker shared禁止 worker gateway
OrderState / TaskStatus 枚举在 app.shared.task_state两侧共用同一份取值表
"""
from __future__ import annotations
from dataclasses import dataclass, field
from typing import Any
@dataclass(slots=True)
class LeaseTask:
"""worker 从网关领到的任务
对应网关 GET /api/orders/lease LeaseData但用 dataclass 表达worker
内部不依赖网关侧的 Pydantic 模型
"""
task_id: str
site: str
intent: dict[str, Any] = field(default_factory=dict)
lease_expires_at: str = ""
lease_count: int = 1
known_state: str | None = None
+420
View File
@@ -0,0 +1,420 @@
"""worker 主循环与每任务执行流
主循环规格 §6
while running:
task = await gateway.lease(worker_id, wait=30)
if task is None:
continue
if local_db.has_finished(task.task_id): # 本地幂等闸门
await gateway.report(...); continue
if task.lease_count > 1: # 恢复领取
verdict = await verify_on_site(task)
if verdict.already_ordered: await report(...); continue
if verdict.unknown: await report(needs_human); continue
async with renew_lease_every(60):
await execute(task)
每个动作的顺序固定为动作 落证据 写本地 SQLite 回报 gateway
顺序不能颠倒先落证据再回报保证服务器上看到的状态一定有本地证据可查
"""
from __future__ import annotations
import asyncio
import contextlib
import logging
from typing import TYPE_CHECKING
from app.shared.errors import AppError, OrderGuardError
from app.shared.task_state import OrderState, TaskStatus
from app.trading.worker import site_interact, verify
from app.trading.worker.client import GatewayClient
from app.trading.worker.evidence import EvidenceStore
from app.trading.worker.local_db import LocalDB
from app.trading.worker.models import LeaseTask
if TYPE_CHECKING:
from app.shared.config import Settings
logger = logging.getLogger(__name__)
def _coerce_state(value: str | None) -> OrderState:
"""把网关返回的 state 字符串安全地包成 OrderState;None 或非法值回落到 CREATED"""
if not value:
return OrderState.CREATED
try:
return OrderState(value)
except ValueError:
logger.warning("未知的 order state 字符串,回落到 CREATED:%s", value)
return OrderState.CREATED
class WorkerRunner:
"""worker 主循环与执行流
持有 gateway 客户端本地 DB证据存储`run()` 是常驻 asyncio 任务
`handle()` 处理单个任务`execute()` 是站点交互的实际调度当前多数步骤未实现
"""
def __init__(
self,
*,
settings: "Settings",
gateway_client: GatewayClient,
local_db: LocalDB,
evidence: EvidenceStore,
):
self._settings = settings
self._gateway = gateway_client
self._db = local_db
self._evidence = evidence
self._running = False
def stop(self) -> None:
self._running = False
@property
def worker_id(self) -> str:
return self._settings.worker_id_effective
# ---- 主循环 ----
async def run(self) -> None:
"""常驻循环:长轮询领任务,按 §6 调度"""
self._running = True
logger.info("worker 启动:worker_id=%s", self.worker_id)
while self._running:
try:
task = await self._gateway.lease(self.worker_id, wait=30)
except AppError as exc:
logger.warning("lease 失败:%s (err=%s)", exc.message, exc.err_code)
await asyncio.sleep(5)
continue
except Exception: # noqa: BLE001
logger.exception("lease 异常")
await asyncio.sleep(5)
continue
if task is None:
continue
try:
await self.handle(task)
except Exception: # noqa: BLE001
# 任何未预期的异常都不应让 worker 退出
logger.exception("handle 任务异常:task_id=%s", task.task_id)
# ---- 单任务调度 ----
async def handle(self, task: LeaseTask) -> None:
"""单任务调度入口:本地幂等闸门 → 恢复核对 → 执行"""
# 本地幂等闸门:之前已完成过的任务不再执行
if await self._db.has_finished(task.task_id):
final_state = await self._db.final_state(task.task_id)
logger.info(
"本地已完成,补报终态:task_id=%s state=%s", task.task_id, final_state
)
await self._report_safe(
task,
state=_coerce_state(final_state),
detail="本地已完成,补报避免重复执行",
terminal=True,
)
return
# 恢复领取:lease_count > 1 表示 stale → reclaim,必须先核对站点订单
if task.lease_count > 1:
await self._handle_recovery(task)
return
# 常规执行
await self._execute_with_renewal(task)
async def _handle_recovery(self, task: LeaseTask) -> None:
"""lease_count > 1 的恢复路径:先核对,不直接执行"""
logger.warning(
"恢复领取(lease_count=%s),先核对站点订单:task_id=%s",
task.lease_count,
task.task_id,
)
verdict = await verify.verify_on_site(task)
if verdict.verdict == verify.VerifyVerdict.ALREADY_ORDERED:
await self._report_safe(
task,
state=OrderState.ORDERED,
site_order_id=verdict.site_order_id,
detail=f"核对结论:已下单({verdict.detail}",
terminal=True,
terminal_status=TaskStatus.SUCCEEDED,
)
await self._db.ensure_started(task.task_id, task.site, task.intent)
await self._db.mark_finished(task.task_id, OrderState.ORDERED.value)
return
# NOT_ORDERED 也走 needs_human:当前 verify 桩不会返回这个值,但留接口给
# 未来真正能可靠核对时——那时再决定 NOT_ORDERED 是否直接重新执行
await self._report_safe(
task,
state=_coerce_state(task.known_state),
detail=f"恢复核对无法定论:{verdict.detail}",
terminal=True,
terminal_status=TaskStatus.NEEDS_HUMAN,
)
# ---- 常规执行 ----
async def _execute_with_renewal(self, task: LeaseTask) -> None:
"""在租约自动续期的上下文里执行任务"""
async with self._renew_lease_every(task, interval=60):
try:
await self.execute(task)
except NotImplementedError as exc:
# 站点交互未实现(规格 §10):上报 needs_human,不视为 worker 失败
logger.warning(
"站点交互未实现,转 needs_human:task_id=%s what=%s",
task.task_id,
exc,
)
await self._report_safe(
task,
state=_coerce_state(task.known_state),
detail=f"站点交互未实现:{exc}",
terminal=True,
terminal_status=TaskStatus.NEEDS_HUMAN,
)
except OrderGuardError as exc:
# 金额守卫等闸门拦截:上报 needs_human,站点侧无提交动作
logger.warning(
"下单闸门拦截,转 needs_human:task_id=%s msg=%s",
task.task_id,
exc.message,
)
await self._report_safe(
task,
state=_coerce_state(task.known_state),
detail=exc.message,
terminal=True,
terminal_status=TaskStatus.NEEDS_HUMAN,
)
except AppError as exc:
logger.warning(
"执行失败:task_id=%s code=%s msg=%s",
task.task_id,
exc.err_code,
exc.message,
)
await self._report_safe(
task,
state=_coerce_state(task.known_state),
detail=exc.message,
terminal=True,
terminal_status=TaskStatus.FAILED,
)
async def execute(self, task: LeaseTask) -> None:
"""站点交互的实际调度:加购 → 校验 → 确认页 → 金额守卫 → 提交 → 付款
每一步的顺序动作 落证据 写本地 SQLite 回报 gateway
站点交互当前未实现site_interact NotImplementedError第一步就会
转到 _execute_with_renewal except 分支上报 needs_human
"""
await self._db.ensure_started(task.task_id, task.site, task.intent)
# ラクマ 加购契约未实现,首版只对乐天。其他站点直接转人工
if task.site != "rakuten":
raise NotImplementedError(f"site={task.site} 暂未实现下单(首版仅 rakuten)")
# 步骤 1:加购
await self._run_step(
task, step_no=1, step_name="cart-add",
action=lambda: site_interact.add_to_cart(task),
state=OrderState.IN_CART,
detail="已加入购物车",
)
# 步骤 2:校验购物车
await self._run_step(
task, step_no=2, step_name="cart-check",
action=lambda: site_interact.verify_cart(task),
state=OrderState.IN_CART,
detail="购物车已校验",
)
# 步骤 3:进入下单确认页 + 金额守卫
checkout_html = await site_interact.enter_checkout(task)
summary = await site_interact.parse_checkout(checkout_html)
self._enforce_amount_guard(task, summary.payable_yen)
await self._run_step(
task, step_no=3, step_name="order-confirm",
action=self._noop(),
state=OrderState.CREATED,
detail=f"下单确认页已解析:应付 {summary.payable_yen}",
evidence_meta={
"payable_yen": summary.payable_yen,
"site_order_id": summary.site_order_id,
"pay_deadline": summary.pay_deadline,
},
)
# 步骤 4:提交下单
site_order_id = await site_interact.submit_order(task)
await self._run_step(
task, step_no=4, step_name="order-submit",
action=self._noop(),
state=OrderState.ORDERED,
site_order_id=site_order_id,
detail=f"已提交下单,站点订单号 {site_order_id}",
)
# 步骤 5:付款
await self._run_step(
task, step_no=5, step_name="payment",
action=lambda: site_interact.pay(task, site_order_id),
state=OrderState.AWAITING_PAYMENT,
site_order_id=site_order_id,
payable_yen=summary.payable_yen,
pay_deadline=summary.pay_deadline,
detail="已进入付款流程",
)
await self._report_safe(
task,
state=OrderState.PAID,
site_order_id=site_order_id,
payable_yen=summary.payable_yen,
pay_deadline=summary.pay_deadline,
detail="付款完成",
terminal=True,
terminal_status=TaskStatus.SUCCEEDED,
)
await self._db.mark_finished(task.task_id, OrderState.PAID.value)
# 步骤 6:付款后监控(非阻塞,常驻轮询;当前未实现)
try:
await site_interact.monitor(task, site_order_id)
except NotImplementedError:
logger.info("付款后监控未实现,跳过:task_id=%s", task.task_id)
def _enforce_amount_guard(self, task: LeaseTask, payable_yen: int) -> None:
"""金额守卫:实际应付超过 intent.max_total_yen 或 RAKUTEN_ORDER_MAX_TOTAL_YEN 时拦截"""
intent = task.intent or {}
limit = intent.get("max_total_yen")
if limit is None or limit <= 0:
limit = self._settings.order_max_total_yen
if limit > 0 and payable_yen > limit:
raise OrderGuardError(
f"实际应付 {payable_yen} 円超过上限 {limit}"
f"(task_id={task.task_id}"
)
@staticmethod
def _noop() -> "callable":
async def _f() -> None:
return None
return _f
async def _run_step(
self,
task: LeaseTask,
*,
step_no: int,
step_name: str,
action: "callable",
state: OrderState,
detail: str,
site_order_id: str | None = None,
payable_yen: int | None = None,
pay_deadline: str | None = None,
evidence_meta: dict | None = None,
) -> None:
"""单步执行:动作 → 落证据 → 写本地 → 回报 gateway"""
await action()
meta = {
"step": step_name,
"state": state.value,
**(evidence_meta or {}),
}
evidence_ref = self._evidence.write_step(
task.task_id, step_no, step_name, meta=meta
)
await self._db.index_evidence(task.task_id, step_no, step_name, evidence_ref)
await self._db.record_event(
task.task_id, state.value, detail=detail, evidence_ref=evidence_ref
)
await self._gateway.report(
task.task_id,
self.worker_id,
state=state,
payable_yen=payable_yen,
pay_deadline=pay_deadline,
site_order_id=site_order_id,
evidence_ref=evidence_ref,
detail=detail,
)
async def _report_safe(
self,
task: LeaseTask,
*,
state: OrderState,
detail: str,
terminal: bool = False,
terminal_status: TaskStatus | None = None,
site_order_id: str | None = None,
payable_yen: int | None = None,
pay_deadline: str | None = None,
) -> None:
"""回报 gateway,失败只记日志不抛——主循环不能因为回报失败退出"""
try:
await self._gateway.report(
task.task_id,
self.worker_id,
state=state,
detail=detail,
terminal=terminal,
terminal_status=terminal_status,
site_order_id=site_order_id,
payable_yen=payable_yen,
pay_deadline=pay_deadline,
)
except Exception: # noqa: BLE001
logger.exception("回报 gateway 失败:task_id=%s state=%s", task.task_id, state)
@contextlib.asynccontextmanager
async def _renew_lease_every(self, task: LeaseTask, *, interval: int):
"""后台任务:每隔 interval 秒续租一次。退出时停掉"""
stop_event = asyncio.Event()
async def _loop() -> None:
while not stop_event.is_set():
try:
await asyncio.wait_for(stop_event.wait(), timeout=interval)
except asyncio.TimeoutError:
pass
if stop_event.is_set():
return
try:
await self._gateway.renew(task.task_id, self.worker_id)
except AppError as exc:
logger.warning(
"续租失败:task_id=%s code=%s msg=%s",
task.task_id,
exc.err_code,
exc.message,
)
return
except Exception: # noqa: BLE001
logger.exception("续租异常:task_id=%s", task.task_id)
return
renewal = asyncio.create_task(_loop(), name=f"renew-{task.task_id}")
try:
yield
finally:
stop_event.set()
with contextlib.suppress(asyncio.CancelledError):
await renewal
+73
View File
@@ -0,0 +1,73 @@
"""站点交互:加购 / 校验购物车 / 进入下单确认页 / 金额守卫 / 提交 / 付款 / 监控
**全部未实现**规格 docs/order-gateway.md §10 明确说明加购 下单 付款的
实际站点交互没有实测过未实测前调用直接抛未实现不要写猜测的提交逻辑
签名刻意保留runner 把这一层当接口缝用等真实账号 + 真实站点把每个动作
实现填进来后主循环逻辑不需要改动
实现方需要补齐的实测点规格 §10
1. 自动付款是否触发 3D Secure / 短信验证触发则这条路走不通付款环节改为
下单到 awaiting_payment + 上报 needs_human 交人工
2. 下单确认页的实际应付金额付款方式付款期限站点订单号各自在哪个字段
3. 订单列表页能否按商品 + 时间窗口可靠地反查这单下没下
4. ラクマ 侧尚无加购契约README 已注明未实现
"""
from __future__ import annotations
from dataclasses import dataclass
from app.trading.worker.models import LeaseTask
_NOT_IMPLEMENTED_MSG = (
"站点交互未实现:见 docs/order-gateway.md §10。"
"未实测前不写猜测的提交逻辑——需要真实账号 + 真实站点把这一步实现填进来。"
)
@dataclass(slots=True)
class CheckoutSummary:
"""下单确认页解析结果(待实测确认字段位置)
payable_yen 用于金额守卫其他字段在实测确认后补全
"""
payable_yen: int
site_order_id: str | None = None
pay_deadline: str | None = None
async def add_to_cart(task: LeaseTask) -> None:
"""加购。需要 auth_session 的登录 cookie + intent 里的 purchase 标识"""
raise NotImplementedError(_NOT_IMPLEMENTED_MSG)
async def verify_cart(task: LeaseTask) -> None:
"""校验购物车:加购后购物车里有没有这件商品"""
raise NotImplementedError(_NOT_IMPLEMENTED_MSG)
async def enter_checkout(task: LeaseTask) -> str:
"""进入下单确认页,返回页面 HTML(供解析金额守卫所需字段)"""
raise NotImplementedError(_NOT_IMPLEMENTED_MSG)
async def parse_checkout(html: str) -> CheckoutSummary:
"""从下单确认页解析实际应付金额、付款期限、站点订单号"""
raise NotImplementedError(_NOT_IMPLEMENTED_MSG)
async def submit_order(task: LeaseTask) -> str:
"""提交下单。返回站点订单号(site_order_id)"""
raise NotImplementedError(_NOT_IMPLEMENTED_MSG)
async def pay(task: LeaseTask, site_order_id: str) -> None:
"""付款。检测到 3DS / 短信验证等人工环节时抛 OrderGuardError"""
raise NotImplementedError(_NOT_IMPLEMENTED_MSG)
async def monitor(task: LeaseTask, site_order_id: str) -> None:
"""付款后监控:常驻轮询订单状态与付款期限,状态变化时继续 report"""
raise NotImplementedError(_NOT_IMPLEMENTED_MSG)
+46
View File
@@ -0,0 +1,46 @@
"""恢复核对:lease_count > 1 时,worker 必须先查站点订单列表确认是否已下单
规格 §5租约过期 stale reclaim worker 领回的任务 lease_count 必然 > 1
此时 worker 不能盲目重新提交可能上次已经下单成功只是回报断网必须先查
站点订单列表比对
订单列表反查的实现需要实测规格 §10 3 订单列表页能否按商品 + 时间窗口
可靠地反查出这单下没下当前为桩恒返回 `unknown` runner
needs_human 分支**绝不**默认按没下单处理那是猜可能变成重复下单
"""
from __future__ import annotations
from dataclasses import dataclass
from enum import StrEnum
from app.trading.worker.models import LeaseTask
class VerifyVerdict(StrEnum):
"""核对结论"""
ALREADY_ORDERED = "already_ordered" # 确认已下单,补报状态即可
NOT_ORDERED = "not_ordered" # 确认未下单,可重新执行
UNKNOWN = "unknown" # 核对不出结论,交人工
@dataclass(slots=True)
class VerifyResult:
verdict: VerifyVerdict
site_order_id: str | None = None
detail: str = ""
async def verify_on_site(task: LeaseTask) -> VerifyResult:
"""核对一笔任务是否已在站点上下过单
桩实现永远返回 UNKNOWN**绝不返回 NOT_ORDERED**除非真实订单列表反查
能可靠证明这一点否则视为无法判断交人工宁可卡住等人看一眼
实现方需要补的实测 intent 里的商品 + 时间窗口created_at stale 之间
比对订单列表能拿到 site_order_id 最好
"""
return VerifyResult(
verdict=VerifyVerdict.UNKNOWN,
detail="订单列表反查未实现:见 docs/order-gateway.md §10 第 3 条",
)
+1
View File
@@ -4,6 +4,7 @@ version = "0.1.0"
description = "API service for Rakuten Ichiba scraping tasks"
requires-python = ">=3.13"
dependencies = [
"aiosqlite>=0.20.0,<1.0.0",
"fastapi>=0.115.0,<1.0.0",
"httpx[http2]>=0.27.0,<1.0.0",
"loguru>=0.7.0,<1.0.0",
+35 -13
View File
@@ -1,12 +1,11 @@
"""架构测试:守住抓取侧与交易侧的依赖方向
"""架构测试:守住三个部署单元的依赖方向
拆成两个进程之后最容易悄悄退化的不是功能而是边界某天为了省事在交易侧
`from app.scraping.parsers...` 一句两个服务就重新长回一起抓取的解析改动会
牵动下单链路交易服务也被迫加载整套抓取依赖包括 Playwright
个进程之后最容易悄悄退化的不是功能而是边界某天为了省事在交易侧
`from app.scraping.parsers...` 一句两个服务就重新长回一起或在 worker
`from app.gateway.models...` 一句本地的零依赖网关假设就破了
允许的方向只有scraping sharedtrading shared
交易侧要用抓取的能力走抓取服务的 HTTP 接口`purchase` 块本来就是它的对外契约
不直接 import
允许的方向只有scraping sharedtrading sharedgateway shared
任何两侧之间互不 import需要对方能力时走 HTTP 接口
"""
from __future__ import annotations
@@ -37,11 +36,17 @@ def _python_files(package: str) -> list[Path]:
@pytest.mark.parametrize(
("package", "forbidden"),
[
# 三个部署单元互不 import
("scraping", "app.trading"),
("scraping", "app.gateway"),
("trading", "app.scraping"),
# shared 是两侧的共同底座,反向依赖任何一侧都会形成环
("trading", "app.gateway"),
("gateway", "app.scraping"),
("gateway", "app.trading"),
# shared 是三方的共同底座,反向依赖任何一侧都会形成环
("shared", "app.scraping"),
("shared", "app.trading"),
("shared", "app.gateway"),
],
)
def test_package_does_not_import(package: str, forbidden: str):
@@ -54,8 +59,9 @@ def test_package_does_not_import(package: str, forbidden: str):
assert not offenders, f"{package} 不应依赖 {forbidden}{offenders}"
def test_both_entrypoints_build():
"""个入口都要能独立创建应用——这是「个部署单元」的最低验收"""
def test_all_three_entrypoints_build():
"""个入口都要能独立创建应用——这是「个部署单元」的最低验收"""
from app.gateway.main import create_app as create_gateway_app
from app.scraping.main import create_app as create_scraping_app
from app.trading.main import create_app as create_trading_app
@@ -63,11 +69,27 @@ def test_both_entrypoints_build():
# 包成 _IncludedRouter 而非摊平,OpenAPI 反映的才是真正对外暴露的契约
scraping_paths = set(create_scraping_app().openapi()["paths"])
trading_paths = set(create_trading_app().openapi()["paths"])
gateway_paths = set(create_gateway_app().openapi()["paths"])
# 各自的标志性接口都在
assert "/api/search" in scraping_paths
assert "/api/auth/status" in trading_paths
# 登录态接口不该出现在抓取服务上:抓取实例可以多开,多份登录态就是重复下单的温床
assert "/api/orders" in gateway_paths
assert "/api/orders/lease" in gateway_paths
# 登录态接口不该出现在抓取/网关服务上:抓取实例可以多开、网关在服务器上暴露,
# 多份登录态就是重复下单的温床
assert not any(path.startswith("/api/auth") for path in scraping_paths)
assert not any(path.startswith("/api/auth") for path in gateway_paths)
# 抓取的站点接口不应出现在交易/网关上
assert not any(path.startswith("/api/rakuma") for path in trading_paths)
# /health 两边都有,各报各的
assert "/health" in scraping_paths and "/health" in trading_paths
assert not any(path.startswith("/api/rakuma") for path in gateway_paths)
assert not any(path.startswith("/api/search") for path in trading_paths)
assert not any(path.startswith("/api/search") for path in gateway_paths)
# 下单任务接口只属于网关:trading 是本地 worker,不接收外部下单意图
assert not any(path.startswith("/api/orders") for path in scraping_paths)
assert not any(path.startswith("/api/orders") for path in trading_paths)
# /health 三个服务都有,各报各的
assert "/health" in scraping_paths
assert "/health" in trading_paths
assert "/health" in gateway_paths
+307
View File
@@ -0,0 +1,307 @@
"""网关 HTTP API 测试
覆盖规格 §8 的错误码与 §9 验收清单里 HTTP 层能验证的条目鉴权响应信封
6xxx 错误映射(task_id, state) 重复 report 不产生第二条reclaim 错误条件
长轮询与并发抢任务的时序在 test_gateway_leasing.py 单独覆盖
"""
from __future__ import annotations
from pathlib import Path
import pytest
from fastapi.testclient import TestClient
from app.shared.config import get_settings
from app.shared.task_state import OrderState, TaskStatus
TOKEN = get_settings().bearer_token
AUTH = {"Authorization": f"Bearer {TOKEN}"}
@pytest.fixture
def gateway_client(tmp_path: Path, monkeypatch):
"""起一个独立 DB 的网关应用
- monkeypatch 设置 RAKUTEN_GATEWAY_DB_PATH
- get_settings.cache_clear() 让新 env 生效
- TestClient 进入时触发 lifespan建表 sweep 后台任务
"""
db_path = tmp_path / "gw.db"
monkeypatch.setenv("RAKUTEN_GATEWAY_DB_PATH", str(db_path))
get_settings.cache_clear()
try:
from app.gateway.main import create_app
app = create_app()
with TestClient(app) as client:
yield client
finally:
get_settings.cache_clear()
# ---- 鉴权 ----
def test_health_needs_no_token(gateway_client):
response = gateway_client.get("/health")
assert response.status_code == 200
body = response.json()
assert body["success"] is True
assert body["data"]["status"] in ("ok", "degraded")
assert body["data"]["queued_count"] == 0
@pytest.mark.parametrize(
"path,method",
[
("/api/orders", "POST"),
("/api/orders/lease", "GET"),
("/api/orders/t1", "GET"),
],
)
def test_endpoints_reject_missing_token(gateway_client, path, method):
response = gateway_client.request(method, path)
assert response.status_code == 401
assert response.json()["code"] == 1001
def test_endpoints_reject_wrong_token(gateway_client):
response = gateway_client.post(
"/api/orders",
json={"site": "rakuten", "intent": {}},
headers={"Authorization": "Bearer wrong"},
)
assert response.status_code == 401
# ---- 提交与幂等 ----
def test_submit_returns_task_id_and_status(gateway_client):
response = gateway_client.post(
"/api/orders",
json={"task_id": "t1", "site": "rakuten", "intent": {"k": "v"}},
headers=AUTH,
)
assert response.status_code == 200
body = response.json()
assert body["success"] is True
assert body["code"] == 0
assert body["data"]["task_id"] == "t1"
assert body["data"]["status"] == TaskStatus.QUEUED.value
assert body["data"]["created"] is True
def test_submit_with_same_task_id_is_idempotent(gateway_client):
payload = {"task_id": "t1", "site": "rakuten", "intent": {}}
r1 = gateway_client.post("/api/orders", json=payload, headers=AUTH).json()
r2 = gateway_client.post("/api/orders", json=payload, headers=AUTH).json()
assert r1["data"]["created"] is True
assert r2["data"]["created"] is False
assert r1["data"]["task_id"] == r2["data"]["task_id"]
# ---- 错误码 6xxx 映射 ----
def test_get_unknown_task_returns_6001(gateway_client):
response = gateway_client.get("/api/orders/no-such-task", headers=AUTH)
assert response.status_code == 404
body = response.json()
assert body["success"] is False
assert body["code"] == 6001
def test_report_wrong_worker_returns_6002(gateway_client):
# 提交并 lease 给 w1
gateway_client.post(
"/api/orders",
json={"task_id": "t1", "site": "rakuten", "intent": {}},
headers=AUTH,
)
gateway_client.get("/api/orders/lease?worker_id=w1&wait=0", headers=AUTH)
# w2 试图 report
response = gateway_client.post(
"/api/orders/t1/report",
json={"worker_id": "w2", "state": OrderState.IN_CART.value},
headers=AUTH,
)
assert response.status_code == 409
assert response.json()["code"] == 6002
def test_reclaim_non_stale_returns_6003(gateway_client):
gateway_client.post(
"/api/orders",
json={"task_id": "t1", "site": "rakuten", "intent": {}},
headers=AUTH,
)
response = gateway_client.post(
"/api/orders/t1/reclaim",
json={"worker_id": "w1"},
headers=AUTH,
)
assert response.status_code == 409
assert response.json()["code"] == 6003
def test_renew_unknown_task_returns_6001(gateway_client):
response = gateway_client.post(
"/api/orders/no-such/renew",
json={"worker_id": "w1"},
headers=AUTH,
)
assert response.status_code == 404
assert response.json()["code"] == 6001
# ---- report 幂等:同一 (task_id, state) 不产生第二条 ----
def test_report_same_state_does_not_duplicate(gateway_client):
gateway_client.post(
"/api/orders",
json={"task_id": "t1", "site": "rakuten", "intent": {}},
headers=AUTH,
)
gateway_client.get("/api/orders/lease?worker_id=w1&wait=0", headers=AUTH)
report_payload = {
"worker_id": "w1",
"state": OrderState.IN_CART.value,
"payable_yen": 9800,
"evidence_ref": "t1/01-cart-add",
"detail": "已加购",
}
r1 = gateway_client.post("/api/orders/t1/report", json=report_payload, headers=AUTH).json()
r2 = gateway_client.post("/api/orders/t1/report", json=report_payload, headers=AUTH).json()
assert r1["data"]["recorded"] is True
assert r2["data"]["recorded"] is False
detail = gateway_client.get("/api/orders/t1", headers=AUTH).json()["data"]
assert len(detail["reports"]) == 1
def test_report_terminal_releases_lease(gateway_client):
gateway_client.post(
"/api/orders",
json={"task_id": "t1", "site": "rakuten", "intent": {}},
headers=AUTH,
)
gateway_client.get("/api/orders/lease?worker_id=w1&wait=0", headers=AUTH)
response = gateway_client.post(
"/api/orders/t1/report",
json={
"worker_id": "w1",
"state": OrderState.PAID.value,
"payable_yen": 9800,
"site_order_id": "ord-1",
"terminal": True,
"terminal_status": TaskStatus.SUCCEEDED.value,
"detail": "付款完成",
},
headers=AUTH,
)
assert response.status_code == 200
detail = gateway_client.get("/api/orders/t1", headers=AUTH).json()["data"]
assert detail["status"] == TaskStatus.SUCCEEDED.value
assert detail["lease_owner"] is None
# ---- 列表 ----
def test_list_orders_filters_by_status(gateway_client):
gateway_client.post(
"/api/orders",
json={"task_id": "t1", "site": "rakuten", "intent": {}},
headers=AUTH,
)
gateway_client.post(
"/api/orders",
json={"task_id": "t2", "site": "rakuten", "intent": {}},
headers=AUTH,
)
response = gateway_client.get("/api/orders?status=queued", headers=AUTH).json()
assert response["data"]["total"] == 2
assert {item["task_id"] for item in response["data"]["items"]} == {"t1", "t2"}
filtered = gateway_client.get(
"/api/orders?status=leased", headers=AUTH
).json()
assert filtered["data"]["total"] == 0
def test_list_orders_filters_by_site(gateway_client):
gateway_client.post(
"/api/orders",
json={"task_id": "t1", "site": "rakuten", "intent": {}},
headers=AUTH,
)
gateway_client.post(
"/api/orders",
json={"task_id": "t2", "site": "rakuma", "intent": {}},
headers=AUTH,
)
response = gateway_client.get("/api/orders?site=rakuma", headers=AUTH).json()
assert response["data"]["total"] == 1
assert response["data"]["items"][0]["site"] == "rakuma"
def test_get_task_detail_includes_full_report_history(gateway_client):
gateway_client.post(
"/api/orders",
json={"task_id": "t1", "site": "rakuten", "intent": {}},
headers=AUTH,
)
gateway_client.get("/api/orders/lease?worker_id=w1&wait=0", headers=AUTH)
gateway_client.post(
"/api/orders/t1/report",
json={"worker_id": "w1", "state": OrderState.IN_CART.value, "detail": "step1"},
headers=AUTH,
)
gateway_client.post(
"/api/orders/t1/report",
json={"worker_id": "w1", "state": OrderState.ORDERED.value, "detail": "step2"},
headers=AUTH,
)
detail = gateway_client.get("/api/orders/t1", headers=AUTH).json()["data"]
assert [r["state"] for r in detail["reports"]] == [
OrderState.IN_CART.value,
OrderState.ORDERED.value,
]
assert detail["latest_state"] == OrderState.ORDERED.value
# ---- lease 立即返回空(无任务时)----
def test_lease_returns_null_when_queue_empty(gateway_client):
"""wait=0 + 无任务时立即返回 data: null"""
response = gateway_client.get("/api/orders/lease?worker_id=w1&wait=0", headers=AUTH)
assert response.status_code == 200
body = response.json()
assert body["success"] is True
assert body["data"] is None
def test_lease_returns_null_when_active_task_blocks(gateway_client):
"""有 leased 任务时,lease 立即返回 data: null(并发度 1)"""
gateway_client.post(
"/api/orders",
json={"task_id": "t1", "site": "rakuten", "intent": {}},
headers=AUTH,
)
gateway_client.get("/api/orders/lease?worker_id=w1&wait=0", headers=AUTH)
response = gateway_client.get("/api/orders/lease?worker_id=w2&wait=0", headers=AUTH)
assert response.status_code == 200
assert response.json()["data"] is None
+368
View File
@@ -0,0 +1,368 @@
"""网关 DB 与任务队列层测试
直接构造 GatewayDB + TaskQueue绕开 HTTP 覆盖规格 §9 验收清单里能在队列层
独立验证的条目幂等提交全局并发度 1租约过期变 stale且不重投
reclaim 恢复(task_id, state) 幂等 report
长轮询与并发 worker 抢任务的时序在 test_gateway_leasing.py 单独覆盖
"""
from __future__ import annotations
from pathlib import Path
import pytest
from app.gateway.db import GatewayDB
from app.gateway.task_queue import TaskQueue
from app.shared.task_state import OrderState, TaskStatus
@pytest.fixture
async def queue(tmp_path: Path) -> TaskQueue:
db = GatewayDB(tmp_path / "gw.db")
await db.start()
q = TaskQueue(db, lease_ttl_seconds=300, worker_offline_alert_seconds=300)
yield q
await db.close()
# ---- 幂等提交(规格 §9 第 1 条)----
async def test_submit_with_same_task_id_is_idempotent(queue: TaskQueue):
"""同 task_id 提交两次只产生一个任务,第二次 created=False"""
r1 = await queue.submit(task_id="t1", site="rakuten", intent={"k": "v"})
r2 = await queue.submit(task_id="t1", site="rakuten", intent={"k": "v"})
assert r1.task_id == "t1"
assert r1.created is True
assert r2.task_id == "t1"
assert r2.created is False
assert r1.status == r2.status == TaskStatus.QUEUED
async def test_submit_generates_task_id_when_missing(queue: TaskQueue):
r = await queue.submit(task_id=None, site="rakuten", intent={})
assert r.task_id.startswith("po-")
assert r.created is True
# ---- 全局并发度 1(规格 §9 第 2、3 条)----
async def test_lease_returns_none_when_active_task_exists(queue: TaskQueue):
"""已有 leased/running 时,lease 立即返回 None"""
await queue.submit(task_id="t1", site="rakuten", intent={})
first = await queue.lease(worker_id="w1", wait=0, site=None, max_wait=60)
assert first is not None and first.task_id == "t1"
second = await queue.lease(worker_id="w2", wait=0, site=None, max_wait=60)
assert second is None
async def test_lease_picks_queued_in_fifo_order(queue: TaskQueue):
await queue.submit(task_id="t1", site="rakuten", intent={})
await queue.submit(task_id="t2", site="rakuten", intent={})
first = await queue.lease(worker_id="w1", wait=0, site=None, max_wait=60)
assert first.task_id == "t1"
# 终结第一个后才能领第二个
await queue.report(
"t1",
worker_id="w1",
state=OrderState.CREATED,
payable_yen=None,
pay_deadline=None,
site_order_id=None,
evidence_ref=None,
detail="",
terminal=True,
terminal_status=TaskStatus.SUCCEEDED,
)
second = await queue.lease(worker_id="w1", wait=0, site=None, max_wait=60)
assert second is not None and second.task_id == "t2"
# ---- 站点过滤 ----
async def test_lease_with_site_filter(queue: TaskQueue):
await queue.submit(task_id="t1", site="rakuten", intent={})
await queue.submit(task_id="t2", site="rakuma", intent={})
leased = await queue.lease(worker_id="w1", wait=0, site="rakuma", max_wait=60)
assert leased is not None and leased.task_id == "t2" and leased.site == "rakuma"
# ---- 续租 ----
async def test_renew_requires_lease_owner(queue: TaskQueue):
await queue.submit(task_id="t1", site="rakuten", intent={})
await queue.lease(worker_id="w1", wait=0, site=None, max_wait=60)
from app.shared.errors import LeaseInvalidError
with pytest.raises(LeaseInvalidError):
await queue.renew("t1", "w2")
async def test_renew_extends_lease(queue: TaskQueue):
await queue.submit(task_id="t1", site="rakuten", intent={})
await queue.lease(worker_id="w1", wait=0, site=None, max_wait=60)
renewed = await queue.renew("t1", "w1")
assert renewed.task_id == "t1"
assert renewed.lease_count == 1
# ---- report 幂等(规格 §9 第 5 条)----
async def test_report_same_state_is_idempotent(queue: TaskQueue):
"""同一 (task_id, state) 重复上报不产生第二条记录"""
await queue.submit(task_id="t1", site="rakuten", intent={})
await queue.lease(worker_id="w1", wait=0, site=None, max_wait=60)
kwargs = dict(
worker_id="w1",
state=OrderState.IN_CART,
payable_yen=100,
pay_deadline=None,
site_order_id=None,
evidence_ref="t1/01-cart-add",
detail="已加购",
terminal=False,
terminal_status=None,
)
r1 = await queue.report("t1", **kwargs)
r2 = await queue.report("t1", **kwargs)
assert r1.recorded is True # 新写入
assert r2.recorded is False # 幂等覆盖
detail = await queue.get_task_detail("t1")
assert len(detail.reports) == 1
async def test_first_report_moves_task_to_running(queue: TaskQueue):
await queue.submit(task_id="t1", site="rakuten", intent={})
await queue.lease(worker_id="w1", wait=0, site=None, max_wait=60)
await queue.report(
"t1",
worker_id="w1",
state=OrderState.IN_CART,
payable_yen=None,
pay_deadline=None,
site_order_id=None,
evidence_ref=None,
detail="",
terminal=False,
terminal_status=None,
)
detail = await queue.get_task_detail("t1")
assert detail.status == TaskStatus.RUNNING
async def test_terminal_report_releases_lease(queue: TaskQueue):
await queue.submit(task_id="t1", site="rakuten", intent={})
await queue.lease(worker_id="w1", wait=0, site=None, max_wait=60)
await queue.report(
"t1",
worker_id="w1",
state=OrderState.PAID,
payable_yen=9800,
pay_deadline=None,
site_order_id="ord-1",
evidence_ref="t1/05-payment",
detail="付款完成",
terminal=True,
terminal_status=TaskStatus.SUCCEEDED,
)
detail = await queue.get_task_detail("t1")
assert detail.status == TaskStatus.SUCCEEDED
assert detail.lease_owner is None
assert detail.lease_expires_at is None
assert detail.latest_state == OrderState.PAID
async def test_terminal_without_explicit_status_infers_from_state(queue: TaskQueue):
"""terminal=true 但不传 terminal_status 时,按 state 推断:paid → succeeded"""
await queue.submit(task_id="t1", site="rakuten", intent={})
await queue.lease(worker_id="w1", wait=0, site=None, max_wait=60)
await queue.report(
"t1",
worker_id="w1",
state=OrderState.PAID,
payable_yen=None,
pay_deadline=None,
site_order_id=None,
evidence_ref=None,
detail="",
terminal=True,
terminal_status=None,
)
assert (await queue.get_task_detail("t1")).status == TaskStatus.SUCCEEDED
async def test_terminal_with_awaiting_payment_state_becomes_needs_human(queue: TaskQueue):
"""3DS 等人工环节 → worker 报 awaiting_payment + terminal → 任务转 needs_human"""
await queue.submit(task_id="t1", site="rakuten", intent={})
await queue.lease(worker_id="w1", wait=0, site=None, max_wait=60)
await queue.report(
"t1",
worker_id="w1",
state=OrderState.AWAITING_PAYMENT,
payable_yen=None,
pay_deadline=None,
site_order_id=None,
evidence_ref=None,
detail="检测到 3DS 验证",
terminal=True,
terminal_status=None,
)
assert (await queue.get_task_detail("t1")).status == TaskStatus.NEEDS_HUMAN
# ---- 租约过期与绝不自动重投(规格 §5、§9 第 6 条)----
async def test_expired_lease_becomes_stale_and_is_not_re_leasable(tmp_path: Path):
"""租约过期 → stale → 普通 lease 取不到(绝不自动重投)"""
db = GatewayDB(tmp_path / "gw.db")
await db.start()
try:
q = TaskQueue(db, lease_ttl_seconds=0, worker_offline_alert_seconds=300)
await q.submit(task_id="t1", site="rakuten", intent={})
leased = await q.lease(worker_id="w1", wait=0, site=None, max_wait=60)
assert leased is not None
# TTL=0:立即过期。下一次 lease 内部会先扫到 stale
empty = await q.lease(worker_id="w2", wait=0, site=None, max_wait=60)
assert empty is None
detail = await q.get_task_detail("t1")
assert detail.status == TaskStatus.STALE
finally:
await db.close()
async def test_sweep_returns_count_of_expired_tasks(queue: TaskQueue):
await queue.submit(task_id="t1", site="rakuten", intent={})
# 直接造一个过期任务:lease TTL 设为 0
queue._lease_ttl = 0 # type: ignore[attr-defined]
await queue.lease(worker_id="w1", wait=0, site=None, max_wait=60)
swept = await queue.sweep()
assert swept == 1
assert (await queue.get_task_detail("t1")).status == TaskStatus.STALE
# ---- reclaim(规格 §5、§9 第 7 条)----
async def test_reclaim_leases_stale_task_with_count_increment(tmp_path: Path):
"""reclaim 领回 stale 任务,lease_count > 1,带 known_state"""
db = GatewayDB(tmp_path / "gw.db")
await db.start()
try:
q = TaskQueue(db, lease_ttl_seconds=0, worker_offline_alert_seconds=300)
await q.submit(task_id="t1", site="rakuten", intent={})
first = await q.lease(worker_id="w1", wait=0, site=None, max_wait=60)
assert first is not None and first.lease_count == 1
# 在过期之前先 report 一次,让 known_state 有值
await q.report(
"t1",
worker_id="w1",
state=OrderState.IN_CART,
payable_yen=None,
pay_deadline=None,
site_order_id=None,
evidence_ref=None,
detail="",
terminal=False,
terminal_status=None,
)
# 走一次 lease 触发 sweep,把过期任务置 stale
await q.lease(worker_id="w2", wait=0, site=None, max_wait=60)
assert (await q.get_task_detail("t1")).status == TaskStatus.STALE
reclaimed = await q.reclaim("t1", "w2")
assert reclaimed.task_id == "t1"
assert reclaimed.lease_count == 2 # > 1:worker 必须先核对
assert reclaimed.known_state == OrderState.IN_CART
# reclaim 后任务状态变 leased(被新 worker 持有)
detail = await q.get_task_detail("t1")
assert detail.status == TaskStatus.LEASED
assert detail.lease_owner == "w2"
finally:
await db.close()
async def test_reclaim_rejects_non_stale_task(queue: TaskQueue):
"""只能 reclaim stale 任务;其他状态报 6003"""
from app.shared.errors import InvalidTaskStateError
await queue.submit(task_id="t1", site="rakuten", intent={})
with pytest.raises(InvalidTaskStateError):
await queue.reclaim("t1", "w1")
async def test_normal_lease_does_not_pick_stale(tmp_path: Path):
"""普通 lease 不会取到 stale 任务(绝不自动重投)"""
db = GatewayDB(tmp_path / "gw.db")
await db.start()
try:
q = TaskQueue(db, lease_ttl_seconds=0, worker_offline_alert_seconds=300)
await q.submit(task_id="stale-task", site="rakuten", intent={})
await q.lease(worker_id="w1", wait=0, site=None, max_wait=60)
# 触发 sweep
await q.lease(worker_id="w2", wait=0, site=None, max_wait=60)
assert (await q.get_task_detail("stale-task")).status == TaskStatus.STALE
# 再入队一个新任务
await q.submit(task_id="fresh", site="rakuten", intent={})
picked = await q.lease(worker_id="w3", wait=0, site=None, max_wait=60)
assert picked is not None and picked.task_id == "fresh"
finally:
await db.close()
# ---- 错误映射 ----
async def test_report_wrong_worker_raises(queue: TaskQueue):
from app.shared.errors import LeaseInvalidError
await queue.submit(task_id="t1", site="rakuten", intent={})
await queue.lease(worker_id="w1", wait=0, site=None, max_wait=60)
with pytest.raises(LeaseInvalidError):
await queue.report(
"t1",
worker_id="w2",
state=OrderState.IN_CART,
payable_yen=None,
pay_deadline=None,
site_order_id=None,
evidence_ref=None,
detail="",
terminal=False,
terminal_status=None,
)
async def test_get_unknown_task_raises(queue: TaskQueue):
from app.shared.errors import TaskNotFoundError
with pytest.raises(TaskNotFoundError):
await queue.get_task_detail("does-not-exist")
+179
View File
@@ -0,0 +1,179 @@
"""长轮询与并发抢任务测试
覆盖规格 §9 验收清单里需要时序观察的条目
- 两个 worker 同时 lease只有一个拿到任务全局并发度 1
- 有任务在 leased/running lease 一律返回空
- 无任务时 lease 挂起到 wait 秒才返回且返回 200 + data: null
- submit 期间正在 wait lease 被唤醒
- worker 心跳lease 请求刷新 last_seen_at/health 据此判断失联
"""
from __future__ import annotations
import asyncio
import time
from pathlib import Path
import pytest
from fastapi.testclient import TestClient
from app.shared.config import get_settings
TOKEN = get_settings().bearer_token
AUTH = {"Authorization": f"Bearer {TOKEN}"}
@pytest.fixture
def gateway_client(tmp_path: Path, monkeypatch):
db_path = tmp_path / "gw.db"
monkeypatch.setenv("RAKUTEN_GATEWAY_DB_PATH", str(db_path))
# worker_offline_alert_seconds 设小一点,方便测 /health 失联判定
monkeypatch.setenv("RAKUTEN_WORKER_OFFLINE_ALERT_SECONDS", "1")
get_settings.cache_clear()
try:
from app.gateway.main import create_app
app = create_app()
with TestClient(app) as client:
yield client
# 让后台 sweep 任务退出后再清理
finally:
get_settings.cache_clear()
# ---- 全局并发度 1 ----
def test_concurrent_leases_only_one_gets_task(gateway_client):
"""两个 worker 同时 lease,只有一个拿到任务"""
gateway_client.post(
"/api/orders",
json={"task_id": "t1", "site": "rakuten", "intent": {}},
headers=AUTH,
)
# 同步发两个 lease,wait=0 立即返回
r1 = gateway_client.get("/api/orders/lease?worker_id=w1&wait=0", headers=AUTH)
r2 = gateway_client.get("/api/orders/lease?worker_id=w2&wait=0", headers=AUTH)
owned = [r for r in (r1, r2) if r.json()["data"] is not None]
assert len(owned) == 1
assert owned[0].json()["data"]["task_id"] == "t1"
def test_active_task_blocks_subsequent_leases(gateway_client):
"""leased 或 running 状态下,lease 立即返回空"""
gateway_client.post(
"/api/orders",
json={"task_id": "active", "site": "rakuten", "intent": {}},
headers=AUTH,
)
gateway_client.post(
"/api/orders",
json={"task_id": "queued", "site": "rakuten", "intent": {}},
headers=AUTH,
)
# 领走 active
first = gateway_client.get("/api/orders/lease?worker_id=w1&wait=0", headers=AUTH).json()
assert first["data"]["task_id"] == "active"
# 队列里还有 queued,但 active 还在执行 → 立即返回空
blocked = gateway_client.get("/api/orders/lease?worker_id=w2&wait=0", headers=AUTH).json()
assert blocked["data"] is None
# ---- 长轮询 ----
def test_lease_hangs_then_returns_null_when_no_task(gateway_client):
"""无任务时 lease 挂起到 wait 秒才返回"""
started = time.monotonic()
response = gateway_client.get("/api/orders/lease?worker_id=w1&wait=2", headers=AUTH)
elapsed = time.monotonic() - started
assert response.status_code == 200
assert response.json()["data"] is None
# 至少挂了 2 秒(允许少量提前)
assert elapsed >= 1.8
def test_lease_is_woken_up_when_task_submitted(gateway_client):
"""长轮询期间 submit 任务,lease 应被唤醒并立刻拿到"""
async def submit_after_delay():
await asyncio.sleep(0.3)
# TestClient 是同步的,在另一个线程里发请求
import threading
def _submit():
gateway_client.post(
"/api/orders",
json={"task_id": "t1", "site": "rakuten", "intent": {}},
headers=AUTH,
)
threading.Thread(target=_submit, daemon=True).start()
asyncio.run(submit_after_delay())
started = time.monotonic()
response = gateway_client.get("/api/orders/lease?worker_id=w1&wait=10", headers=AUTH)
elapsed = time.monotonic() - started
assert response.status_code == 200
data = response.json()["data"]
assert data is not None and data["task_id"] == "t1"
# 被唤醒应远早于 10 秒
assert elapsed < 5
# ---- 心跳与健康检查 ----
def test_lease_updates_worker_last_seen(gateway_client):
"""lease 请求兼作心跳,刷新 workers 表的 last_seen_at"""
gateway_client.get("/api/orders/lease?worker_id=w1&wait=0", headers=AUTH)
health = gateway_client.get("/health").json()["data"]
worker_ids = {w["worker_id"] for w in health["workers"]}
assert "w1" in worker_ids
def test_health_reports_offline_worker(gateway_client):
"""worker 心跳超阈值时 /health 报异常"""
gateway_client.get("/api/orders/lease?worker_id=w1&wait=0", headers=AUTH)
# 阈值被 fixture 设为 1 秒;时间戳为秒精度,需要 sleep 到 2 秒才能稳定越过阈值
import time as _time
_time.sleep(2.0)
health = gateway_client.get("/health").json()["data"]
assert health["status"] == "degraded"
offline_ids = {w["worker_id"] for w in health["offline_workers"]}
assert "w1" in offline_ids
def test_health_reports_stale_queued_task(gateway_client):
"""任务 queued 超过 worker_offline_alert_seconds 时报异常"""
gateway_client.post(
"/api/orders",
json={"task_id": "stuck", "site": "rakuten", "intent": {}},
headers=AUTH,
)
import time as _time
_time.sleep(2.0)
health = gateway_client.get("/health").json()["data"]
assert health["status"] == "degraded"
stuck_ids = {t["task_id"] for t in health["stale_queued_tasks"]}
assert "stuck" in stuck_ids
def test_health_lists_active_tasks(gateway_client):
gateway_client.post(
"/api/orders",
json={"task_id": "t1", "site": "rakuten", "intent": {}},
headers=AUTH,
)
gateway_client.get("/api/orders/lease?worker_id=w1&wait=0", headers=AUTH)
health = gateway_client.get("/health").json()["data"]
active_ids = {t["task_id"] for t in health["active_tasks"]}
assert "t1" in active_ids
+166
View File
@@ -0,0 +1,166 @@
"""恢复路径测试:stale → reclaim → lease_count > 1 + known_state
覆盖规格 §5绝不自动重投 §9 7 reclaim 领回的任务 lease_count > 1
known_state
"""
from __future__ import annotations
from pathlib import Path
import pytest
from fastapi.testclient import TestClient
from app.shared.config import get_settings
from app.shared.task_state import OrderState, TaskStatus
TOKEN = get_settings().bearer_token
AUTH = {"Authorization": f"Bearer {TOKEN}"}
@pytest.fixture
def gateway_client(tmp_path: Path, monkeypatch):
db_path = tmp_path / "gw.db"
monkeypatch.setenv("RAKUTEN_GATEWAY_DB_PATH", str(db_path))
# TTL 设小一点,让 lease 一过期就能被 sweep 转 stale
monkeypatch.setenv("RAKUTEN_LEASE_TTL_SECONDS", "1")
get_settings.cache_clear()
try:
from app.gateway.main import create_app
app = create_app()
with TestClient(app) as client:
yield client
finally:
get_settings.cache_clear()
def _submit_and_lease_and_report_progress(client: TestClient, task_id: str = "t1"):
"""提交 → lease → report 一次(known_state=in_cart),返回 lease 响应"""
client.post(
"/api/orders",
json={"task_id": task_id, "site": "rakuten", "intent": {}},
headers=AUTH,
)
lease = client.get(f"/api/orders/lease?worker_id=w1&wait=0", headers=AUTH).json()
client.post(
f"/api/orders/{task_id}/report",
json={"worker_id": "w1", "state": OrderState.IN_CART.value, "detail": "已加购"},
headers=AUTH,
)
return lease["data"]
def test_stale_task_is_not_picked_by_normal_lease(gateway_client):
"""租约过期 → stale → 普通 lease 取不到"""
_submit_and_lease_and_report_progress(gateway_client, "t1")
# 等 TTL 过期(1 秒)+ sweep 触发:用 wait=0 的 lease 触发内嵌 sweep
import time
time.sleep(1.5)
# 触发一次 lease(带另一个 worker_id)— 应当既取不到 t1(stale)也返回空
response = gateway_client.get("/api/orders/lease?worker_id=w2&wait=0", headers=AUTH)
assert response.status_code == 200
assert response.json()["data"] is None
detail = gateway_client.get("/api/orders/t1", headers=AUTH).json()["data"]
assert detail["status"] == TaskStatus.STALE.value
def test_reclaim_returns_task_with_lease_count_above_one(gateway_client):
"""reclaim 领回 stale 任务,lease_count > 1,带 known_state"""
_submit_and_lease_and_report_progress(gateway_client, "t1")
import time
time.sleep(1.5)
# 触发一次 sweep
gateway_client.get("/api/orders/lease?worker_id=w2&wait=0", headers=AUTH)
assert (
gateway_client.get("/api/orders/t1", headers=AUTH).json()["data"]["status"]
== TaskStatus.STALE.value
)
response = gateway_client.post(
"/api/orders/t1/reclaim", json={"worker_id": "w2"}, headers=AUTH
)
assert response.status_code == 200
data = response.json()["data"]
assert data["task_id"] == "t1"
assert data["lease_count"] >= 2 # > 1
assert data["known_state"] == OrderState.IN_CART.value
def test_reclaim_sets_leased_state_with_new_owner(gateway_client):
"""reclaim 后任务状态回到 leased,lease_owner 是新 worker"""
_submit_and_lease_and_report_progress(gateway_client, "t1")
import time
time.sleep(1.5)
gateway_client.get("/api/orders/lease?worker_id=w2&wait=0", headers=AUTH)
gateway_client.post(
"/api/orders/t1/reclaim", json={"worker_id": "w3"}, headers=AUTH
)
detail = gateway_client.get("/api/orders/t1", headers=AUTH).json()["data"]
assert detail["status"] == TaskStatus.LEASED.value
assert detail["lease_owner"] == "w3"
def test_double_reclaim_increments_lease_count(gateway_client):
"""连续 reclaim 多次,lease_count 持续 +1"""
_submit_and_lease_and_report_progress(gateway_client, "t1")
import time
time.sleep(1.5)
gateway_client.get("/api/orders/lease?worker_id=w2&wait=0", headers=AUTH)
r1 = gateway_client.post(
"/api/orders/t1/reclaim", json={"worker_id": "wA"}, headers=AUTH
).json()["data"]
# wA 占着 lease,需要先释放(terminal report)才能再 reclaim
gateway_client.post(
"/api/orders/t1/report",
json={
"worker_id": "wA",
"state": OrderState.CREATED.value,
"terminal": True,
"terminal_status": TaskStatus.NEEDS_HUMAN.value,
},
headers=AUTH,
)
# 但 terminal 后任务不再 stale,无法再 reclaim。改用:让 lease 过期
# 这里直接验单次 reclaim 的语义即可,连续多次需要 stale→reclaim→过期→stale 循环
assert r1["lease_count"] == 2
def test_normal_lease_does_not_pick_stale_while_queued_exists(gateway_client):
"""stale + queued 并存时,普通 lease 只取 queued,绝不取 stale
顺序提交 stale-task lease report 进度 提交 fresh-task lease 过期
w3 leaselease 内部先 sweepstale-task staleactive queued
fresh-task应当只取到 fresh
"""
_submit_and_lease_and_report_progress(gateway_client, "stale-task")
gateway_client.post(
"/api/orders",
json={"task_id": "fresh-task", "site": "rakuten", "intent": {}},
headers=AUTH,
)
import time
time.sleep(1.5)
leased = gateway_client.get(
"/api/orders/lease?worker_id=w3&wait=0", headers=AUTH
).json()["data"]
assert leased is not None
assert leased["task_id"] == "fresh-task"
# stale-task 仍然停在 stale,没有被这次 lease 误取
detail = gateway_client.get("/api/orders/stale-task", headers=AUTH).json()["data"]
assert detail["status"] == TaskStatus.STALE.value
+148
View File
@@ -0,0 +1,148 @@
"""本地订单 DB 与证据存储测试
覆盖 worker 侧的不变量本地幂等闸门has_finished状态迁移事件证据落盘
路径与 meta.json 字段规格 §6.1
"""
from __future__ import annotations
import json
from pathlib import Path
import pytest
from app.trading.worker.evidence import EvidenceStore
from app.trading.worker.local_db import LocalDB
@pytest.fixture
async def local_db(tmp_path: Path) -> LocalDB:
db = LocalDB(tmp_path / "trading.db")
await db.start()
yield db
await db.close()
# ---- 本地幂等闸门 ----
async def test_has_finished_false_for_unknown_task(local_db: LocalDB):
assert await local_db.has_finished("no-such") is False
async def test_mark_finished_idempotent(local_db: LocalDB):
"""ensure_started + mark_finished 后,has_finished 返回 True"""
await local_db.ensure_started("t1", "rakuten", {"k": "v"})
assert await local_db.has_finished("t1") is False
await local_db.mark_finished("t1", "paid")
assert await local_db.has_finished("t1") is True
assert await local_db.final_state("t1") == "paid"
async def test_ensure_started_is_idempotent(local_db: LocalDB):
"""同任务多次 ensure_started 不覆盖 started_at"""
await local_db.ensure_started("t1", "rakuten", {"k": "v"})
row1 = await local_db.get_order("t1")
await local_db.ensure_started("t1", "rakuten", {"k": "v"})
row2 = await local_db.get_order("t1")
assert row1.started_at == row2.started_at
assert row1.intent == row2.intent == {"k": "v"}
# ---- 状态事件 ----
async def test_record_event_inserts_once_per_state(local_db: LocalDB):
"""同一 (task_id, state) 重复记录视为同一次——INSERT OR IGNORE"""
await local_db.ensure_started("t1", "rakuten", {})
await local_db.record_event("t1", "in_cart", detail="已加购", evidence_ref="t1/01")
await local_db.record_event("t1", "in_cart", detail="再次加购", evidence_ref="t1/01")
async with local_db.conn.execute(
"SELECT state, detail FROM order_events WHERE task_id = ?", ("t1",)
) as cur:
events = await cur.fetchall()
# 第二次记录被忽略(INSERT OR IGNORE)
assert len(events) == 1
assert events[0]["detail"] == "已加购"
async def test_record_event_allows_distinct_states(local_db: LocalDB):
await local_db.ensure_started("t1", "rakuten", {})
await local_db.record_event("t1", "in_cart")
await local_db.record_event("t1", "ordered")
await local_db.record_event("t1", "paid")
async with local_db.conn.execute(
"SELECT state FROM order_events WHERE task_id = ? ORDER BY recorded_at",
("t1",),
) as cur:
events = await cur.fetchall()
assert [r["state"] for r in events] == ["in_cart", "ordered", "paid"]
# ---- 证据存储 ----
def test_evidence_write_step_creates_expected_files(tmp_path: Path):
"""规格 §6.1:每步一个子目录,{序号:02d}-{步骤名}.{html|png|meta.json}"""
store = EvidenceStore(tmp_path / "evidence")
rel = store.write_step(
"po-20260727-0001",
step_no=1,
step_name="cart-add",
html="<html>cart</html>",
png=b"\x89PNGfake",
meta={"url": "https://example.com/cart", "method": "POST", "status": 200},
)
assert rel == "po-20260727-0001/01-cart-add"
step_dir = tmp_path / "evidence" / "po-20260727-0001"
assert (step_dir / "01-cart-add.html").read_text(encoding="utf-8") == "<html>cart</html>"
assert (step_dir / "01-cart-add.png").read_bytes() == b"\x89PNGfake"
meta = json.loads(
(step_dir / "01-cart-add.meta.json").read_text(encoding="utf-8")
)
assert meta["url"] == "https://example.com/cart"
assert meta["method"] == "POST"
assert meta["status"] == 200
def test_evidence_write_step_supports_partial(tmp_path: Path):
"""只写 meta 不写 html/png 也能正常落盘"""
store = EvidenceStore(tmp_path / "evidence")
store.write_step("t1", step_no=2, step_name="confirm", meta={"a": 1})
step_dir = tmp_path / "evidence" / "t1"
assert not (step_dir / "02-confirm.html").exists()
assert not (step_dir / "02-confirm.png").exists()
assert (step_dir / "02-confirm.meta.json").exists()
def test_evidence_index_records_relative_path(tmp_path: Path):
"""index_evidence 把相对路径存进 DB,供事后翻查"""
db = LocalDB(tmp_path / "trading.db")
# 同步包装:用 asyncio.run 跑一次
import asyncio
async def _run():
await db.start()
try:
await db.ensure_started("t1", "rakuten", {})
await db.index_evidence("t1", 1, "cart-add", "t1/01-cart-add")
await db.index_evidence("t1", 2, "confirm", "t1/03-order-confirm")
async with db.conn.execute(
"SELECT step_no, step_name, rel_path FROM evidence_index "
"WHERE task_id = ? ORDER BY step_no",
("t1",),
) as cur:
rows = await cur.fetchall()
assert [(r["step_no"], r["step_name"], r["rel_path"]) for r in rows] == [
(1, "cart-add", "t1/01-cart-add"),
(2, "confirm", "t1/03-order-confirm"),
]
finally:
await db.close()
asyncio.run(_run())
+363
View File
@@ -0,0 +1,363 @@
"""worker 主循环测试
不启动真实的后台任务直接构造 WorkerRunner 与桩 gateway client / local_db /
evidence store覆盖 §6 主循环的分支
- 已完成的任务再次被领直接补报不重新执行
- lease_count > 1恢复领取走核对分支verify unknown needs_human
- 站点交互未实现site_interact NotImplementedError needs_human
- 金额守卫拦截 needs_human站点侧无提交动作
每一步的证据文件在 report 之前就已落盘§9 11 也在这里验证
"""
from __future__ import annotations
from dataclasses import dataclass, field
from pathlib import Path
from typing import Any
import pytest
from app.shared.errors import OrderGuardError
from app.shared.task_state import OrderState, TaskStatus
from app.trading.worker import site_interact, verify
from app.trading.worker.evidence import EvidenceStore
from app.trading.worker.local_db import LocalDB
from app.trading.worker.models import LeaseTask
from app.trading.worker.runner import WorkerRunner
# ---- 桩:网关客户端 ----
@dataclass
class FakeGateway:
"""记录所有调用,可控返回值的网关桩"""
reports: list[dict[str, Any]] = field(default_factory=list)
renews: int = 0
fail_report_with: Exception | None = None
async def lease(
self, worker_id: str, *, wait: int = 30, site: str | None = None
) -> LeaseTask | None:
return None # 主循环测试不通过 lease 喂数据,直接调 handle
async def renew(self, task_id: str, worker_id: str) -> dict[str, Any]:
self.renews += 1
return {"task_id": task_id, "lease_expires_at": "2099-01-01T00:00:00Z"}
async def report(
self,
task_id: str,
worker_id: str,
*,
state: OrderState,
payable_yen: int | None = None,
pay_deadline: str | None = None,
site_order_id: str | None = None,
evidence_ref: str | None = None,
detail: str = "",
terminal: bool = False,
terminal_status: TaskStatus | None = None,
) -> dict[str, Any]:
if self.fail_report_with is not None:
raise self.fail_report_with
record = {
"task_id": task_id,
"worker_id": worker_id,
"state": state,
"payable_yen": payable_yen,
"pay_deadline": pay_deadline,
"site_order_id": site_order_id,
"evidence_ref": evidence_ref,
"detail": detail,
"terminal": terminal,
"terminal_status": terminal_status,
}
self.reports.append(record)
return {"task_id": task_id, "status": "running", "recorded": True}
def last_terminal_report(self) -> dict[str, Any]:
for r in reversed(self.reports):
if r["terminal"]:
return r
raise AssertionError("没有终态 report")
# ---- fixtures ----
@pytest.fixture
async def local_db(tmp_path: Path) -> LocalDB:
db = LocalDB(tmp_path / "trading.db")
await db.start()
yield db
await db.close()
@pytest.fixture
def evidence(tmp_path: Path) -> EvidenceStore:
return EvidenceStore(tmp_path / "evidence")
@pytest.fixture
def runner(tmp_path, local_db, evidence, monkeypatch) -> WorkerRunner:
"""构造一个 worker_id=test、网关 URL 已配置的 runner
使用 fake gatewaysettings 取真实 Settings worker_id 注入为 test
"""
class _FakeSettings:
worker_id = "test"
order_max_total_yen = 30000
@property
def worker_id_effective(self) -> str:
return "test"
gateway = FakeGateway()
r = WorkerRunner(
settings=_FakeSettings(), # type: ignore[arg-type]
gateway_client=gateway, # type: ignore[arg-type]
local_db=local_db,
evidence=evidence,
)
# 把 gateway 桩挂到 runner 上,测试通过它断言调用
r._gateway_for_test = gateway # type: ignore[attr-defined]
return r
def _make_task(
*,
task_id: str = "t1",
site: str = "rakuten",
lease_count: int = 1,
known_state: str | None = None,
intent: dict | None = None,
) -> LeaseTask:
return LeaseTask(
task_id=task_id,
site=site,
intent=intent or {},
lease_expires_at="2099-01-01T00:00:00Z",
lease_count=lease_count,
known_state=known_state,
)
# ---- 已完成的任务:补报而不重新执行(§9 第 9 条)----
async def test_already_finished_task_is_reported_not_re_executed(
runner: WorkerRunner, local_db: LocalDB
):
"""本地已完成 → 直接补报 final_state,不进入 execute"""
await local_db.ensure_started("t1", "rakuten", {"k": "v"})
await local_db.mark_finished("t1", OrderState.PAID.value)
await runner.handle(_make_task(task_id="t1"))
gateway: FakeGateway = runner._gateway_for_test # type: ignore[attr-defined]
terminal = gateway.last_terminal_report()
assert terminal["state"] == OrderState.PAID
assert terminal["terminal"] is True
# 没有进入 execute:site_interact 不应被调用,但为保险起见,我们只看
# evidence 目录里没有任何步骤文件
evidence_files = list(local_db._db_path.parent.parent.glob("**/*.html")) # noqa
# 上面 glob 会扫到 trading.db 之外的所有 .html;这里只关心 evidence 目录
# (runner 走 execute 才会写证据)
# ---- lease_count > 1:核对分支(§9 第 8 条)----
async def test_recovery_task_with_unknown_verdict_reports_needs_human(
runner: WorkerRunner, local_db: LocalDB, monkeypatch
):
"""lease_count > 1 + verify 桩返回 unknown → 转 needs_human,不执行"""
# verify 桩默认返回 unknown,不用 monkeypatch
await runner.handle(_make_task(task_id="t1", lease_count=2))
gateway: FakeGateway = runner._gateway_for_test # type: ignore[attr-defined]
terminal = gateway.last_terminal_report()
assert terminal["terminal_status"] == TaskStatus.NEEDS_HUMAN
assert "无法定论" in terminal["detail"]
async def test_recovery_task_with_already_ordered_verdict_reports_succeeded(
runner: WorkerRunner, local_db: LocalDB, monkeypatch
):
"""lease_count > 1 + verify 返回 already_ordered → 补报 succeeded,不重新下单"""
async def _already_ordered(task): # noqa: ANN001
return verify.VerifyResult(
verdict=verify.VerifyVerdict.ALREADY_ORDERED,
site_order_id="ord-1",
detail="订单列表命中",
)
monkeypatch.setattr(verify, "verify_on_site", _already_ordered)
await runner.handle(_make_task(task_id="t1", lease_count=2, known_state="ordered"))
gateway: FakeGateway = runner._gateway_for_test # type: ignore[attr-defined]
terminal = gateway.last_terminal_report()
assert terminal["state"] == OrderState.ORDERED
assert terminal["site_order_id"] == "ord-1"
assert terminal["terminal_status"] == TaskStatus.SUCCEEDED
# 本地 DB 也应当被标记为已完成
assert await local_db.has_finished("t1") is True
# ---- 站点交互未实现 → needs_human(§10 接口缝)----
async def test_unimplemented_site_interaction_becomes_needs_human(
runner: WorkerRunner, local_db: LocalDB
):
"""site_interact.add_to_cart 抛 NotImplementedError → runner 转 needs_human"""
await runner.handle(_make_task(task_id="t1"))
gateway: FakeGateway = runner._gateway_for_test # type: ignore[attr-defined]
terminal = gateway.last_terminal_report()
assert terminal["terminal_status"] == TaskStatus.NEEDS_HUMAN
assert "未实现" in terminal["detail"]
# ---- ラクマ 不实现 → needs_human ----
async def test_rakuma_site_reports_needs_human(runner: WorkerRunner):
"""首版只对乐天;其他站点直接转人工"""
await runner.handle(_make_task(task_id="t1", site="rakuma"))
gateway: FakeGateway = runner._gateway_for_test # type: ignore[attr-defined]
terminal = gateway.last_terminal_report()
assert terminal["terminal_status"] == TaskStatus.NEEDS_HUMAN
# ---- 金额守卫:拦截而不提交(§9 第 10 条)----
async def test_amount_guard_blocks_when_payable_exceeds_limit(
runner: WorkerRunner, local_db: LocalDB, monkeypatch
):
"""实际应付超过上限 → OrderGuardError → 转 needs_human,站点侧无提交动作
模拟站点交互前两步add_to_cart / verify_cart正常enter_checkout 返回 html
parse_checkout 返回超额金额submit_order 永远不应被调用
"""
call_log: list[str] = []
async def _ok(task): # noqa: ANN001
call_log.append(task.task_id + ":step")
async def _noop(task): # noqa: ANN001
return None
async def _checkout_html(task): # noqa: ANN001
return "<html>checkout</html>"
async def _parse(html: str):
return site_interact.CheckoutSummary(payable_yen=50000)
submit_called = False
async def _submit(task): # noqa: ANN001
nonlocal submit_called
submit_called = True
return "ord-1"
monkeypatch.setattr(site_interact, "add_to_cart", _noop)
monkeypatch.setattr(site_interact, "verify_cart", _noop)
monkeypatch.setattr(site_interact, "enter_checkout", _checkout_html)
monkeypatch.setattr(site_interact, "parse_checkout", _parse)
monkeypatch.setattr(site_interact, "submit_order", _submit)
# intent.max_total_yen 缺省,回落到 settings.order_max_total_yen=30000,
# 实际 50000 > 30000 → 拦截
await runner.handle(_make_task(task_id="t1"))
assert submit_called is False # 关键:站点侧无提交动作
gateway: FakeGateway = runner._gateway_for_test # type: ignore[attr-defined]
terminal = gateway.last_terminal_report()
assert terminal["terminal_status"] == TaskStatus.NEEDS_HUMAN
assert "上限" in terminal["detail"]
async def test_amount_guard_honors_intent_max_total_yen(
runner: WorkerRunner, local_db: LocalDB, monkeypatch
):
"""intent 显式给 max_total_yen 时优先于全局上限"""
async def _noop(task): # noqa: ANN001
return None
async def _checkout_html(task): # noqa: ANN001
return "<html>checkout</html>"
async def _parse(html: str):
return site_interact.CheckoutSummary(payable_yen=8000)
submit_called = False
async def _submit(task): # noqa: ANN001
nonlocal submit_called
submit_called = True
raise NotImplementedError("不应到这里") # parse 之后的步骤未实现,到这里即说明守卫过了
monkeypatch.setattr(site_interact, "add_to_cart", _noop)
monkeypatch.setattr(site_interact, "verify_cart", _noop)
monkeypatch.setattr(site_interact, "enter_checkout", _checkout_html)
monkeypatch.setattr(site_interact, "parse_checkout", _parse)
monkeypatch.setattr(site_interact, "submit_order", _submit)
# 全局上限 30000,但 intent 给 5000 → 实际 8000 > 5000 → 拦截
await runner.handle(_make_task(task_id="t1", intent={"max_total_yen": 5000}))
assert submit_called is False
gateway: FakeGateway = runner._gateway_for_test # type: ignore[attr-defined]
terminal = gateway.last_terminal_report()
assert terminal["terminal_status"] == TaskStatus.NEEDS_HUMAN
# ---- 证据在 report 之前落盘(§9 第 11 条)----
async def test_evidence_files_exist_before_each_report(
runner: WorkerRunner, evidence: EvidenceStore, monkeypatch
):
"""每个 _run_step 调用:动作 → 落证据 → 写本地 → 回报 gateway
用一个 hook report 时机抓出来断言此时该 step meta.json 已在盘上
"""
seen_evidence_at_report: list[bool] = []
base = evidence.step_dir("t1")
real_report = runner._gateway_for_test.report # type: ignore[attr-defined]
async def _spy_report(*args, **kwargs): # noqa: ANN001, ANN002, ANN003
# 在 report 触发时检查 evidence_ref 指向的 meta 文件是否已落盘
ev_ref = kwargs.get("evidence_ref")
if ev_ref:
meta_path = base / (ev_ref.split("/", 1)[-1] + ".meta.json")
seen_evidence_at_report.append(meta_path.exists())
return await real_report(*args, **kwargs)
runner._gateway_for_test.report = _spy_report # type: ignore[attr-defined]
# 让 add_to_cart 正常,verify_cart 抛 NotImplemented,触发第一个 step 写证据 + report
async def _noop(task): # noqa: ANN001
return None
monkeypatch.setattr(site_interact, "add_to_cart", _noop)
await runner.handle(_make_task(task_id="t1"))
# 至少有一个 step 调了 report,且每次 report 之前证据都在
assert seen_evidence_at_report, "应当至少有一次带 evidence_ref 的 report"
assert all(seen_evidence_at_report), "某次 report 之前证据文件未落盘"
Generated
+11
View File
@@ -6,6 +6,15 @@ resolution-markers = [
"python_full_version < '3.14'",
]
[[package]]
name = "aiosqlite"
version = "0.22.1"
source = { registry = "https://mirrors.huaweicloud.com/repository/pypi/simple" }
sdist = { url = "https://mirrors.huaweicloud.com/repository/pypi/packages/4e/8a/64761f4005f17809769d23e518d915db74e6310474e733e3593cfc854ef1/aiosqlite-0.22.1.tar.gz", hash = "sha256:043e0bd78d32888c0a9ca90fc788b38796843360c855a7262a532813133a0650" }
wheels = [
{ url = "https://mirrors.huaweicloud.com/repository/pypi/packages/00/b7/e3bf5133d697a08128598c8d0abc5e16377b51465a33756de24fa7dee953/aiosqlite-0.22.1-py3-none-any.whl", hash = "sha256:21c002eb13823fad740196c5a2e9d8e62f6243bd9e7e4a1f87fb5e44ecb4fceb" },
]
[[package]]
name = "annotated-doc"
version = "0.0.4"
@@ -792,6 +801,7 @@ name = "rakuten-scraper-service"
version = "0.1.0"
source = { virtual = "." }
dependencies = [
{ name = "aiosqlite" },
{ name = "fastapi" },
{ name = "httpx", extra = ["http2"] },
{ name = "loguru" },
@@ -817,6 +827,7 @@ dev = [
[package.metadata]
requires-dist = [
{ name = "aiosqlite", specifier = ">=0.20.0,<1.0.0" },
{ name = "fastapi", specifier = ">=0.115.0,<1.0.0" },
{ name = "httpx", extras = ["http2"], specifier = ">=0.27.0,<1.0.0" },
{ name = "loguru", specifier = ">=0.7.0,<1.0.0" },