feat(report): orchestration skeleton with staged pipeline and refund-on-failure
POST /reports 骨架：资格校验→终身配额预扣→tk_task→cpu 队列；动态装配 流水线（dotted-path 注册、缺阶段即 CONFIG_ERROR）；中间态逐段落库、失败 退款、derived_features 跨重试/重生成复用（T-P3-07/09）。 实现：packages/report 的 pipeline.py（Stage 契约保持不变，新增 assemble_stages/ run_pipeline/StageFailureError，ReportContext 增可选 task_id 参数、向后兼容）、 repo.py、schemas.py、service.py（重放优先于资格校验、reserve+建行+建任务同事务、 提交后才投递、并发同键落败回滚）；apps/api/routers/reports.py 填桩（send_task 按 任务名投递，不 import apps.worker）；apps/worker/report_tasks.py（worker 侧自建 gateway 装配、状态归一 QUEUED→RUNNING、成功 commit 配额+成本聚合写 model_info、 基础设施异常重试一次、重试耗尽兜底失败退款）。测试：单测 6 个（装配缺失/空配置/ 协议违约/顺序执行/失败保留产物），集成 6 场景（预扣 lifetime/PENDING、成功 committed+两阶段键、同 request_key 幂等不重复扣、二次首生成 40920、阶段失败 refund+derived_features 保留、regenerate 走 m: 月配额 version+1 复用视觉中间态）。 偏差说明： 1) apps/worker/celery_app.py 的 include 只含 apps.worker.tasks（禁改文件），真 worker 进程注册 meida.report.generate 需协调者合并时把 include 补上 apps.worker.report_tasks；API 侧已按任务名 send_task，两侧只靠名字衔接，测试 不受影响。2) ReportContext 增加可选 task_id 关键字参数（失败结案要推动 tk_task），Stage 协议/DEFAULT_STAGE_PATHS/stage_names 未动，T16/T17 兼容。 3) JSONB 深变更用 flag_modified 显式标脏（重新赋相等新 dict 会被 flush 的按值 比较跳过，集成测试抓到过该坑）。4) POST 后 tk_task 保持 PENDING（任务文档 §4 场景 1 口径），QUEUED→RUNNING 由 worker 入口归一。
This commit is contained in:
+181
-6
@@ -1,9 +1,10 @@
|
||||
"""报告流水线的阶段契约（第四批 T15/T16/T17 的共同接口）。
|
||||
"""报告流水线的阶段契约与编排（第四批 T15/T16/T17 的共同接口）。
|
||||
|
||||
本文件由协调提交只定 **接口**；编排实现（assemble_stages / run_pipeline）
|
||||
归 T15。T16 提供 vision_stage / diagnosis_stage，T17 提供 copy_stage /
|
||||
validation_stage——三个任务互不改对方文件，靠 Settings.report_pipeline_stages
|
||||
里的 dotted path 动态装配。
|
||||
阶段契约（``ReportContext`` / ``Stage`` / ``DEFAULT_STAGE_PATHS``）由协调提交
|
||||
定死，T15 实现编排（``assemble_stages`` / ``run_pipeline``）。T16 提供
|
||||
vision_stage / diagnosis_stage，T17 提供 copy_stage / validation_stage——
|
||||
三个任务互不改对方文件，靠 Settings.report_pipeline_stages 里的 dotted
|
||||
path 动态装配。
|
||||
|
||||
设计动机（为什么动态 import）：三个阶段模块由三个并行分支各自创建，
|
||||
静态 import 会在任一分支缺文件时炸掉编排；按配置动态装配让每个分支
|
||||
@@ -13,13 +14,24 @@ validation_stage——三个任务互不改对方文件，靠 Settings.report_pi
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import importlib
|
||||
from typing import Any, Protocol, runtime_checkable
|
||||
|
||||
from sqlalchemy.ext.asyncio import AsyncSession
|
||||
from sqlalchemy.orm.attributes import flag_modified
|
||||
|
||||
from packages.billing import repo as billing_repo
|
||||
from packages.billing.service import BillingService, Reservation
|
||||
from packages.domain.task.state import TaskStatus
|
||||
from packages.llm.gateway import LLMGateway
|
||||
from packages.shared.config import Settings
|
||||
from packages.shared.models import RpInput, RpReport
|
||||
from packages.shared.errors import AppError, ErrorCode
|
||||
from packages.shared.logging import get_logger
|
||||
from packages.shared.models import Ledger, RpInput, RpReport
|
||||
from packages.task import repo as task_repo
|
||||
from packages.task.service import TaskService
|
||||
|
||||
logger = get_logger(__name__)
|
||||
|
||||
|
||||
class ReportContext:
|
||||
@@ -37,12 +49,16 @@ class ReportContext:
|
||||
session: AsyncSession,
|
||||
settings: Settings,
|
||||
gateway: LLMGateway | None = None,
|
||||
task_id: int | None = None,
|
||||
) -> None:
|
||||
self.report = report
|
||||
self.report_input = report_input
|
||||
self.session = session
|
||||
self.settings = settings
|
||||
self.gateway = gateway
|
||||
# 编排层专用：失败结案要推动 tk_task。阶段实现不需要感知它，
|
||||
# 保持可选以保证旧构造调用（T16/T17 的测试）不受影响。
|
||||
self.task_id = task_id
|
||||
|
||||
|
||||
@runtime_checkable
|
||||
@@ -70,3 +86,162 @@ DEFAULT_STAGE_PATHS: tuple[str, ...] = (
|
||||
def stage_names(result: dict[str, Any]) -> list[str]:
|
||||
"""result JSONB 里已出现的阶段键（诊断/调试用的小工具）。"""
|
||||
return [k for k in result if isinstance(result[k], dict)]
|
||||
|
||||
|
||||
class StageFailureError(Exception):
|
||||
"""阶段异常且已按失败结案（rp_report failed + tk_task failed + 退款已提交）。
|
||||
|
||||
celery 包装层据此区分两类异常：``StageFailureError`` 是业务性失败，重跑
|
||||
也不会好，不再重试；其它异常（DB/网络/装配）发生在结算之外，值得用
|
||||
重试换恢复。
|
||||
"""
|
||||
|
||||
def __init__(self, stage_name: str, original: BaseException) -> None:
|
||||
self.stage_name = stage_name
|
||||
self.original = original
|
||||
super().__init__(f"stage={stage_name} error={type(original).__name__}: {original}")
|
||||
|
||||
|
||||
def assemble_stages(settings: Settings) -> list[Stage]:
|
||||
"""按 ``settings.report_pipeline_stages`` 的 dotted path 动态装配阶段列表。
|
||||
|
||||
动态 import 是刻意的插件式装配（见模块 docstring）：T16/T17 各自建阶段
|
||||
模块，不碰本文件；import-linter 看不到这条边，阶段模块自身的依赖仍受
|
||||
分层契约约束，这里只做装配不做业务。
|
||||
|
||||
模块/类不存在、实例化失败、或不符合 Stage 协议时，收集**全部**缺失项
|
||||
抛 ``CONFIG_ERROR`` —— 绝不静默跳过：静默等于流水线空跑还扣用户配额。
|
||||
"""
|
||||
if not settings.report_pipeline_stages:
|
||||
raise AppError(
|
||||
ErrorCode.CONFIG_ERROR,
|
||||
message="报告流水线阶段配置为空",
|
||||
data={"missing": list(DEFAULT_STAGE_PATHS)},
|
||||
)
|
||||
|
||||
missing: list[str] = []
|
||||
stages: list[Stage] = []
|
||||
for dotted in settings.report_pipeline_stages:
|
||||
module_path, sep, attr = dotted.partition(":")
|
||||
stage: Any = None
|
||||
if sep and module_path and attr:
|
||||
try:
|
||||
# 阶段构造约定无参。__init__ 抛错视同配置坏了（比如阶段自身
|
||||
# 校验它的依赖），归入 CONFIG_ERROR 而不是让 worker 崩溃。
|
||||
stage = getattr(importlib.import_module(module_path), attr)()
|
||||
except Exception:
|
||||
stage = None
|
||||
if stage is None or not isinstance(stage, Stage):
|
||||
missing.append(dotted)
|
||||
else:
|
||||
stages.append(stage)
|
||||
|
||||
if missing:
|
||||
raise AppError(
|
||||
ErrorCode.CONFIG_ERROR,
|
||||
message="报告流水线阶段装配失败：缺失或不可用",
|
||||
data={"missing": missing},
|
||||
)
|
||||
return stages
|
||||
|
||||
|
||||
def reservation_from_ledger(ledger: Ledger) -> Reservation:
|
||||
"""从 reserved 流水重建 Reservation（worker 成功 commit / 失败 refund 都要它）。"""
|
||||
return Reservation(
|
||||
tx_id=ledger.tx_id,
|
||||
user_id=ledger.user_id,
|
||||
quota_key=ledger.quota_key or "",
|
||||
period_key=ledger.period_key or "",
|
||||
amount=abs(ledger.delta),
|
||||
remaining=ledger.balance_after,
|
||||
idempotency_key=ledger.idempotency_key,
|
||||
)
|
||||
|
||||
|
||||
async def load_reservation(session: AsyncSession, task_id: int) -> Reservation | None:
|
||||
"""按 ``tk_task.reserve_tx_id`` 取回预扣；无预扣或流水缺失返回 None。"""
|
||||
task = await task_repo.get(session, task_id)
|
||||
if task is None or task.reserve_tx_id is None:
|
||||
return None
|
||||
ledger = await billing_repo.get_ledger(session, task.reserve_tx_id)
|
||||
if ledger is None:
|
||||
return None
|
||||
return reservation_from_ledger(ledger)
|
||||
|
||||
|
||||
async def run_pipeline(ctx: ReportContext, stages: list[Stage]) -> None:
|
||||
"""逐阶段执行，每阶段产物即刻 flush 落库（T-P3-09 中间态持久化）。
|
||||
|
||||
中间态的意义：阶段半程失败或 worker 崩溃后重试，已完成阶段（尤其
|
||||
``rp_input.derived_features`` 里的视觉结果）不必重算——「失败不需重新
|
||||
上传照片」的结构保证就落在这几次 flush 里。
|
||||
|
||||
阶段异常 → ``_settle_stage_failure``（事务纪律见其 docstring）后包成
|
||||
:class:`StageFailureError` 上抛；正常返回即全部阶段成功，成功结案归调用方
|
||||
（worker：commit 配额 + succeeded + 成本聚合）。
|
||||
"""
|
||||
# 防御：default=dict 只在 INSERT 时生效，测试/手工构造的瞬态对象可能为 None。
|
||||
if ctx.report.result is None:
|
||||
ctx.report.result = {}
|
||||
if ctx.report_input.derived_features is None:
|
||||
ctx.report_input.derived_features = {}
|
||||
|
||||
for stage in stages:
|
||||
try:
|
||||
await stage.run(ctx)
|
||||
except Exception as exc:
|
||||
await _settle_stage_failure(ctx, stage.name, exc)
|
||||
raise StageFailureError(stage.name, exc) from exc
|
||||
# JSONB 列没有 Mutable 追踪：阶段对 dict 的深变更 ORM 看不见。且不能
|
||||
# 靠「重新赋一个相等的新 dict」——flush 按值比较会认为没变而跳过
|
||||
# UPDATE。flag_modified 强制该列进入下一条 UPDATE（中间态逐段落库
|
||||
# 的关键）。derived_features 同理。
|
||||
ctx.report.result.setdefault(stage.name, {})
|
||||
flag_modified(ctx.report, "result")
|
||||
flag_modified(ctx.report_input, "derived_features")
|
||||
await ctx.session.flush()
|
||||
logger.info(
|
||||
"report_stage_done",
|
||||
report_id=ctx.report.report_id,
|
||||
task_id=ctx.task_id,
|
||||
stage=stage.name,
|
||||
)
|
||||
|
||||
|
||||
async def _settle_stage_failure(ctx: ReportContext, stage_name: str, exc: Exception) -> None:
|
||||
"""阶段失败的结案：report failed + tk_task failed + 退款，提交后才返回。
|
||||
|
||||
事务纪律（对齐 AgentService）：退款必须先于异常上抛提交，否则外层回滚
|
||||
会把退款一起卷走（FR-BILL-05）。结案本身失败只记日志、不吞原始异常——
|
||||
此时额度挂在 reserved 状态，靠流水对账兜底。
|
||||
"""
|
||||
summary = {"stage": stage_name, "type": type(exc).__name__, "message": str(exc)[:200]}
|
||||
try:
|
||||
ctx.report.status = "failed"
|
||||
ctx.report.result["error"] = summary
|
||||
flag_modified(ctx.report, "result") # 深变更必须显式标脏，见 run_pipeline 的注释
|
||||
if ctx.task_id is not None:
|
||||
await TaskService(ctx.session).transition(
|
||||
task_id=ctx.task_id,
|
||||
target=TaskStatus.FAILED,
|
||||
stage_text=f"阶段失败：{stage_name}",
|
||||
error_message=f"{type(exc).__name__}: {exc}",
|
||||
)
|
||||
reservation = await load_reservation(ctx.session, ctx.task_id)
|
||||
if reservation is not None:
|
||||
await BillingService(ctx.session).refund(
|
||||
reservation, reason=f"report_stage_failed:{stage_name}"
|
||||
)
|
||||
await ctx.session.commit()
|
||||
logger.warning(
|
||||
"report_stage_failed",
|
||||
report_id=ctx.report.report_id,
|
||||
task_id=ctx.task_id,
|
||||
stage=stage_name,
|
||||
error_type=type(exc).__name__,
|
||||
)
|
||||
except Exception:
|
||||
await ctx.session.rollback()
|
||||
logger.exception(
|
||||
"report_failure_settle_failed", report_id=ctx.report.report_id, stage=stage_name
|
||||
)
|
||||
|
||||
Reference in New Issue
Block a user