qlib/backend/app/api/jobs.py

"""异步研究 Job API（Phase 4）：提交 / 查询 / SSE 进度。

POST /api/jobs          创建 Job（BackgroundTasks 后台执行），立即返回 job_id
GET  /api/jobs/{id}     状态 + 结果（成功时内嵌 result）
GET  /api/jobs/{id}/events  SSE 进度（queued→running→success|failed）

执行模式见 config job.mode：subprocess 时研究任务在独立子进程跑（内存隔离），
API worker 不被重任务拖垮（内存优化专项）。
"""

from __future__ import annotations

import asyncio
import json
from datetime import datetime
from typing import Annotated

from fastapi import APIRouter, BackgroundTasks, HTTPException, Query
from fastapi.responses import StreamingResponse

from app.api.deps import DbSession, ExperimentRepoDep, JobRepoDep
from app.application.services.job_executor import new_id, run_job_background, terminate_active
from app.domain.entities.research import (
    BacktestResult,
    FactorTestReport,
    JobRecord,
    JobStatus,
    ResearchSpec,
)
from app.infrastructure.persistence.sqlalchemy.repositories.jobs_impl import (
    SqlAlchemyJobRepository,
)
from app.infrastructure.persistence.sqlalchemy.session import SessionLocal

router = APIRouter(prefix="/jobs", tags=["jobs"])


def _decode_result(kind: str, result_json: str | None):
    from app.domain.entities.selection import SelectionResult

    if result_json is None:
        return None
    if kind == "selection":
        return SelectionResult.model_validate_json(result_json)
    model = BacktestResult if kind == "backtest" else FactorTestReport
    return model.model_validate_json(result_json)


def _job_view(job: JobRecord, experiment_repo=None) -> dict:
    """Job 视图：`result` 契约不变（成功时内嵌**完整**结果）。

    结果来源（2026-09 起完整结果只在 experiment 存一份，job.result_json 不再重复写）：
    1. `job.experiment_id` 有值且能读到归档 → 解码 experiment.result_json；
    2. 否则回退解码 `job.result_json`（老记录 / 归档被删除前的历史数据）；
    3. 归档被删除且 job 侧无副本 → `result=None`，并给出
       `result_unavailable_reason` 如实说明原因（AGENT §7：不静默给空结果）。
    """
    view = job.model_dump()
    view.pop("result_json", None)
    view["spec"] = json.loads(job.spec_json)

    result = None
    source = None
    if job.experiment_id and experiment_repo is not None:
        exp = experiment_repo.get(job.experiment_id)
        if exp is not None:
            result = _decode_result(job.kind, exp.result_json)
            source = "experiment"
        else:
            view["result_unavailable_reason"] = (
                f"归档 {job.experiment_id} 已不存在（可能已被删除）；"
                "完整结果仅存于归档，Job 记录本身不再保存结果副本"
            )
    if result is None and job.result_json:
        result = _decode_result(job.kind, job.result_json)
        source = "job"
    view["result"] = result
    view["result_source"] = source
    return view


@router.post("", summary="创建异步研究 Job")
def create_job(
    spec: ResearchSpec,
    background: BackgroundTasks,
    session: DbSession,
    job_repo: JobRepoDep,
) -> dict:
    job = JobRecord(
        id=new_id("JOB"),
        kind=spec.type,
        spec_json=spec.model_dump_json(),
        status=JobStatus.QUEUED,
        created_at=datetime.now(),
    )
    job_repo.create(job)
    session.commit()
    background.add_task(run_job_background, job.id)
    return {"job_id": job.id, "status": job.status}


@router.get("", summary="Job 列表")
def list_jobs(
    job_repo: JobRepoDep,
    experiment_repo: ExperimentRepoDep,
    kind: Annotated[str | None, Query(description="按类型过滤（backtest/factor_test）")] = None,
    limit: Annotated[int, Query(ge=1, le=200)] = 20,
) -> list[dict]:
    return [
        _job_view(j, experiment_repo) for j in job_repo.list_recent(kind=kind, limit=limit)
    ]


@router.post("/{job_id}/cancel", summary="取消 Job（queued/running）")
def cancel_job(job_id: str, session: DbSession, job_repo: JobRepoDep) -> dict:
    job = job_repo.get(job_id)
    if job is None:
        raise HTTPException(status_code=404, detail=f"Job {job_id} 不存在")
    if job.status not in (JobStatus.QUEUED, JobStatus.RUNNING):
        return {"job_id": job_id, "status": job.status, "cancelled": False}
    job.status = JobStatus.CANCELLED
    job.stage = None
    job_repo.update(job)
    session.commit()
    terminate_active(job_id)  # 终止研究子进程（若有）；父进程兜底已跳过 CANCELLED
    return {"job_id": job_id, "status": JobStatus.CANCELLED, "cancelled": True}


@router.get("/{job_id}", summary="查询 Job 状态与结果")
def get_job(job_id: str, job_repo: JobRepoDep, experiment_repo: ExperimentRepoDep) -> dict:
    job = job_repo.get(job_id)
    if job is None:
        raise HTTPException(status_code=404, detail=f"Job {job_id} 不存在")
    return _job_view(job, experiment_repo)


@router.get("/{job_id}/events", summary="Job 进度 SSE")
async def job_events(job_id: str) -> StreamingResponse:
    async def gen():
        while True:
            with SessionLocal() as session:
                job = SqlAlchemyJobRepository(session).get(job_id)
            if job is None:
                yield "event: error\ndata: job not found\n\n"
                return
            payload = json.dumps(
                {"job_id": job.id, "status": job.status, "stage": job.stage}, ensure_ascii=False
            )
            yield f"data: {payload}\n\n"
            if job.status in (JobStatus.SUCCESS, JobStatus.FAILED, JobStatus.CANCELLED):
                return
            await asyncio.sleep(0.4)

    return StreamingResponse(gen(), media_type="text/event-stream")