o2o-site-AEO/solution/backend/services/rollback_service.py

"""대상 버전 스냅샷 로드 → payload 재조립 → 렌더 → 게이트 → 공개 주소 전환"""
import os
import uuid

from common.database.db_session_manager import DB_SESSION_MNG
from common.database.model.models import places, site_publish_logs, site_versions, sites
from common.enums import (
    BuildStatus,
    DBWRType,
    ErrorType,
    PlaceCategory,
    PublishAction,
    PublishResult,
    SiteStatus,
)
from common.logger import LOG
from common.utils.gtime import GTime
from crud.place_crud import PlaceCRUD
from crud.site_crud import SiteCRUD
from services import alert_service, azure_static, indexnow, publish_gate, render_service, site_payload
from services.build_service import _actor_uuid, ensure_site, load_channel_links

_site_crud = SiteCRUD()
_place_crud = PlaceCRUD()

RENDER_TIMEOUT_SEC_ENV = "RENDER_TIMEOUT_SEC"


class RollbackAborted(RuntimeError):
    """재시도해도 소용없는 중단 — 잡의 last_error 로 남는다."""


async def _log(site_id, version_id, result: PublishResult, gate=None, actor=None):
    row = site_publish_logs(
        site_id=site_id,
        site_version_id=version_id,
        action=PublishAction.ROLLBACK.value,
        result=result.value,
        reject_reason=(gate.reason.value if gate is not None and gate.reason else None),
        detail=(gate.as_log() if gate is not None and not gate.passed else None),
        # ★ build_service 와 같은 함수를 쓴다 — 모양이 틀린 actor 하나가 되돌리기를
        #   죽이면 안 된다(그쪽 주석에 실측이 있다).
        actor_user_id=_actor_uuid(actor),
    )
    await DB_SESSION_MNG.execute_lambda_run([site_publish_logs.DBType()], [lambda s: _site_crud.add_log(s, row)])


async def run_rollback(job: dict) -> dict:
    """ROLLBACK 잡 핸들러."""
    payload = job["payload"]
    place_id = payload["place_id"]
    owner_user_id = payload["owner_user_id"]
    target_version = int(payload["target_version"])
    render_timeout_sec = float(os.environ.get(RENDER_TIMEOUT_SEC_ENV) or 180)

    err, place = await DB_SESSION_MNG.execute_lambda(
        places.DBType(),
        DBWRType.DB_READ.value,
        lambda s: _place_crud.get_place(s, uuid.UUID(owner_user_id), uuid.UUID(place_id)),
    )
    if err != ErrorType.SUCCESS or place is None:
        raise RollbackAborted(f"사업장을 찾을 수 없다: {place_id}")

    site = await ensure_site(place_id)

    v_err, version = await DB_SESSION_MNG.execute_lambda(
        site_versions.DBType(),
        DBWRType.DB_READ.value,
        lambda s: _site_crud.get_version_by_number(s, site.site_id, target_version),
    )
    if v_err != ErrorType.SUCCESS or version is None:
        raise RollbackAborted(f"버전 {target_version} 을 찾을 수 없다")
    if version.build_status != BuildStatus.BUILT.value or not version.snapshot:
        raise RollbackAborted(f"버전 {target_version} 은 성공적으로 빌드된 적이 없어 되돌릴 수 없다")
    if site.current_version_id == version.site_version_id:
        raise RollbackAborted(f"이미 지금 공개 중인 버전이다(v{target_version})")

    result = {
        "place_id": place_id, "site_id": str(site.site_id),
        "target_version": target_version, "site_version_id": str(version.site_version_id),
    }

    async def _fail(reason: str, gate: publish_gate.GateResult | None = None):
        await _log(site.site_id, version.site_version_id, PublishResult.REJECTED if gate else PublishResult.FAILED,
                   gate, payload.get("requested_by"))
        result["rolled_back"] = False
        result["error"] = reason
        LOG.w(f"[rollback] place={place_id} v{target_version} 실패: {reason}")
        # 게이트 반려는 알리지 않는다 — build_service._fail 과 같은 규칙(운영자를 부를 일이 아니다).
        if gate is None:
            await alert_service.send_alert(
                kind="build_failed",
                title=f"롤백 실패 — {place_id} → v{target_version}",
                detail=f"place_id={place_id} target_version={target_version}\n{reason}",
                dedupe_key=f"build_failed:{place_id}",
            )
        return result

    # snapshot 을 그대로 옮긴다 — 재수집하지 않는다(파일 머리주석 참조).
    links = await load_channel_links(place_id)
    payload_dict = await site_payload.prepare_site_payload(
        place, version.snapshot, site, version, links, publish=True
    )
    payload_path = site_payload.write_payload(payload_dict)
    result["payload_path"] = payload_path

    try:
        report = await render_service.render_site(payload_path, target_version, render_timeout_sec)
    except render_service.RenderFailed as ex:
        return await _fail(str(ex))

    mismatches = list(report.get("mismatches") or [])
    unique_count_raw = report.get("uniqueContentCount")

    # 렌더러가 그때와 달라졌을 수 있으므로 게이트를 다시 돈다(파일 머리주석 참조) — fact 검증 게이트(1차)는 건너뛴다.
    gate = publish_gate.evaluate(
        PlaceCategory(place.category), version.snapshot.get("facts") or [], unique_count_raw, mismatches
    )
    result["gate"] = {"passed": gate.passed, **gate.as_log()}
    if not gate.passed:
        return await _fail(f"{gate.reason.name}: {gate.as_log()}", gate)
    if not report.get("ok"):
        return await _fail(str(report.get("error") or "렌더 실패"))

    slug = site_payload.publish_slug(place, site)
    try:
        await render_service.activate_site(slug, target_version)
        await azure_static.publish(slug)
    except (render_service.RenderFailed, azure_static.AzurePublishError) as ex:
        return await _fail(str(ex))
    await indexnow.submit(slug)

    now = GTime.UTC()
    site_update = {
        "status": SiteStatus.PUBLISHED.value,
        "current_version_id": version.site_version_id,
        "published_at": now,
    }
    await DB_SESSION_MNG.execute_lambda_claim(
        sites.DBType(), lambda s: _site_crud.update_site(s, site.site_id, site_update)
    )
    await _log(site.site_id, version.site_version_id, PublishResult.SUCCESS, None, payload.get("requested_by"))
    await alert_service.resolve_alert(f"build_failed:{place_id}", f"롤백 성공 — {place_id} → v{target_version}")

    result["rolled_back"] = True
    LOG.i(f"[rollback] place={place_id} v{target_version} 로 되돌림")
    return result