"""Scheduled jobs that own the DNS-request execution plane.

Before 2026-08-14 the cloud write ran inline inside the Feishu approval
webhook. That coupled a multi-second provider call to a callback Feishu
gives ~3 seconds before it redelivers, and — worse — it made the request
handler the only thing that could ever move a row out of ``executing``.
When one of those handlers died mid-write (DNSPod rejected the TTL, the
fail-path then blew up on an expired ORM attribute), the request sat at
``executing`` forever: the requester saw 处理中 indefinitely, no audit row,
no notification, and no admin button that would touch it.

Two crons replace that single point of failure:

  ``cmdb.dns_request_executor`` (every minute)
      Drains rows the approval callback left at ``approved`` and performs
      the provider write in the worker, where nothing is redelivering and
      no HTTP timeout applies.

  ``cmdb.dns_request_execution_reaper`` (every 10 minutes)
      Sweeps rows stuck at ``executing`` past ``stale_minutes``. ``create``
      is reconciled against the live provider first — if the record is
      already there, the interrupted write DID land and the row is
      completed rather than re-driven into a duplicate. Everything else is
      idempotent and simply re-armed to ``approved``. Rows that exhaust
      ``max_attempts`` land in ``failed`` with a message the admin resume
      gate recognises.
"""
from __future__ import annotations

import logging
from typing import Any

from sqlalchemy.ext.asyncio import AsyncSession

from app.domains.cmdb.dns_request_service import DnsRequestService
from app.scheduled_jobs.registry import register_scheduled_job

logger = logging.getLogger(__name__)


@register_scheduled_job(
    key="cmdb.dns_request_executor",
    display_name="DNS 申请执行器",
    description=(
        "每分钟扫描 status='approved' 的 DNS 解析申请并执行云侧写入"
        "(腾讯云 DNSPod / Route53 等),成功置 executed,provider 报错置"
        " failed 并回写飞书评论。审批回调本身不再做云写:回调只落"
        " approved 后立即返回,避免飞书 3 秒超时重投与请求被中断导致"
        "行卡在 executing。config: batch_limit(默认 50)。"
    ),
)
async def run_dns_request_executor(
    db: AsyncSession, config: dict[str, Any],
) -> dict[str, Any]:
    """Drain the ``approved`` backlog.

    Returns ``checked`` / ``executed`` / ``failed`` / ``errors``.
    ``failed`` is a normal outcome (the provider rejected the write and
    the row now says so); ``errors`` is not — it counts rows that threw
    out of the executor itself and is what should be watched.
    """
    batch_limit = int(config.get("batch_limit") or 50)
    summary = await DnsRequestService(db).execute_approved_backlog(
        batch_limit=batch_limit,
    )
    # A tick where every row threw is a systematic break (credentials
    # rotated, provider unreachable, a regression in the executor), not
    # bad luck. Raise so the failure notifier pages instead of letting a
    # green cron hide a fully broken execution plane.
    if summary["checked"] and summary["errors"] == summary["checked"]:
        raise RuntimeError(
            f"cmdb.dns_request_executor: all {summary['checked']} rows "
            "raised out of the executor"
        )
    return summary


@register_scheduled_job(
    key="cmdb.dns_request_execution_reaper",
    display_name="DNS 申请执行卡死回收",
    description=(
        "每 10 分钟扫描 status='executing' 且超过 stale_minutes(默认 10)"
        "未更新的 DNS 申请——这些行的执行进程已经不在了。create 先对账"
        "云端:记录已存在说明写入其实成功,补齐 Helix 侧台账并置 executed"
        "(绝不重复创建);其余动作幂等,直接回退 approved 交给执行器重跑。"
        "重试超过 max_attempts(默认 3)则置 failed 并写入可 resume 的错误"
        "信息,管理员在申请详情里点「重跑」即可,无需重新发起审批。"
        "config: stale_minutes、max_attempts、batch_limit(默认 50)。"
    ),
)
async def run_dns_request_execution_reaper(
    db: AsyncSession, config: dict[str, Any],
) -> dict[str, Any]:
    """Un-strand rows parked at ``executing``.

    Returns ``checked`` / ``healed`` (cloud write had landed, bookkeeping
    completed) / ``rearmed`` / ``exhausted`` (gave up, now resumable) /
    ``errors``.
    """
    summary = await DnsRequestService(db).reconcile_stranded_executions(
        stale_minutes=int(config.get("stale_minutes") or 10),
        max_attempts=int(config.get("max_attempts") or 3),
        batch_limit=int(config.get("batch_limit") or 50),
    )
    if summary["checked"] and summary["errors"] == summary["checked"]:
        raise RuntimeError(
            f"cmdb.dns_request_execution_reaper: all {summary['checked']} "
            "rows failed to reconcile"
        )
    if summary["exhausted"]:
        # Loud on purpose: an exhausted row means an approved DNS change
        # never reached the provider and now needs a human.
        logger.warning(
            "cmdb.dns_request_execution_reaper: %s request(s) gave up after "
            "repeated interrupted executions — awaiting an admin resume",
            summary["exhausted"],
        )
    return summary