AiDOPWarehouse/server/Plugins/Admin.NET.Plugin.AiDOP/Service/S8/S8TimeoutAutoEscalationService.cs

using Admin.NET.Plugin.AiDOP.Const.S8;
using Admin.NET.Plugin.AiDOP.Entity.S8;
using Admin.NET.Plugin.AiDOP.Infrastructure;
using Microsoft.Extensions.Logging;

namespace Admin.NET.Plugin.AiDOP.Service.S8;

/// <summary>
/// S8-TIMEOUT-AUTO-ESCALATION-JOB-1（P4-1）：扫描 sla_deadline 已超时且未关闭/未已升级的异常，
/// 通过 <see cref="S8TaskFlowService.UpgradeAsync"/> 启动 EXCEPTION_ESCALATION ApprovalFlow，与人工升级链路 100% 等价。
///
/// 设计要点：
/// - 不依赖 timeout_flag；扫描公式与读端 IsCurrentlyTimeout 一致：sla_deadline IS NOT NULL AND sla_deadline &lt; now
///   AND status NOT IN ('CLOSED','RECOVERED','ESCALATED')。
/// - status 进一步限制在 ASSIGNED / IN_PROGRESS（与 <see cref="Infrastructure.S8.S8StatusRules"/> 允许 ESCALATED 的转移一致）；
///   NEW / PENDING_VERIFICATION / REJECTED 等状态不通过本 Job 自动升级。
/// - 已有 active_flow_instance_id 的异常跳过（防重）；UpgradeAsync 内部还会再校验一次，双层保险。
/// - exception_type.escalate_role_code 为空 / 非法时跳过并 LogInformation；不写脏数据，不补默认。
/// - 状态 / Timeline 由 <see cref="ExceptionEscalationBizHandler"/>.OnFlowStarted 写入；本服务不重复维护。
/// - 通知层走 <see cref="S8NotificationLayerResolver"/>；当前 baseline notify_channel="log"，无外部副作用。
/// </summary>
public class S8TimeoutAutoEscalationService : ITransient
{
    private readonly SqlSugarRepository<AdoS8Exception> _rep;
    private readonly SqlSugarRepository<AdoS8ExceptionType> _typeRep;
    private readonly S8TaskFlowService _taskFlow;
    private readonly S8NotificationLayerResolver _layerResolver;
    private readonly ILogger<S8TimeoutAutoEscalationService> _logger;

    public S8TimeoutAutoEscalationService(
        SqlSugarRepository<AdoS8Exception> rep,
        SqlSugarRepository<AdoS8ExceptionType> typeRep,
        S8TaskFlowService taskFlow,
        S8NotificationLayerResolver layerResolver,
        ILogger<S8TimeoutAutoEscalationService> logger)
    {
        _rep = rep;
        _typeRep = typeRep;
        _taskFlow = taskFlow;
        _layerResolver = layerResolver;
        _logger = logger;
    }

    /// <summary>
    /// 扫描一次。返回成功触发升级的异常数量；调用方负责调度 / 限流。
    /// </summary>
    public async Task<int> RunOnceAsync(int batchSize = 50, CancellationToken ct = default)
    {
        var now = DateTime.Now;
        var candidates = await _rep.AsQueryable()
            .Where(x => !x.IsDeleted
                && x.SlaDeadline != null
                && x.SlaDeadline < now
                && (x.Status == "ASSIGNED" || x.Status == "IN_PROGRESS")
                && (x.ActiveFlowInstanceId == null || x.ActiveFlowInstanceId == 0))
            .OrderBy(x => x.SlaDeadline)
            .Take(batchSize)
            .ToListAsync();

        if (candidates.Count == 0) return 0;

        // exception_type.escalate_role_code 批量映射（本租户覆盖优先于平台默认）
        var typeCodes = candidates
            .Select(c => c.ExceptionTypeCode)
            .Where(c => !string.IsNullOrWhiteSpace(c))
            .Distinct()
            .Select(c => c!)
            .ToList();
        // S8-P0-1-SCHEDULER-TRUSTED-SCOPE-1：原实现按 TypeCode 全库取「factory_id 最大」的一行，
        // 无租户/工厂谓词 —— 本 Job 是跨租户扫描，会把 B 租户的 escalate_role_code 用到 A 租户的异常上。
        // 现改为逐异常按其自身 TenantId 解析：候选 = 本租户覆盖 ∪ 平台默认 (tenant_id = 0)。
        //
        // S8-TENANT-ONLY-BATCH6：precedence 由 OrderByDescending(FactoryId) 改为「租户行优先」。
        // 本批之后租户覆盖行的 factory_id 恒为 0，与平台默认相等，按 factory 排序会退化成任意序，
        // 可能拿平台默认的 escalate_role_code 去升级一条本租户已自定义的异常。
        var typeRows = typeCodes.Count == 0
            ? new List<AdoS8ExceptionType>()
            : (await _typeRep.AsQueryable().ClearFilter()
                .Where(t => typeCodes.Contains(t.TypeCode))
                .ToListAsync())
                .OrderByDescending(t => t.TenantId != S8ConfigScope.GlobalTenantId)
                .ToList();

        AdoS8ExceptionType? ResolveTypeForScope(string typeCode, long tenantId) =>
            typeRows.FirstOrDefault(t => t.TypeCode == typeCode
                                      && (t.TenantId == tenantId || t.TenantId == S8ConfigScope.GlobalTenantId));

        var processed = 0;
        foreach (var e in candidates)
        {
            ct.ThrowIfCancellationRequested();
            var type = string.IsNullOrWhiteSpace(e.ExceptionTypeCode)
                ? null
                : ResolveTypeForScope(e.ExceptionTypeCode!, e.TenantId);
            if (type == null || string.IsNullOrWhiteSpace(type.EscalateRoleCode))
            {
                _logger.LogInformation(
                    "s8_timeout_auto_escalate_skip exceptionId={Id} exceptionCode={Code} reason=escalate_role_empty typeCode={TypeCode}",
                    e.Id, e.ExceptionCode, e.ExceptionTypeCode);
                continue;
            }

            try
            {
                // UpgradeAsync 已内置 ActiveFlowInstanceId / IsAllowedTransition 二次校验；
                // 与 manual upgrade 100% 等价，状态/timeline 由 OnFlowStarted 写入。
                var remark = $"[AUTO] SLA deadline exceeded; auto escalation triggered. sla_deadline={e.SlaDeadline:yyyy-MM-dd HH:mm:ss}; escalate_role_code={type.EscalateRoleCode}";
                await _taskFlow.UpgradeAsync(e.Id, e.TenantId, remark);
                processed++;
                _logger.LogInformation(
                    "s8_timeout_auto_escalate_started exceptionId={Id} exceptionCode={Code} typeCode={TypeCode} escalateRoleCode={Role}",
                    e.Id, e.ExceptionCode, e.ExceptionTypeCode, type.EscalateRoleCode);

                await TryDispatchAsync(e);
            }
            catch (Exception ex)
            {
                _logger.LogWarning(ex,
                    "s8_timeout_auto_escalate_failed exceptionId={Id} status={Status}", e.Id, e.Status);
            }
        }

        if (processed > 0 || candidates.Count > 0)
            _logger.LogInformation(
                "s8_timeout_auto_escalate_summary processed={Processed} candidates={Total}", processed, candidates.Count);
        return processed;
    }

    private async Task TryDispatchAsync(AdoS8Exception e)
    {
        try
        {
            await _layerResolver.DispatchByLayerAsync(new S8NotificationLayerResolver.DispatchByLayerInput
            {
                TenantId = e.TenantId,
                ExceptionId = e.Id,
                ExceptionNo = e.ExceptionCode,
                // 优先 module_code（S1-S7 严格基线），保持与 NotificationLayer baseline 同口径。
                SceneCode = string.IsNullOrWhiteSpace(e.ModuleCode) ? e.SceneCode : e.ModuleCode!,
                Severity = e.Severity,
                Status = "ESCALATED",
                EventCode = S8NotifyEventCode.EscalationTriggered,
                ExceptionRef = e,
                Title = $"[AUTO] 异常升级 - {e.ExceptionCode}",
                Content = "SLA 已超时，系统自动触发升级。",
                SourceRuleCode = e.SourceRuleCode,
            });
        }
        catch (Exception ex)
        {
            _logger.LogWarning(ex, "s8_timeout_auto_escalate_dispatch_failed exceptionId={Id}", e.Id);
        }
    }
}