| 123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196 |
- using System.Text.Json;
- using Admin.NET.Plugin.AiDOP.Entity.S8;
- using Admin.NET.Plugin.AiDOP.Infrastructure.S8;
- using Admin.NET.Plugin.AiDOP.Service.S8.Rules.DataAccess;
- using Admin.NET.Plugin.AiDOP.Service.S8.Rules.Definitions;
- using Microsoft.Extensions.Logging;
- namespace Admin.NET.Plugin.AiDOP.Service.S8.Rules;
- /// <summary>
- /// TIMEOUT 类规则 evaluator。
- /// 判定:dueAt <= now - graceMinutes 且 status 不在 completedStates 内 → HIT。
- /// 不做严重度阶梯、不做 SLA 升级、不做事件触发。
- ///
- /// S8-DATASET-PROVIDER-FOUNDATION-1:取数由 <see cref="S8MonitoringDataGateway"/> 承担,
- /// 本类不直接依赖 DataTable / 数据源 / SQL。
- ///
- /// <para><b>S8-RULE-GOVERNANCE-BATCH1:判定语义改由代码定义供给。</b>
- /// 此前 <c>dueAtField</c> / <c>statusField</c> / <c>completedStates</c> / <c>objectIdField</c> /
- /// <c>exceptionTypeCode</c> 全部从 <c>params_json</c> 解析 —— 而 <c>params_json</c> 是配置页可整块
- /// 覆写的自由文本。于是「这条规则判什么」实际由业务用户决定,改一个字符串就换一套业务含义。
- /// 现在这些一律取自 <see cref="S8RuleDefinition"/>(<see cref="IS8RuleCatalog"/> 供给),
- /// params_json 只剩 graceMinutes 等真正的运行策略。</para>
- ///
- /// <para>连带修复:params_json 为 NULL 不再抛 <c>rule_not_configured</c> —— 定义在代码里,
- /// 一条没有运行参数的规则完全可以按默认值跑。</para>
- /// </summary>
- public class S8TimeoutRuleEvaluator : IS8RuleEvaluator, ITransient
- {
- public const string RuleTypeCode = "TIMEOUT";
- public string RuleType => RuleTypeCode;
- private readonly S8MonitoringDataGateway _dataGateway;
- private readonly IS8RuleCatalog _ruleCatalog;
- private readonly ILogger<S8TimeoutRuleEvaluator> _logger;
- public S8TimeoutRuleEvaluator(
- S8MonitoringDataGateway dataGateway,
- IS8RuleCatalog ruleCatalog,
- ILogger<S8TimeoutRuleEvaluator> logger)
- {
- _dataGateway = dataGateway;
- _ruleCatalog = ruleCatalog;
- _logger = logger;
- }
- public async Task<List<S8RuleHit>> EvaluateAsync(
- long tenantId,
- AdoS8WatchRule rule,
- CancellationToken cancellationToken = default)
- {
- // R5 evaluator 失败语义保护:所有"非命中判定"路径均改为抛出 S8RuleEvaluatorException,
- // 由 SchedulerService 标记 evaluate_failed 并跳过 recovery reconcile,避免对未确认未命中的 rule 误标 recovered_at。
- //
- // S8-RULE-GOVERNANCE-BATCH1:第一道门是「这条规则在当前代码版本里有没有定义」。
- // 没有定义 = 它的判定语义无处可取,绝不允许猜(历史实现会去 params_json 里猜,
- // 于是任何人建一条规则、写一段 JSON 就能让调度器替他跑)。
- var effective = ResolveEffective(rule);
- // S8-SQL-EVALUATOR-GUARD-P2-1:每次评估解析 maxRows(env 优先,回退代码默认)。
- // S8-STANDARD-DATASET-HARD-CUTOVER-1:不再解析 commandTimeout —— S8 已不执行 SQL,
- // 超时归 Provider 自己的取数实现负责。
- var maxRows = S8EvaluatorGuard.ResolveMaxRows(_logger);
- var data = await _dataGateway.LoadAsync(
- tenantId, rule, RuleTypeCode, maxRows, cancellationToken);
- return EvaluateRows(data.RowSet, effective, tenantId, DateTime.Now);
- }
- /// <summary>
- /// 解析生效规则:定义必须存在、类型必须匹配、TIMEOUT 语义必须齐备。
- /// 三者任一不满足都是**发布态问题**,不是数据问题,故 fail-fast 而非静默返空命中。
- /// </summary>
- private S8EffectiveRule ResolveEffective(AdoS8WatchRule rule)
- {
- var definition = _ruleCatalog.TryGet(rule.RuleCode)
- ?? throw new S8RuleEvaluatorException(
- S8RuleCatalog.ReasonNotFound,
- $"规则 {rule.RuleCode} 在当前版本中没有代码定义,不予执行");
- if (!string.Equals(definition.RuleType, RuleTypeCode, StringComparison.Ordinal))
- throw new S8RuleEvaluatorException(
- "rule_type_mismatch",
- $"规则 {rule.RuleCode} 的代码定义类型为 {definition.RuleType},不能由 TIMEOUT evaluator 执行");
- if (definition.Timeout == null)
- throw new S8RuleEvaluatorException(
- "rule_semantics_missing",
- $"规则 {rule.RuleCode} 的代码定义缺少 TIMEOUT 判定语义");
- return S8EffectiveRule.Resolve(rule, definition);
- }
- /// <summary>
- /// TIMEOUT 判定核心:只消费 canonical 行,不接触 SQL / DataTable / 数据源。
- /// 判定算法与迁移前逐行一致;internal 暴露供契约测试直接驱动。
- /// </summary>
- internal static List<S8RuleHit> EvaluateRows(
- S8MonitoringRowSet rowSet,
- S8EffectiveRule effective,
- long tenantId,
- DateTime detectedAt)
- {
- var definition = effective.Definition;
- var semantics = definition.Timeout!;
- var parameters = effective.Parameters;
- var rule = effective.Row;
- var hits = new List<S8RuleHit>();
- var threshold = detectedAt.AddMinutes(-parameters.GraceMinutes);
- // 源对象类型来自定义。此前是 rule.SourceObjectType ?? rule.WatchObjectType 两列回落 ——
- // 两列都是用户可写的,而该值直接进 dedup_key,改动即造成历史异常断代。
- var sourceObjectType = definition.SourceObjectType;
- // 判定列来自定义。Provider 按 canonical 名产出行,因此这些值就是 canonical 名;
- // 若结果集确实没有该列,读取自然得到 null → 该行不命中,不再退回"某个 JSON 里写的列名"。
- var statusColumn = semantics.StatusColumn;
- var dueAtColumn = semantics.DueAtColumn;
- var objectCodeColumn = semantics.RelatedObjectCodeColumn;
- var objectIdColumn = semantics.SourceObjectIdColumn;
- foreach (var row in rowSet.Rows)
- {
- var status = row.GetString(statusColumn) ?? string.Empty;
- if (semantics.CompletedStates.Contains(status, StringComparer.OrdinalIgnoreCase))
- continue;
- var due = row.GetDateTime(dueAtColumn);
- if (due == null || due > threshold) continue;
- var relatedObjectCode = row.GetString(objectCodeColumn) ?? string.Empty;
- if (string.IsNullOrWhiteSpace(relatedObjectCode)) continue;
- var sourceObjectId = row.GetString(objectIdColumn) ?? relatedObjectCode;
- var dedupKey = BuildDedupKey(tenantId, definition.RuleCode, sourceObjectType, sourceObjectId);
- hits.Add(new S8RuleHit
- {
- SourceRuleId = rule.Id,
- SourceRuleCode = definition.RuleCode,
- SourceObjectType = sourceObjectType,
- SourceObjectId = sourceObjectId,
- RelatedObjectCode = relatedObjectCode,
- ExceptionTypeCode = definition.ExceptionTypeCode,
- SceneCode = definition.SceneCode,
- Severity = parameters.Severity,
- DedupKey = dedupKey,
- SourcePayload = BuildPayload(row, sourceObjectType, sourceObjectId, due.Value, status, definition, parameters),
- DetectedAt = detectedAt,
- Title = $"[超时] {sourceObjectType} {sourceObjectId} 已超期至 {due.Value:yyyy-MM-dd HH:mm:ss}(状态 {status})",
- OccurrenceDeptId = row.GetLong(S8CanonicalColumns.OccurrenceDeptId) ?? parameters.DefaultOccurrenceDeptId,
- ResponsibleDeptId = row.GetLong(S8CanonicalColumns.ResponsibleDeptId) ?? parameters.DefaultResponsibleDeptId
- });
- }
- return hits;
- }
- /// <summary>
- /// 构造 dedup_key:<c>T{tenant}:R{ruleCode}:{sourceObjectType}:{sourceObjectId}</c>。
- ///
- /// <para><b>S8-TENANT-ONLY-BATCH5:去掉了 <c>:F{factory}</c> 段。</b>
- /// 租户隔离由首段 <c>T{tenant}</c> 与各查询的 <c>tenant_id</c> 谓词双重保证;
- /// 工厂段既不提供隔离,又让同一个业务对象在不同工厂元数据下被当成两个异常。</para>
- ///
- /// <para>真库证据(迁移前取证):<c>dwd_supplier_delivery</c> 全表 39854 行 / 4 租户 / 97 个快照,
- /// <c>(tenant_id, stat_date, po_no, po_line)</c> 重复组为 0,跨工厂碰撞组为 0。
- /// 若将来某个规则的 SourceObjectId 在租户内不唯一,唯一性应由该规则的
- /// <b>SourceObjectId contract</b> 负责(例如把工厂编号并进对象标识),
- /// 而不是把工厂重新变回隔离维度。</para>
- ///
- /// internal 暴露供测试。
- /// </summary>
- internal static string BuildDedupKey(long tenantId, string ruleCode, string sourceObjectType, string sourceObjectId) =>
- $"T{tenantId}:R{ruleCode}:{sourceObjectType}:{sourceObjectId}";
- private static string BuildPayload(
- S8MonitoringRow row, string sourceObjectType, string sourceObjectId, DateTime dueAt, string status,
- Definitions.S8RuleDefinition definition, S8RuleRuntimeParameters parameters)
- {
- var payload = new Dictionary<string, object?>(StringComparer.OrdinalIgnoreCase);
- foreach (var kv in row.Values)
- payload[kv.Key] = kv.Value;
- payload["__ruleType"] = RuleTypeCode;
- payload["__sourceObjectType"] = sourceObjectType;
- payload["__sourceObjectId"] = sourceObjectId;
- payload["__dueAt"] = dueAt;
- payload["__status"] = status;
- payload["__graceMinutes"] = parameters.GraceMinutes;
- payload["__exceptionTypeCode"] = definition.ExceptionTypeCode;
- return JsonSerializer.Serialize(payload);
- }
- }
|