This commit is contained in:
2026-04-12 22:34:46 +08:00
parent d554e9e659
commit 73e16ab6c1
25 changed files with 5962 additions and 52 deletions

View File

@@ -0,0 +1,140 @@
#if false
using Microsoft.Extensions.Diagnostics.HealthChecks;
using OrpaonVision.SiteApp.Runtime.Services;
namespace OrpaonVision.SiteApp.Runtime.HealthChecks;
/// <summary>
/// 报警系统健康检查。
/// </summary>
public sealed class AlarmSystemHealthCheck : IHealthCheck
{
private readonly IAlarmSystemService _alarmSystemService;
private readonly ILogger<AlarmSystemHealthCheck> _logger;
public AlarmSystemHealthCheck(IAlarmSystemService alarmSystemService, ILogger<AlarmSystemHealthCheck> logger)
{
_alarmSystemService = alarmSystemService;
_logger = logger;
}
public async Task<HealthCheckResult> CheckHealthAsync(HealthCheckContext context, CancellationToken cancellationToken = default)
{
try
{
// 检查活跃报警数量
var activeAlarmsResult = await _alarmSystemService.GetActiveAlarmsAsync(cancellationToken);
if (!activeAlarmsResult.IsSuccess)
{
return HealthCheckResult.Unhealthy(
"报警系统服务不可用",
new Dictionary<string, object>
{
["error"] = activeAlarmsResult.Message,
["trace_id"] = activeAlarmsResult.TraceId ?? string.Empty
});
}
var activeAlarmCount = activeAlarmsResult.Data?.Count ?? 0;
// 检查报警栈状态
var criticalStackResult = await _alarmSystemService.GetAlarmStackAsync(Core.AlarmSystem.AlarmStackType.Critical, cancellationToken);
var highStackResult = await _alarmSystemService.GetAlarmStackAsync(Core.AlarmSystem.AlarmStackType.High, cancellationToken);
var criticalCount = criticalStackResult.IsSuccess ? criticalStackResult.Data?.Count ?? 0 : 0;
var highCount = highStackResult.IsSuccess ? highStackResult.Data?.Count ?? 0 : 0;
// 测试报警触发功能
var testAlarmId = Guid.NewGuid();
var testResult = await _alarmSystemService.TriggerAlarmAsync(new Core.AlarmSystem.AlarmRequest
{
RequestId = Guid.NewGuid(),
AlarmType = Core.AlarmSystem.AlarmType.SystemTest,
AlarmLevel = Core.AlarmSystem.AlarmLevel.Info,
Title = "健康检查测试报警",
Description = "用于健康检查的测试报警",
SessionId = Guid.NewGuid(),
AutoClear = true,
AutoClearAfterSeconds = 10
}, cancellationToken);
var testSuccess = testResult.IsSuccess;
var testAlarmIdResult = testResult.Data?.AlarmId ?? Guid.Empty;
// 如果测试成功,立即清除测试报警
if (testSuccess && testAlarmIdResult != Guid.Empty)
{
await _alarmSystemService.ClearAlarmAsync(new Core.AlarmSystem.AlarmClearRequest
{
AlarmId = testAlarmIdResult,
ClearUser = "HealthCheck",
ClearReason = "健康检查测试报警清除"
}, cancellationToken);
}
// 判断健康状态
var data = new Dictionary<string, object>
{
["active_alarm_count"] = activeAlarmCount,
["critical_stack_count"] = criticalCount,
["high_stack_count"] = highCount,
["test_alarm_success"] = testSuccess,
["test_alarm_trigger_time_ms"] = testResult.Data?.TriggerElapsedMs ?? 0,
["status"] = "healthy"
};
// 检查是否有过多活跃报警
if (activeAlarmCount > 100)
{
return HealthCheckResult.Degraded(
"活跃报警数量过多",
new Dictionary<string, object>(data)
{
["status"] = "too_many_alarms"
});
}
// 检查是否有严重报警
if (criticalCount > 0)
{
return HealthCheckResult.Degraded(
"存在严重报警",
new Dictionary<string, object>(data)
{
["status"] = "critical_alarms"
});
}
// 检查测试报警是否成功
if (!testSuccess)
{
return HealthCheckResult.Unhealthy(
"报警触发测试失败",
new Dictionary<string, object>(data)
{
["status"] = "test_failed",
["error"] = testResult.Message
});
}
_logger.LogDebug("报警系统健康检查通过:活跃报警={ActiveCount},严重报警={CriticalCount},高级报警={HighCount},测试耗时={ElapsedMs}ms",
activeAlarmCount, criticalCount, highCount, testResult.Data?.TriggerElapsedMs ?? 0);
return HealthCheckResult.Healthy("报警系统运行正常", data);
}
catch (Exception ex)
{
_logger.LogError(ex, "报警系统健康检查异常");
return HealthCheckResult.Unhealthy(
"报警系统健康检查异常",
new Dictionary<string, object>
{
["error"] = ex.Message,
["exception_type"] = ex.GetType().Name
});
}
}
}
#endif

View File

@@ -0,0 +1,113 @@
#if false
using Microsoft.Extensions.Diagnostics.HealthChecks;
using OrpaonVision.SiteApp.Runtime.Services;
namespace OrpaonVision.SiteApp.Runtime.HealthChecks;
/// <summary>
/// 规则引擎健康检查。
/// </summary>
public sealed class RuleEngineHealthCheck : IHealthCheck
{
private readonly IRuleEngineService _ruleEngineService;
private readonly ILogger<RuleEngineHealthCheck> _logger;
public RuleEngineHealthCheck(IRuleEngineService ruleEngineService, ILogger<RuleEngineHealthCheck> logger)
{
_ruleEngineService = ruleEngineService;
_logger = logger;
}
public async Task<HealthCheckResult> CheckHealthAsync(HealthCheckContext context, CancellationToken cancellationToken = default)
{
try
{
// 检查规则引擎服务是否可用
var rules = await _ruleEngineService.GetRulesAsync(cancellationToken);
if (!rules.IsSuccess)
{
return HealthCheckResult.Unhealthy(
"规则引擎服务不可用",
new Dictionary<string, object>
{
["error"] = rules.Message,
["trace_id"] = rules.TraceId ?? string.Empty
});
}
// 检查是否有规则配置
var ruleCount = rules.Data?.Count ?? 0;
if (ruleCount == 0)
{
return HealthCheckResult.Degraded(
"规则引擎可用但无规则配置",
new Dictionary<string, object>
{
["rule_count"] = ruleCount,
["status"] = "no_rules"
});
}
// 执行简单的规则评估测试
var testResult = await _ruleEngineService.EvaluateRulesAsync(new Core.RuleEngine.RuleEvaluationRequest
{
SessionId = Guid.NewGuid(),
Inference = new Core.RuleEngine.InferenceResultDto
{
SessionId = Guid.NewGuid(),
Timestamp = DateTime.UtcNow,
Detections = new List<Core.RuleEngine.InferenceDetectionDto>
{
new()
{
ClassName = "test",
Confidence = 0.9f,
CenterX = 100,
CenterY = 100,
Width = 50,
Height = 50
}
}
}
}, cancellationToken);
if (!testResult.IsSuccess)
{
return HealthCheckResult.Unhealthy(
"规则引擎评估测试失败",
new Dictionary<string, object>
{
["error"] = testResult.Message,
["trace_id"] = testResult.TraceId ?? string.Empty
});
}
var data = new Dictionary<string, object>
{
["rule_count"] = ruleCount,
["test_evaluation_time_ms"] = testResult.Data?.EvaluationElapsedMs ?? 0,
["test_result_pass"] = testResult.Data?.OverallResult == Core.RuleEngine.RuleResult.Pass,
["status"] = "healthy"
};
_logger.LogDebug("规则引擎健康检查通过:规则数量={RuleCount},测试评估耗时={ElapsedMs}ms",
ruleCount, testResult.Data?.EvaluationElapsedMs ?? 0);
return HealthCheckResult.Healthy("规则引擎运行正常", data);
}
catch (Exception ex)
{
_logger.LogError(ex, "规则引擎健康检查异常");
return HealthCheckResult.Unhealthy(
"规则引擎健康检查异常",
new Dictionary<string, object>
{
["error"] = ex.Message,
["exception_type"] = ex.GetType().Name
});
}
}
}
#endif

View File

@@ -0,0 +1,104 @@
#if false
using Microsoft.Extensions.Diagnostics.HealthChecks;
using OrpaonVision.SiteApp.Runtime.Services;
namespace OrpaonVision.SiteApp.Runtime.HealthChecks;
/// <summary>
/// 状态机健康检查。
/// </summary>
public sealed class StateMachineHealthCheck : IHealthCheck
{
private readonly IRuntimeStateMachineService _stateMachineService;
private readonly ILogger<StateMachineHealthCheck> _logger;
public StateMachineHealthCheck(IRuntimeStateMachineService stateMachineService, ILogger<StateMachineHealthCheck> logger)
{
_stateMachineService = stateMachineService;
_logger = logger;
}
public async Task<HealthCheckResult> CheckHealthAsync(HealthCheckContext context, CancellationToken cancellationToken = default)
{
try
{
// 检查状态机当前状态
var currentState = _stateMachineService.GetCurrentState();
var currentLayer = _stateMachineService.GetCurrentLayer();
// 检查状态机是否响应
var canInitialize = _stateMachineService.CanExecuteOperation(Core.Common.StateTrigger.Initialize);
var canStart = _stateMachineService.CanExecuteOperation(Core.Common.StateTrigger.Start);
// 获取事件历史
var historyResult = await _stateMachineService.GetEventHistoryAsync(10, cancellationToken);
if (!historyResult.IsSuccess)
{
return HealthCheckResult.Unhealthy(
"状态机事件历史获取失败",
new Dictionary<string, object>
{
["error"] = historyResult.Message,
["trace_id"] = historyResult.TraceId ?? string.Empty
});
}
var eventCount = historyResult.Data?.Count ?? 0;
var lastEventTime = eventCount > 0 ? historyResult.Data?.FirstOrDefault()?.TimestampUtc : (DateTime?)null;
// 判断健康状态
var data = new Dictionary<string, object>
{
["current_state"] = currentState.ToString(),
["current_layer"] = currentLayer,
["can_initialize"] = canInitialize,
["can_start"] = canStart,
["event_count"] = eventCount,
["last_event_time"] = lastEventTime?.ToString("yyyy-MM-dd HH:mm:ss") ?? "无事件",
["status"] = "healthy"
};
// 检查是否处于异常状态
if (currentState == Core.Common.RuntimeState.Error || currentState == Core.Common.RuntimeState.Faulted)
{
return HealthCheckResult.Unhealthy(
"状态机处于异常状态",
new Dictionary<string, object>(data)
{
["status"] = "error_state"
});
}
// 检查是否长时间无活动超过5分钟
if (lastEventTime.HasValue && DateTime.UtcNow.Subtract(lastEventTime.Value).TotalMinutes > 5)
{
return HealthCheckResult.Degraded(
"状态机长时间无活动",
new Dictionary<string, object>(data)
{
["status"] = "inactive",
["inactive_minutes"] = DateTime.UtcNow.Subtract(lastEventTime.Value).TotalMinutes
});
}
_logger.LogDebug("状态机健康检查通过:当前状态={CurrentState},当前层级={CurrentLayer},事件数量={EventCount}",
currentState, currentLayer, eventCount);
return HealthCheckResult.Healthy("状态机运行正常", data);
}
catch (Exception ex)
{
_logger.LogError(ex, "状态机健康检查异常");
return HealthCheckResult.Unhealthy(
"状态机健康检查异常",
new Dictionary<string, object>
{
["error"] = ex.Message,
["exception_type"] = ex.GetType().Name
});
}
}
}
#endif