tyto/backend/internal/alerts/manager.go

package alerts

import (
	"fmt"
	"sync"
	"time"

	"tyto/internal/models"
)

// Manager handles alert threshold checking and tracking
type Manager struct {
	mu            sync.RWMutex
	config        models.AlertConfig
	activeAlerts  map[string]*models.Alert
	alertHistory  []models.Alert
	thresholdHits map[string]time.Time // Track when threshold was first exceeded
	maxHistory    int
}

// NewManager creates a new alert manager with default config
func NewManager() *Manager {
	return &Manager{
		config:        models.DefaultAlertConfig(),
		activeAlerts:  make(map[string]*models.Alert),
		alertHistory:  make([]models.Alert, 0),
		thresholdHits: make(map[string]time.Time),
		maxHistory:    100,
	}
}

// GetConfig returns current alert configuration
func (m *Manager) GetConfig() models.AlertConfig {
	m.mu.RLock()
	defer m.mu.RUnlock()
	return m.config
}

// SetConfig updates alert configuration
func (m *Manager) SetConfig(config models.AlertConfig) {
	m.mu.Lock()
	defer m.mu.Unlock()
	m.config = config
}

// GetActiveAlerts returns all currently active alerts
func (m *Manager) GetActiveAlerts() []models.Alert {
	m.mu.RLock()
	defer m.mu.RUnlock()

	alerts := make([]models.Alert, 0, len(m.activeAlerts))
	for _, alert := range m.activeAlerts {
		alerts = append(alerts, *alert)
	}
	return alerts
}

// GetAlertHistory returns recent alert history
func (m *Manager) GetAlertHistory() []models.Alert {
	m.mu.RLock()
	defer m.mu.RUnlock()

	history := make([]models.Alert, len(m.alertHistory))
	copy(history, m.alertHistory)
	return history
}

// AcknowledgeAlert marks an alert as acknowledged
func (m *Manager) AcknowledgeAlert(alertID string) bool {
	m.mu.Lock()
	defer m.mu.Unlock()

	// Search by alert ID field, not map key
	for _, alert := range m.activeAlerts {
		if alert.ID == alertID {
			alert.Acknowledged = true
			return true
		}
	}
	return false
}

// CheckMetrics evaluates metrics against thresholds and triggers/resolves alerts
func (m *Manager) CheckMetrics(metrics models.AllMetrics) []models.Alert {
	m.mu.Lock()
	defer m.mu.Unlock()

	var newAlerts []models.Alert
	now := time.Now()

	for _, threshold := range m.config.Thresholds {
		if !threshold.Enabled {
			continue
		}

		var value float64
		var label string

		switch threshold.Type {
		case models.AlertTypeCPU:
			value = metrics.CPU.TotalUsage
			label = "CPU Usage"
		case models.AlertTypeMemory:
			if metrics.Memory.Total > 0 {
				value = float64(metrics.Memory.Used) / float64(metrics.Memory.Total) * 100
			}
			label = "Memory Usage"
		case models.AlertTypeTemperature:
			// Find max temperature
			for _, sensor := range metrics.Temperature.Sensors {
				if sensor.Temperature > value {
					value = sensor.Temperature
					label = fmt.Sprintf("Temperature (%s)", sensor.Name)
				}
			}
			if label == "" {
				label = "Temperature"
			}
		case models.AlertTypeDisk:
			// Find max disk usage
			for _, mount := range metrics.Disk.Mounts {
				if mount.UsedPercent > value {
					value = mount.UsedPercent
					label = fmt.Sprintf("Disk (%s)", mount.MountPoint)
				}
			}
			if label == "" {
				label = "Disk Usage"
			}
		case models.AlertTypeGPU:
			if metrics.GPU.Available {
				value = float64(metrics.GPU.Utilization)
				label = "GPU Usage"
			}
		}

		// Check thresholds
		alertKey := string(threshold.Type)
		var severity models.AlertSeverity
		var thresholdValue float64

		if value >= threshold.CriticalValue {
			severity = models.AlertSeverityCritical
			thresholdValue = threshold.CriticalValue
		} else if value >= threshold.WarningValue {
			severity = models.AlertSeverityWarning
			thresholdValue = threshold.WarningValue
		} else {
			// Value is below thresholds - resolve any active alert
			delete(m.thresholdHits, alertKey)
			if alert, exists := m.activeAlerts[alertKey]; exists {
				resolvedAt := now
				alert.ResolvedAt = &resolvedAt
				m.addToHistory(*alert)
				delete(m.activeAlerts, alertKey)
			}
			continue
		}

		// Check duration requirement
		firstHit, hasHit := m.thresholdHits[alertKey]
		if !hasHit {
			m.thresholdHits[alertKey] = now
			if threshold.DurationSeconds > 0 {
				continue // Wait for duration
			}
			firstHit = now
		}

		elapsed := now.Sub(firstHit)
		if elapsed.Seconds() < float64(threshold.DurationSeconds) {
			continue // Duration not met yet
		}

		// Check if we need to create/update alert
		existingAlert, alertExists := m.activeAlerts[alertKey]

		if !alertExists {
			// Create new alert
			alert := &models.Alert{
				ID:          fmt.Sprintf("%s-%d", threshold.Type, now.Unix()),
				Type:        threshold.Type,
				Severity:    severity,
				Message:     fmt.Sprintf("%s is at %.1f%% (threshold: %.1f%%)", label, value, thresholdValue),
				Value:       value,
				Threshold:   thresholdValue,
				TriggeredAt: now,
			}
			m.activeAlerts[alertKey] = alert
			newAlerts = append(newAlerts, *alert)
		} else if existingAlert.Severity != severity {
			// Severity changed - update alert
			existingAlert.Severity = severity
			existingAlert.Value = value
			existingAlert.Threshold = thresholdValue
			existingAlert.Message = fmt.Sprintf("%s is at %.1f%% (threshold: %.1f%%)", label, value, thresholdValue)
		} else {
			// Just update value
			existingAlert.Value = value
		}
	}

	return newAlerts
}

func (m *Manager) addToHistory(alert models.Alert) {
	m.alertHistory = append([]models.Alert{alert}, m.alertHistory...)
	if len(m.alertHistory) > m.maxHistory {
		m.alertHistory = m.alertHistory[:m.maxHistory]
	}
}