feat(ai): Add failure pattern detection for predictive intelligence (Phase 5)

Create internal/ai/patterns package: 1. Pattern Detector (detector.go): - Records historical events (high memory, OOM, restarts, etc.) - Detects recurring failure patterns - Calculates average interval between occurrences - Computes confidence based on pattern consistency - Predicts when failures will occur again - Persists to ai_patterns.json 2. Event types tracked: - high_memory, high_cpu, disk_full - oom, restart, unresponsive - backup_failed 3. Integration: - Wire PatternDetector into router startup - Add to AI context in buildEnrichedContext - FormatForContext generates failure predictions Example AI context now includes: 'OOM events typically occurs every ~10 days (next expected in ~3 days)' This enables proactive alerts before problems recur. All tests passing.
2025-12-12 14:11:28 +00:00 · 2025-12-12 14:11:28 +00:00 · d9d798084e
commit d9d798084e
parent dacdd48e28
7 changed files with 837 additions and 0 deletions
--- a/internal/ai/patrol.go
+++ b/internal/ai/patrol.go
@ -214,6 +214,7 @@ type PatrolService struct {
 	baselineStore     *baseline.Store  // For anomaly detection via learned baselines
 	changeDetector    *ChangeDetector  // For tracking infrastructure changes
 	remediationLog    *RemediationLog  // For tracking remediation actions
+	patternDetector   *PatternDetector // For failure prediction from historical patterns

 	// Cached thresholds (recalculated when thresholdProvider changes)
 	thresholds PatrolThresholds
@ -383,6 +384,21 @@ func (p *PatrolService) GetRemediationLog() *RemediationLog {
 	return p.remediationLog
 }

+// SetPatternDetector sets the pattern detector for failure prediction
+func (p *PatrolService) SetPatternDetector(detector *PatternDetector) {
+	p.mu.Lock()
+	defer p.mu.Unlock()
+	p.patternDetector = detector
+	log.Info().Msg("AI Patrol: Pattern detector set for failure prediction")
+}
+
+// GetPatternDetector returns the pattern detector
+func (p *PatrolService) GetPatternDetector() *PatternDetector {
+	p.mu.RLock()
+	defer p.mu.RUnlock()
+	return p.patternDetector
+}
+
 // GetConfig returns the current patrol configuration
 func (p *PatrolService) GetConfig() PatrolConfig {
 	p.mu.RLock()
@ -1782,6 +1798,18 @@ func (p *PatrolService) buildEnrichedContext(state models.StateSnapshot) string
 			log.Debug().Int("new_changes", len(newChanges)).Msg("AI Patrol: Detected infrastructure changes")
 		}
 	}
+	
+	// Append failure predictions if pattern detector is available
+	p.mu.RLock()
+	patternDetector := p.patternDetector
+	p.mu.RUnlock()
+	
+	if patternDetector != nil {
+		predictionsContext := patternDetector.FormatForContext("")
+		if predictionsContext != "" {
+			formatted += predictionsContext
+		}
+	}

 	log.Debug().
 		Int("resources", infraCtx.TotalResources).
--- a/internal/ai/pattern_exports.go
+++ b/internal/ai/pattern_exports.go
@ -0,0 +1,44 @@
+package ai
+
+import (
+	"github.com/rcourtman/pulse-go-rewrite/internal/ai/patterns"
+)
+
+// PatternDetector is an alias for patterns.Detector
+type PatternDetector = patterns.Detector
+
+// PatternDetectorConfig is an alias for patterns.DetectorConfig
+type PatternDetectorConfig = patterns.DetectorConfig
+
+// HistoricalEvent is an alias for patterns.HistoricalEvent
+type HistoricalEvent = patterns.HistoricalEvent
+
+// FailurePrediction is an alias for patterns.FailurePrediction
+type FailurePrediction = patterns.FailurePrediction
+
+// EventType is an alias for patterns.EventType
+type EventType = patterns.EventType
+
+// Pattern is an alias for patterns.Pattern
+type Pattern = patterns.Pattern
+
+// Event type constants
+const (
+	EventHighMemory   = patterns.EventHighMemory
+	EventHighCPU      = patterns.EventHighCPU
+	EventDiskFull     = patterns.EventDiskFull
+	EventOOM          = patterns.EventOOM
+	EventRestart      = patterns.EventRestart
+	EventUnresponsive = patterns.EventUnresponsive
+	EventBackupFailed = patterns.EventBackupFailed
+)
+
+// NewPatternDetector creates a new pattern detector
+func NewPatternDetector(cfg PatternDetectorConfig) *PatternDetector {
+	return patterns.NewDetector(cfg)
+}
+
+// DefaultPatternConfig returns default pattern detector configuration
+func DefaultPatternConfig() PatternDetectorConfig {
+	return patterns.DefaultConfig()
+}
--- a/internal/ai/patterns/detector.go
+++ b/internal/ai/patterns/detector.go
@ -0,0 +1,531 @@
+// Package patterns provides failure pattern detection for predictive intelligence.
+// It analyzes historical data to identify recurring issues and predict future failures.
+package patterns
+
+import (
+	"encoding/json"
+	"math"
+	"os"
+	"path/filepath"
+	"sort"
+	"sync"
+	"time"
+
+	"github.com/rs/zerolog/log"
+)
+
+// EventType represents the type of event being tracked
+type EventType string
+
+const (
+	EventHighMemory    EventType = "high_memory"    // Memory exceeded threshold
+	EventHighCPU       EventType = "high_cpu"       // CPU exceeded threshold
+	EventDiskFull      EventType = "disk_full"      // Disk space critical
+	EventOOM           EventType = "oom"            // Out of memory kill
+	EventRestart       EventType = "restart"        // Resource restarted
+	EventUnresponsive  EventType = "unresponsive"   // Resource became unresponsive
+	EventBackupFailed  EventType = "backup_failed"  // Backup job failed
+)
+
+// HistoricalEvent represents a recorded event
+type HistoricalEvent struct {
+	ID          string    `json:"id"`
+	ResourceID  string    `json:"resource_id"`
+	EventType   EventType `json:"event_type"`
+	Timestamp   time.Time `json:"timestamp"`
+	Description string    `json:"description,omitempty"`
+	Resolved    bool      `json:"resolved"`
+	ResolvedAt  time.Time `json:"resolved_at,omitempty"`
+	Duration    time.Duration `json:"duration,omitempty"` // How long it lasted
+}
+
+// Pattern represents a detected recurring pattern
+type Pattern struct {
+	ResourceID     string        `json:"resource_id"`
+	EventType      EventType     `json:"event_type"`
+	Occurrences    int           `json:"occurrences"`      // Number of times event occurred
+	AverageInterval time.Duration `json:"average_interval"` // Average time between occurrences
+	StdDevInterval time.Duration `json:"stddev_interval"`  // Standard deviation
+	LastOccurrence time.Time     `json:"last_occurrence"`
+	NextPredicted  time.Time     `json:"next_predicted"`   // When we expect it to happen again
+	Confidence     float64       `json:"confidence"`       // 0-1, based on consistency
+	AverageDuration time.Duration `json:"average_duration,omitempty"` // How long events typically last
+}
+
+// FailurePrediction represents a predicted future failure
+type FailurePrediction struct {
+	ResourceID  string    `json:"resource_id"`
+	EventType   EventType `json:"event_type"`
+	PredictedAt time.Time `json:"predicted_at"`
+	DaysUntil   float64   `json:"days_until"`
+	Confidence  float64   `json:"confidence"`
+	Basis       string    `json:"basis"` // Human-readable explanation
+	Pattern     *Pattern  `json:"pattern,omitempty"`
+}
+
+// Detector tracks historical events and detects patterns
+type Detector struct {
+	mu       sync.RWMutex
+	events   []HistoricalEvent
+	patterns map[string]*Pattern // resourceID:eventType -> pattern
+	
+	// Configuration
+	maxEvents       int
+	minOccurrences  int           // Minimum occurrences to form a pattern
+	patternWindow   time.Duration // How far back to look for patterns
+	predictionLimit time.Duration // How far ahead to predict
+	
+	// Persistence
+	dataDir string
+}
+
+// DetectorConfig configures the pattern detector
+type DetectorConfig struct {
+	MaxEvents       int
+	MinOccurrences  int           // Default: 3
+	PatternWindow   time.Duration // Default: 90 days
+	PredictionLimit time.Duration // Default: 30 days
+	DataDir         string
+}
+
+// DefaultConfig returns default detector configuration
+func DefaultConfig() DetectorConfig {
+	return DetectorConfig{
+		MaxEvents:       5000,
+		MinOccurrences:  3,
+		PatternWindow:   90 * 24 * time.Hour,
+		PredictionLimit: 30 * 24 * time.Hour,
+	}
+}
+
+// NewDetector creates a new pattern detector
+func NewDetector(cfg DetectorConfig) *Detector {
+	if cfg.MaxEvents <= 0 {
+		cfg.MaxEvents = 5000
+	}
+	if cfg.MinOccurrences <= 0 {
+		cfg.MinOccurrences = 3
+	}
+	if cfg.PatternWindow <= 0 {
+		cfg.PatternWindow = 90 * 24 * time.Hour
+	}
+	if cfg.PredictionLimit <= 0 {
+		cfg.PredictionLimit = 30 * 24 * time.Hour
+	}
+	
+	d := &Detector{
+		events:          make([]HistoricalEvent, 0),
+		patterns:        make(map[string]*Pattern),
+		maxEvents:       cfg.MaxEvents,
+		minOccurrences:  cfg.MinOccurrences,
+		patternWindow:   cfg.PatternWindow,
+		predictionLimit: cfg.PredictionLimit,
+		dataDir:         cfg.DataDir,
+	}
+	
+	// Load existing data
+	if cfg.DataDir != "" {
+		if err := d.loadFromDisk(); err != nil {
+			log.Warn().Err(err).Msg("Failed to load pattern history from disk")
+		} else if len(d.events) > 0 {
+			log.Info().Int("events", len(d.events)).Int("patterns", len(d.patterns)).
+				Msg("Loaded pattern history from disk")
+		}
+	}
+	
+	return d
+}
+
+// RecordEvent records a new event for pattern analysis
+func (d *Detector) RecordEvent(event HistoricalEvent) {
+	d.mu.Lock()
+	defer d.mu.Unlock()
+	
+	if event.ID == "" {
+		event.ID = generateEventID()
+	}
+	if event.Timestamp.IsZero() {
+		event.Timestamp = time.Now()
+	}
+	
+	d.events = append(d.events, event)
+	d.trimEvents()
+	
+	// Recompute pattern for this resource/event type
+	key := patternKey(event.ResourceID, event.EventType)
+	d.patterns[key] = d.computePattern(event.ResourceID, event.EventType)
+	
+	// Persist asynchronously
+	go func() {
+		if err := d.saveToDisk(); err != nil {
+			log.Warn().Err(err).Msg("Failed to save pattern history")
+		}
+	}()
+}
+
+// RecordFromAlert records an event from an alert
+func (d *Detector) RecordFromAlert(resourceID string, alertType string, timestamp time.Time) {
+	eventType := mapAlertToEventType(alertType)
+	if eventType == "" {
+		return // Not a trackable event type
+	}
+	
+	d.RecordEvent(HistoricalEvent{
+		ResourceID:  resourceID,
+		EventType:   eventType,
+		Timestamp:   timestamp,
+		Description: alertType,
+	})
+}
+
+// GetPredictions returns failure predictions for all tracked resources
+func (d *Detector) GetPredictions() []FailurePrediction {
+	d.mu.RLock()
+	defer d.mu.RUnlock()
+	
+	var predictions []FailurePrediction
+	now := time.Now()
+	
+	for _, pattern := range d.patterns {
+		// Only predict if pattern has sufficient confidence
+		if pattern.Confidence < 0.3 || pattern.Occurrences < d.minOccurrences {
+			continue
+		}
+		
+		// Check if prediction is within our limit
+		if pattern.NextPredicted.Before(now) || pattern.NextPredicted.After(now.Add(d.predictionLimit)) {
+			continue
+		}
+		
+		daysUntil := pattern.NextPredicted.Sub(now).Hours() / 24
+		
+		predictions = append(predictions, FailurePrediction{
+			ResourceID:  pattern.ResourceID,
+			EventType:   pattern.EventType,
+			PredictedAt: pattern.NextPredicted,
+			DaysUntil:   daysUntil,
+			Confidence:  pattern.Confidence,
+			Basis:       formatPatternBasis(pattern),
+			Pattern:     pattern,
+		})
+	}
+	
+	// Sort by days until (soonest first)
+	sort.Slice(predictions, func(i, j int) bool {
+		return predictions[i].DaysUntil < predictions[j].DaysUntil
+	})
+	
+	return predictions
+}
+
+// GetPredictionsForResource returns failure predictions for a specific resource
+func (d *Detector) GetPredictionsForResource(resourceID string) []FailurePrediction {
+	all := d.GetPredictions()
+	var result []FailurePrediction
+	for _, p := range all {
+		if p.ResourceID == resourceID {
+			result = append(result, p)
+		}
+	}
+	return result
+}
+
+// GetPatterns returns all detected patterns
+func (d *Detector) GetPatterns() map[string]*Pattern {
+	d.mu.RLock()
+	defer d.mu.RUnlock()
+	
+	result := make(map[string]*Pattern)
+	for k, v := range d.patterns {
+		result[k] = v
+	}
+	return result
+}
+
+// computePattern analyzes events to find patterns for a resource/event type
+func (d *Detector) computePattern(resourceID string, eventType EventType) *Pattern {
+	cutoff := time.Now().Add(-d.patternWindow)
+	
+	// Get all events for this resource/type within the window
+	var events []HistoricalEvent
+	for _, e := range d.events {
+		if e.ResourceID == resourceID && e.EventType == eventType && e.Timestamp.After(cutoff) {
+			events = append(events, e)
+		}
+	}
+	
+	if len(events) < d.minOccurrences {
+		return nil
+	}
+	
+	// Sort by timestamp
+	sort.Slice(events, func(i, j int) bool {
+		return events[i].Timestamp.Before(events[j].Timestamp)
+	})
+	
+	// Calculate intervals between events
+	var intervals []time.Duration
+	var durations []time.Duration
+	
+	for i := 1; i < len(events); i++ {
+		interval := events[i].Timestamp.Sub(events[i-1].Timestamp)
+		intervals = append(intervals, interval)
+		
+		if events[i-1].Duration > 0 {
+			durations = append(durations, events[i-1].Duration)
+		}
+	}
+	
+	if len(intervals) == 0 {
+		return nil
+	}
+	
+	// Calculate average and stddev of intervals
+	avgInterval := averageDuration(intervals)
+	stddevInterval := stddevDuration(intervals, avgInterval)
+	
+	// Calculate confidence based on consistency
+	// If stddev is low relative to mean, pattern is more reliable
+	consistency := 1.0
+	if avgInterval > 0 {
+		cv := float64(stddevInterval) / float64(avgInterval) // Coefficient of variation
+		consistency = 1.0 - math.Min(cv, 1.0) // Higher consistency = lower CV
+	}
+	
+	// Adjust confidence based on number of occurrences
+	occurrenceBonus := math.Min(float64(len(events))/10.0, 0.3)
+	confidence := consistency*0.7 + occurrenceBonus
+	
+	// Predict next occurrence
+	lastEvent := events[len(events)-1]
+	nextPredicted := lastEvent.Timestamp.Add(avgInterval)
+	
+	// Calculate average duration if available
+	var avgDuration time.Duration
+	if len(durations) > 0 {
+		avgDuration = averageDuration(durations)
+	}
+	
+	return &Pattern{
+		ResourceID:       resourceID,
+		EventType:        eventType,
+		Occurrences:      len(events),
+		AverageInterval:  avgInterval,
+		StdDevInterval:   stddevInterval,
+		LastOccurrence:   lastEvent.Timestamp,
+		NextPredicted:    nextPredicted,
+		Confidence:       confidence,
+		AverageDuration:  avgDuration,
+	}
+}
+
+// trimEvents removes old events beyond maxEvents
+func (d *Detector) trimEvents() {
+	if len(d.events) > d.maxEvents {
+		d.events = d.events[len(d.events)-d.maxEvents:]
+	}
+}
+
+// saveToDisk persists events and patterns
+func (d *Detector) saveToDisk() error {
+	if d.dataDir == "" {
+		return nil
+	}
+	
+	d.mu.RLock()
+	data := struct {
+		Events   []HistoricalEvent   `json:"events"`
+		Patterns map[string]*Pattern `json:"patterns"`
+	}{
+		Events:   d.events,
+		Patterns: d.patterns,
+	}
+	d.mu.RUnlock()
+	
+	jsonData, err := json.MarshalIndent(data, "", "  ")
+	if err != nil {
+		return err
+	}
+	
+	path := filepath.Join(d.dataDir, "ai_patterns.json")
+	tmpPath := path + ".tmp"
+	if err := os.WriteFile(tmpPath, jsonData, 0600); err != nil {
+		return err
+	}
+	
+	return os.Rename(tmpPath, path)
+}
+
+// loadFromDisk loads events and patterns
+func (d *Detector) loadFromDisk() error {
+	if d.dataDir == "" {
+		return nil
+	}
+	
+	path := filepath.Join(d.dataDir, "ai_patterns.json")
+	jsonData, err := os.ReadFile(path)
+	if err != nil {
+		if os.IsNotExist(err) {
+			return nil
+		}
+		return err
+	}
+	
+	var data struct {
+		Events   []HistoricalEvent   `json:"events"`
+		Patterns map[string]*Pattern `json:"patterns"`
+	}
+	
+	if err := json.Unmarshal(jsonData, &data); err != nil {
+		return err
+	}
+	
+	d.events = data.Events
+	d.patterns = data.Patterns
+	
+	return nil
+}
+
+// FormatForContext formats predictions for AI consumption
+func (d *Detector) FormatForContext(resourceID string) string {
+	var predictions []FailurePrediction
+	if resourceID != "" {
+		predictions = d.GetPredictionsForResource(resourceID)
+	} else {
+		predictions = d.GetPredictions()
+	}
+	
+	if len(predictions) == 0 {
+		return ""
+	}
+	
+	var result string
+	result = "\n## ⏰ Failure Predictions\n"
+	result += "Based on historical patterns:\n"
+	
+	for _, p := range predictions {
+		if len(result) > 2000 { // Limit context size
+			result += "\n... and more\n"
+			break
+		}
+		result += "- " + p.Basis + "\n"
+	}
+	
+	return result
+}
+
+// Helper functions
+
+var eventCounter int64
+
+func generateEventID() string {
+	eventCounter++
+	return time.Now().Format("20060102150405") + "-" + intToStr(int(eventCounter%1000))
+}
+
+func intToStr(n int) string {
+	if n == 0 {
+		return "0"
+	}
+	var result string
+	for n > 0 {
+		result = string(rune('0'+n%10)) + result
+		n /= 10
+	}
+	return result
+}
+
+func patternKey(resourceID string, eventType EventType) string {
+	return resourceID + ":" + string(eventType)
+}
+
+func mapAlertToEventType(alertType string) EventType {
+	switch alertType {
+	case "memory_warning", "memory_critical":
+		return EventHighMemory
+	case "cpu_warning", "cpu_critical":
+		return EventHighCPU
+	case "disk_warning", "disk_critical":
+		return EventDiskFull
+	case "oom", "out_of_memory":
+		return EventOOM
+	case "restart", "restarted":
+		return EventRestart
+	case "unresponsive", "unreachable":
+		return EventUnresponsive
+	case "backup_failed":
+		return EventBackupFailed
+	default:
+		return ""
+	}
+}
+
+func averageDuration(durations []time.Duration) time.Duration {
+	if len(durations) == 0 {
+		return 0
+	}
+	var sum int64
+	for _, d := range durations {
+		sum += int64(d)
+	}
+	return time.Duration(sum / int64(len(durations)))
+}
+
+func stddevDuration(durations []time.Duration, mean time.Duration) time.Duration {
+	if len(durations) < 2 {
+		return 0
+	}
+	var sumSquares float64
+	for _, d := range durations {
+		diff := float64(d - mean)
+		sumSquares += diff * diff
+	}
+	variance := sumSquares / float64(len(durations)-1)
+	return time.Duration(math.Sqrt(variance))
+}
+
+func formatPatternBasis(p *Pattern) string {
+	daysInterval := p.AverageInterval.Hours() / 24
+	daysSinceLast := time.Since(p.LastOccurrence).Hours() / 24
+	daysUntilNext := p.NextPredicted.Sub(time.Now()).Hours() / 24
+	
+	eventName := string(p.EventType)
+	switch p.EventType {
+	case EventHighMemory:
+		eventName = "high memory usage"
+	case EventHighCPU:
+		eventName = "high CPU usage"
+	case EventDiskFull:
+		eventName = "disk space critical"
+	case EventOOM:
+		eventName = "OOM events"
+	case EventRestart:
+		eventName = "restarts"
+	case EventUnresponsive:
+		eventName = "unresponsive periods"
+	case EventBackupFailed:
+		eventName = "backup failures"
+	}
+	
+	if daysUntilNext < 0 {
+		return eventName + " typically occurs every ~" + formatDays(daysInterval) + 
+			" (last: " + formatDays(daysSinceLast) + " ago, overdue)"
+	}
+	
+	return eventName + " typically occurs every ~" + formatDays(daysInterval) + 
+		" (next expected in ~" + formatDays(daysUntilNext) + ")"
+}
+
+func formatDays(days float64) string {
+	if days < 1 {
+		hours := days * 24
+		if hours < 1 {
+			return "less than an hour"
+		}
+		return intToStr(int(hours)) + " hours"
+	}
+	if days < 2 {
+		return "1 day"
+	}
+	return intToStr(int(days)) + " days"
+}
--- a/internal/ai/patterns/detector_test.go
+++ b/internal/ai/patterns/detector_test.go
@ -0,0 +1,206 @@
+package patterns
+
+import (
+	"testing"
+	"time"
+)
+
+func TestDetector_RecordEvent(t *testing.T) {
+	d := NewDetector(DetectorConfig{MinOccurrences: 2})
+
+	// Record first event
+	d.RecordEvent(HistoricalEvent{
+		ResourceID: "vm-100",
+		EventType:  EventHighMemory,
+		Timestamp:  time.Now().Add(-10 * 24 * time.Hour),
+	})
+
+	if len(d.events) != 1 {
+		t.Errorf("Expected 1 event, got %d", len(d.events))
+	}
+}
+
+func TestDetector_PatternDetection(t *testing.T) {
+	d := NewDetector(DetectorConfig{MinOccurrences: 3, PatternWindow: 365 * 24 * time.Hour})
+
+	// Record events with 10-day interval
+	now := time.Now()
+	for i := 5; i >= 0; i-- {
+		d.RecordEvent(HistoricalEvent{
+			ResourceID: "vm-100",
+			EventType:  EventHighMemory,
+			Timestamp:  now.Add(-time.Duration(i*10) * 24 * time.Hour),
+		})
+	}
+
+	// Check that pattern was detected
+	patterns := d.GetPatterns()
+	key := patternKey("vm-100", EventHighMemory)
+	pattern, ok := patterns[key]
+	if !ok {
+		t.Fatal("Expected pattern to be detected")
+	}
+
+	if pattern.Occurrences != 6 {
+		t.Errorf("Expected 6 occurrences, got %d", pattern.Occurrences)
+	}
+
+	// Average interval should be ~10 days
+	avgDays := pattern.AverageInterval.Hours() / 24
+	if avgDays < 9 || avgDays > 11 {
+		t.Errorf("Expected ~10 day interval, got %.1f days", avgDays)
+	}
+}
+
+func TestDetector_GetPredictions(t *testing.T) {
+	d := NewDetector(DetectorConfig{
+		MinOccurrences:  3,
+		PatternWindow:   365 * 24 * time.Hour,
+		PredictionLimit: 30 * 24 * time.Hour,
+	})
+
+	// Record events with regular interval
+	now := time.Now()
+	for i := 3; i >= 0; i-- {
+		d.RecordEvent(HistoricalEvent{
+			ResourceID: "vm-100",
+			EventType:  EventOOM,
+			Timestamp:  now.Add(-time.Duration(i*7) * 24 * time.Hour), // 7-day interval
+		})
+	}
+
+	predictions := d.GetPredictions()
+	
+	// Should have a prediction for OOM
+	found := false
+	for _, p := range predictions {
+		if p.ResourceID == "vm-100" && p.EventType == EventOOM {
+			found = true
+			// Should predict in ~7 days
+			if p.DaysUntil < 5 || p.DaysUntil > 9 {
+				t.Errorf("Expected prediction in ~7 days, got %.1f days", p.DaysUntil)
+			}
+			break
+		}
+	}
+	
+	if !found {
+		t.Error("Expected OOM prediction for vm-100")
+	}
+}
+
+func TestDetector_GetPredictionsForResource(t *testing.T) {
+	d := NewDetector(DetectorConfig{MinOccurrences: 3, PatternWindow: 365 * 24 * time.Hour})
+
+	now := time.Now()
+	// Add pattern for vm-100
+	for i := 3; i >= 0; i-- {
+		d.RecordEvent(HistoricalEvent{
+			ResourceID: "vm-100",
+			EventType:  EventRestart,
+			Timestamp:  now.Add(-time.Duration(i*14) * 24 * time.Hour),
+		})
+	}
+	// Add pattern for vm-200
+	for i := 3; i >= 0; i-- {
+		d.RecordEvent(HistoricalEvent{
+			ResourceID: "vm-200",
+			EventType:  EventHighCPU,
+			Timestamp:  now.Add(-time.Duration(i*5) * 24 * time.Hour),
+		})
+	}
+
+	// Get predictions for vm-100 only
+	predictions := d.GetPredictionsForResource("vm-100")
+	for _, p := range predictions {
+		if p.ResourceID != "vm-100" {
+			t.Errorf("Got prediction for wrong resource: %s", p.ResourceID)
+		}
+	}
+}
+
+func TestDetector_Confidence(t *testing.T) {
+	d := NewDetector(DetectorConfig{MinOccurrences: 3, PatternWindow: 365 * 24 * time.Hour})
+
+	now := time.Now()
+	// Add very consistent pattern (every 7 days exactly)
+	for i := 5; i >= 0; i-- {
+		d.RecordEvent(HistoricalEvent{
+			ResourceID: "consistent-vm",
+			EventType:  EventHighMemory,
+			Timestamp:  now.Add(-time.Duration(i*7*24) * time.Hour),
+		})
+	}
+
+	patterns := d.GetPatterns()
+	pattern := patterns[patternKey("consistent-vm", EventHighMemory)]
+	if pattern == nil {
+		t.Fatal("Expected pattern")
+	}
+
+	// Consistent pattern should have high confidence
+	if pattern.Confidence < 0.5 {
+		t.Errorf("Expected high confidence for consistent pattern, got %.2f", pattern.Confidence)
+	}
+}
+
+func TestDetector_FormatForContext(t *testing.T) {
+	d := NewDetector(DetectorConfig{MinOccurrences: 3, PatternWindow: 365 * 24 * time.Hour})
+
+	now := time.Now()
+	for i := 3; i >= 0; i-- {
+		d.RecordEvent(HistoricalEvent{
+			ResourceID: "vm-100",
+			EventType:  EventOOM,
+			Timestamp:  now.Add(-time.Duration(i*10) * 24 * time.Hour),
+		})
+	}
+
+	context := d.FormatForContext("vm-100")
+	if context == "" {
+		t.Error("Expected non-empty context")
+	}
+	
+	if !contains(context, "OOM") && !contains(context, "oom") {
+		t.Errorf("Expected context to mention OOM: %s", context)
+	}
+}
+
+func TestMapAlertToEventType(t *testing.T) {
+	tests := []struct {
+		alertType string
+		expected  EventType
+	}{
+		{"memory_warning", EventHighMemory},
+		{"memory_critical", EventHighMemory},
+		{"cpu_warning", EventHighCPU},
+		{"cpu_critical", EventHighCPU},
+		{"disk_warning", EventDiskFull},
+		{"disk_critical", EventDiskFull},
+		{"oom", EventOOM},
+		{"restart", EventRestart},
+		{"unresponsive", EventUnresponsive},
+		{"backup_failed", EventBackupFailed},
+		{"unknown_alert", ""},
+	}
+
+	for _, tc := range tests {
+		result := mapAlertToEventType(tc.alertType)
+		if result != tc.expected {
+			t.Errorf("mapAlertToEventType(%q) = %q, want %q", tc.alertType, result, tc.expected)
+		}
+	}
+}
+
+func contains(s, substr string) bool {
+	return len(s) >= len(substr) && (s == substr || len(s) > 0 && containsHelper(s, substr))
+}
+
+func containsHelper(s, substr string) bool {
+	for i := 0; i <= len(s)-len(substr); i++ {
+		if s[i:i+len(substr)] == substr {
+			return true
+		}
+	}
+	return false
+}
--- a/internal/ai/service.go
+++ b/internal/ai/service.go
@ -241,6 +241,17 @@ func (s *Service) SetRemediationLog(remLog *RemediationLog) {
 	}
 }

+// SetPatternDetector sets the pattern detector for failure prediction
+func (s *Service) SetPatternDetector(detector *PatternDetector) {
+	s.mu.RLock()
+	patrol := s.patrolService
+	s.mu.RUnlock()
+
+	if patrol != nil {
+		patrol.SetPatternDetector(detector)
+	}
+}
+
 // StartPatrol starts the background patrol service
 func (s *Service) StartPatrol(ctx context.Context) {
 	s.mu.RLock()
--- a/internal/api/ai_handlers.go
+++ b/internal/api/ai_handlers.go
@ -115,6 +115,11 @@ func (h *AISettingsHandler) SetRemediationLog(remLog *ai.RemediationLog) {
 	h.aiService.SetRemediationLog(remLog)
 }

+// SetPatternDetector sets the pattern detector for failure prediction
+func (h *AISettingsHandler) SetPatternDetector(detector *ai.PatternDetector) {
+	h.aiService.SetPatternDetector(detector)
+}
+
 // StopPatrol stops the background AI patrol service
 func (h *AISettingsHandler) StopPatrol() {
 	h.aiService.StopPatrol()
--- a/internal/api/router.go
+++ b/internal/api/router.go
@ -1458,6 +1458,18 @@ func (r *Router) StartPatrol(ctx context.Context) {
 		if remediationLog != nil {
 			r.aiSettingsHandler.SetRemediationLog(remediationLog)
 		}
+		
+		// Initialize pattern detector for failure prediction
+		patternDetector := ai.NewPatternDetector(ai.PatternDetectorConfig{
+			MaxEvents:       5000,
+			MinOccurrences:  3,
+			PatternWindow:   90 * 24 * time.Hour,
+			PredictionLimit: 30 * 24 * time.Hour,
+			DataDir:         dataDir,
+		})
+		if patternDetector != nil {
+			r.aiSettingsHandler.SetPatternDetector(patternDetector)
+		}

 		r.aiSettingsHandler.StartPatrol(ctx)
 	}