Aiops

AIOps - AI for IT Operations

This skill provides comprehensive patterns for implementing AIOps strategies in 2025, including intelligent monitoring, automated incident response, predictive analytics, and observability best practices. The patterns are designed to be framework-agnostic and applicable across different infrastructure platforms.

When to Use This Skill

Use this skill when you need to:

• Implement AIOps strategies for modern infrastructure
• Build intelligent monitoring and alerting systems
• Create automated incident response workflows
• Deploy predictive maintenance solutions
• Implement self-healing capabilities
• Build observability platforms with AI/ML
• Optimize multi-cloud operations
• Create chaos engineering practices
• Implement generative AI for operations
• Build digital twins for infrastructure

1. AIOps Architecture Patterns

Core AIOps Platform Architecture

# aiops/core/architecture.py
from abc import ABC, abstractmethod
from typing import Dict, List, Optional, Any, Union
from dataclasses import dataclass, field
from datetime import datetime, timedelta
from enum import Enum
import asyncio
import json
import logging

logger = logging.getLogger(__name__)

class AlertSeverity(str, Enum):
    """Alert severity levels"""
    CRITICAL = "critical"
    HIGH = "high"
    MEDIUM = "medium"
    LOW = "low"
    INFO = "info"

class IncidentStatus(str, Enum):
    """Incident status"""
    OPEN = "open"
    INVESTIGATING = "investigating"
    IDENTIFIED = "identified"
    MONITORING = "monitoring"
    RESOLVED = "resolved"
    CLOSED = "closed"

@dataclass
class Metric:
    """Metric data point"""
    name: str
    value: float
    timestamp: datetime
    labels: Dict[str, str] = field(default_factory=dict)
    source: str = ""
    unit: str = ""

@dataclass
class LogEntry:
    """Structured log entry"""
    timestamp: datetime
    level: str
    message: str
    service: str
    trace_id: Optional[str] = None
    span_id: Optional[str] = None
    labels: Dict[str, str] = field(default_factory=dict)
    stack_trace: Optional[str] = None

@dataclass
class Trace:
    """Distributed trace data"""
    trace_id: str
    spans: List[Dict[str, Any]]
    duration: timedelta
    service_map: Dict[str, List[str]]
    error_count: int = 0

@dataclass
class Alert:
    """Alert representation"""
    id: str
    name: str
    severity: AlertSeverity
    status: IncidentStatus
    description: str
    source: str
    timestamp: datetime
    labels: Dict[str, str] = field(default_factory=dict)
    annotations: Dict[str, str] = field(default_factory=dict)
    fingerprint: str = ""
    related_entities: List[str] = field(default_factory=list)

class DataSource(ABC):
    """Abstract data source interface"""

    @abstractmethod
    async def collect_metrics(
        self,
        query: str,
        start_time: datetime,
        end_time: datetime
    ) -> List[Metric]:
        """Collect metrics from data source"""
        pass

    @abstractmethod
    async def query_logs(
        self,
        query: str,
        start_time: datetime,
        end_time: datetime,
        limit: int = 100
    ) -> List[LogEntry]:
        """Query logs from data source"""
        pass

    @abstractmethod
    async def get_trace(self, trace_id: str) -> Optional[Trace]:
        """Get trace data"""
        pass

class AIOpsEngine:
    """Core AIOps processing engine"""

    def __init__(self):
        self.data_sources: List[DataSource] = []
        self.alert_processors: List[AlertProcessor] = []
        self.ml_models: Dict[str, Any] = {}
        self.knowledge_base: KnowledgeBase = KnowledgeBase()
        self.automation_engine = AutomationEngine()

    def register_data_source(self, data_source: DataSource) -> None:
        """Register a data source"""
        self.data_sources.append(data_source)
        logger.info(f"Registered data source: {type(data_source).__name__}")

    def register_alert_processor(self, processor: 'AlertProcessor') -> None:
        """Register an alert processor"""
        self.alert_processors.append(processor)
        logger.info(f"Registered alert processor: {type(processor).__name__}")

    async def process_alert(self, alert: Alert) -> Dict[str, Any]:
        """Process incoming alert through AIOps pipeline"""
        logger.info(f"Processing alert: {alert.id} - {alert.name}")

        # 1. Enrich alert with context
        enriched_alert = await self._enrich_alert(alert)

        # 2. Run through ML models for classification and prediction
        analysis = await self._analyze_alert(enriched_alert)

        # 3. Determine appropriate response
        response_plan = await self._generate_response_plan(enriched_alert, analysis)

        # 4. Execute automated actions if applicable
        if response_plan.get("auto_execute", False):
            await self._execute_response(enriched_alert, response_plan)

        # 5. Update knowledge base
        await self.knowledge_base.update_from_alert(enriched_alert, analysis, response_plan)

        return {
            "alert_id": alert.id,
            "status": "processed",
            "analysis": analysis,
            "response_plan": response_plan
        }

    async def _enrich_alert(self, alert: Alert) -> Alert:
        """Enrich alert with additional context"""
        # Get related metrics
        for source in self.data_sources:
            try:
                metrics = await source.collect_metrics(
                    query=f"{alert.name}[5m]",
                    start_time=alert.timestamp - timedelta(minutes=5),
                    end_time=alert.timestamp
                )

                # Add metrics to alert annotations
                alert.annotations[f"metrics_{type(source).__name__}"] = json.dumps([
                    {"name": m.name, "value": m.value} for m in metrics
                ])
            except Exception as e:
                logger.error(f"Failed to enrich alert with metrics: {e}")

        # Get related logs
        for source in self.data_sources:
            try:
                logs = await source.query_logs(
                    query=f"service:{alert.source} level:error",
                    start_time=alert.timestamp - timedelta(minutes=5),
                    end_time=alert.timestamp,
                    limit=10
                )

                # Add logs to alert context
                alert.annotations["recent_errors"] = json.dumps([
                    {"timestamp": l.timestamp.isoformat(), "message": l.message}
                    for l in logs
                ])
            except Exception as e:
                logger.error(f"Failed to enrich alert with logs: {e}")

        return alert

    async def _analyze_alert(self, alert: Alert) -> Dict[str, Any]:
        """Analyze alert using ML models"""
        analysis = {
            "classification": None,
            "severity_score": 0.0,
            "predicted_impact": "low",
            "recommendations": [],
            "related_incidents": []
        }

        # Classification model
        if "classification" in self.ml_models:
            features = self._extract_features(alert)
            analysis["classification"] = self.ml_models["classification"].predict([features])[0]

        # Severity prediction
        if "severity" in self.ml_models:
            features = self._extract_features(alert)
            analysis["severity_score"] = self.ml_models["severity"].predict_proba([features])[0][1]

        # Impact prediction
        if "impact" in self.ml_models:
            features = self._extract_features(alert)
            analysis["predicted_impact"] = self.ml_models["impact"].predict([features])[0]

        # Find related incidents from knowledge base
        analysis["related_incidents"] = await self.knowledge_base.find_similar_incidents(alert)

        return analysis

    async def _generate_response_plan(
        self,
        alert: Alert,
        analysis: Dict[str, Any]
    ) -> Dict[str, Any]:
        """Generate automated response plan"""
        plan = {
            "priority": alert.severity,
            "actions": [],
            "escalation_rules": [],
            "auto_execute": False,
            "estimated_resolution_time": None
        }

        # Determine if auto-execution is safe
        if (analysis["classification"] in ["known_issue", "false_positive"] and
            alert.severity in [AlertSeverity.LOW, AlertSeverity.MEDIUM]):
            plan["auto_execute"] = True

        # Generate actions based on classification
        classification = analysis.get("classification")

        if classification == "resource_exhaustion":
            plan["actions"].append({
                "type": "scale",
                "target": alert.source,
                "params": {"replicas": "+20%"}
            })
            plan["actions"].append({
                "type": "notify",
                "channel": "devops",
                "message": f"Resource exhaustion detected in {alert.source}"
            })

        elif classification == "known_issue":
            # Apply known fix
            known_fix = await self.knowledge_base.get_known_fix(alert.name)
            if known_fix:
                plan["actions"].append(known_fix)

        elif classification == "false_positive":
            plan["actions"].append({
                "type": "suppress",
                "duration": "1h",
                "reason": "False positive detected"
            })

        # Add escalation rules
        if alert.severity == AlertSeverity.CRITICAL:
            plan["escalation_rules"].append({
                "condition": "no_response_5m",
                "action": "escalate_to_oncall",
                "timeout": 300
            })

        return plan

    def _extract_features(self, alert: Alert) -> List[float]:
        """Extract features for ML models"""
        # Simple feature extraction - in production, use more sophisticated methods
        features = [
            float(alert.severity == AlertSeverity.CRITICAL),
            float(alert.severity == AlertSeverity.HIGH),
            float(alert.severity == AlertSeverity.MEDIUM),
            len(alert.description),
            len(alert.labels),
            alert.timestamp.hour / 24.0,
            alert.timestamp.weekday() / 7.0
        ]
        return features

    async def _execute_response(
        self,
        alert: Alert,
        response_plan: Dict[str, Any]
    ) -> None:
        """Execute automated response actions"""
        for action in response_plan.get("actions", []):
            try:
                await self.automation_engine.execute(action)
                logger.info(f"Executed action for alert {alert.id}: {action['type']}")
            except Exception as e:
                logger.error(f"Failed to execute action: {e}")

Monitoring Platform Integrations

# aiops/integrations/monitoring.py
import aiohttp
import asyncio
from typing import List, Dict, Any, Optional
from datetime import datetime
from ..core.architecture import DataSource, Metric, LogEntry, Trace, Alert

class PrometheusDataSource(DataSource):
    """Prometheus metrics integration"""

    def __init__(self, url: str):
        self.url = url.rstrip("/")
        self.session = None

    async def _get_session(self) -> aiohttp.ClientSession:
        if not self.session:
            self.session = aiohttp.ClientSession()
        return self.session

    async def collect_metrics(
        self,
        query: str,
        start_time: datetime,
        end_time: datetime
    ) -> List[Metric]:
        """Query Prometheus for metrics"""
        session = await self._get_session()

        params = {
            "query": query,
            "start": start_time.timestamp(),
            "end": end_time.timestamp(),
            "step": "15s"
        }

        url = f"{self.url}/api/v1/query_range"

        async with session.get(url, params=params) as response:
            data = await response.json()
            metrics = []

            if data["status"] == "success":
                result = data["data"]
                if result["resultType"] == "matrix":
                    for series in result["result"]:
                        labels = series["metric"]
                        values = series["values"]

                        for timestamp, value in values:
                            metrics.append(Metric(
                                name=query,
                                value=float(value),
                                timestamp=datetime.fromtimestamp(float(timestamp)),
                                labels=labels,
                                source="prometheus"
                            ))

            return metrics

    async def query_logs(
        self,
        query: str,
        start_time: datetime,
        end_time: datetime,
        limit: int = 100
    ) -> List[LogEntry]:
        """Prometheus doesn't store logs - use Loki integration"""
        return []

    async def get_trace(self, trace_id: str) -> Optional[Trace]:
        """Prometheus doesn't store traces - use Jaeger/Tempo integration"""
        return None

class LokiDataSource(DataSource):
    """Loki log aggregation integration"""

    def __init__(self, url: str):
        self.url = url.rstrip("/")
        self.session = None

    async def _get_session(self) -> aiohttp.ClientSession:
        if not self.session:
            self.session = aiohttp.ClientSession()
        return self.session

    async def collect_metrics(
        self,
        query: str,
        start_time: datetime,
        end_time: datetime
    ) -> List[Metric]:
        """Loki doesn't store metrics"""
        return []

    async def query_logs(
        self,
        query: str,
        start_time: datetime,
        end_time: datetime,
        limit: int = 100
    ) -> List[LogEntry]:
        """Query Loki for logs"""
        session = await self._get_session()

        params = {
            "query": query,
            "start": start_time.isoformat(),
            "end": end_time.isoformat(),
            "limit": str(limit)
        }

        url = f"{self.url}/loki/api/v1/query_range"

        async with session.get(url, params=params) as response:
            data = await response.json()
            logs = []

            if data["status"] == "success":
                result = data["data"]
                for series in result["result"]:
                    stream = series["stream"]
                    for timestamp, line in series["values"]:
                        logs.append(LogEntry(
                            timestamp=datetime.fromtimestamp(float(timestamp) / 1e9),
                            level=stream.get("level", "info"),
                            message=line,
                            service=stream.get("service", "unknown"),
                            labels=stream
                        ))

            return logs

    async def get_trace(self, trace_id: str) -> Optional[Trace]:
        """Loki doesn't store traces"""
        return None

class JaegerDataSource(DataSource):
    """Jaeger distributed tracing integration"""

    def __init__(self, url: str):
        self.url = url.rstrip("/")
        self.session = None

    async def _get_session(self) -> aiohttp.ClientSession:
        if not self.session:
            self.session = aiohttp.ClientSession()
        return self.session

    async def collect_metrics(
        self,
        query: str,
        start_time: datetime,
        end_time: datetime
    ) -> List[Metric]:
        """Jaeger doesn't store metrics"""
        return []

    async def query_logs(
        self,
        query: str,
        start_time: datetime,
        end_time: datetime,
        limit: int = 100
    ) -> List[LogEntry]:
        """Jaeger doesn't store logs"""
        return []

    async def get_trace(self, trace_id: str) -> Optional[Trace]:
        """Get trace from Jaeger"""
        session = await self._get_session()

        url = f"{self.url}/api/traces/{trace_id}"

        async with session.get(url) as response:
            data = await response.json()

            if data["data"] and data["data"][0]["spans"]:
                spans = data["data"][0]["spans"]
                service_map = {}
                error_count = 0

                for span in spans:
                    service = span["process"]["serviceName"]
                    if service not in service_map:
                        service_map[service] = []
                    service_map[service].append(span["spanID"])

                    if span.get("tags"):
                        for tag in span["tags"]:
                            if tag.get("key") == "error" and tag.get("value"):
                                error_count += 1

                return Trace(
                    trace_id=trace_id,
                    spans=spans,
                    duration=timedelta(
                        microseconds=data["data"][0]["duration"]
                    ),
                    service_map=service_map,
                    error_count=error_count
                )

            return None

class DatadogDataSource(DataSource):
    """Datadog APM integration"""

    def __init__(self, api_key: str, app_key: str, site: str = "datadoghq.com"):
        self.api_key = api_key
        self.app_key = app_key
        self.site = site
        self.base_url = f"https://api.{site}"
        self.session = None

    async def _get_session(self) -> aiohttp.ClientSession:
        if not self.session:
            headers = {
                "DD-API-KEY": self.api_key,
                "DD-APPLICATION-KEY": self.app_key
            }
            self.session = aiohttp.ClientSession(headers=headers)
        return self.session

    async def collect_metrics(
        self,
        query: str,
        start_time: datetime,
        end_time: datetime
    ) -> List[Metric]:
        """Query Datadog metrics"""
        session = await self._get_session()

        params = {
            "query": query,
            "from": int(start_time.timestamp()),
            "to": int(end_time.timestamp())
        }

        url = f"{self.base_url}/api/v1/query"

        async with session.get(url, params=params) as response:
            data = await response.json()
            metrics = []

            if "series" in data:
                for series in data["series"]:
                    for point in series["pointlist"]:
                        metrics.append(Metric(
                            name=series["metric"],
                            value=point[1],
                            timestamp=datetime.fromtimestamp(point[0]),
                            labels={k: str(v) for k, v in series["tag_set"].items()},
                            source="datadog",
                            unit=series.get("unit", "")
                        ))

            return metrics

    async def query_logs(
        self,
        query: str,
        start_time: datetime,
        end_time: datetime,
        limit: int = 100
    ) -> List[LogEntry]:
        """Query Datadog logs"""
        session = await self._get_session()

        payload = {
            "query": query,
            "time": {
                "from": start_time.isoformat(),
                "to": end_time.isoformat()
            },
            "limit": limit
        }

        url = f"{self.base_url}/api/v2/logs/events/search"

        async with session.post(url, json=payload) as response:
            data = await response.json()
            logs = []

            for log in data.get("data", []):
                attributes = log.get("attributes", {})

                logs.append(LogEntry(
                    timestamp=datetime.fromisoformat(attributes["timestamp"]),
                    level=attributes.get("status", "info"),
                    message=attributes.get("message", ""),
                    service=attributes.get("service", "unknown"),
                    trace_id=attributes.get("trace_id"),
                    labels=attributes.get("tags", [])
                ))

            return logs

    async def get_trace(self, trace_id: str) -> Optional[Trace]:
        """Get trace from Datadog"""
        session = await self._get_session()

        url = f"{self.base_url}/api/v1/trace/{trace_id}"

        async with session.get(url) as response:
            data = await response.json()

            # Parse Datadog trace format
            # Implementation depends on Datadog API response structure
            return None

2. Machine Learning for Operations

Anomaly Detection Models

# aiops/ml/anomaly_detection.py
import numpy as np
import pandas as pd
from typing import List, Dict, Tuple, Optional
from datetime import datetime, timedelta
from sklearn.ensemble import IsolationForest
from sklearn.preprocessing import StandardScaler
from sklearn.decomposition import PCA
import joblib
import asyncio
from collections import deque

class AnomalyDetector:
    """Generic anomaly detection for metrics and logs"""

    def __init__(self, model_type: str = "isolation_forest"):
        self.model_type = model_type
        self.models: Dict[str, Any] = {}
        self.scalers: Dict[str, StandardScaler] = {}
        self.is_trained: Dict[str, bool] = {}
        self.feature_history: Dict[str, deque] = {}

    async def detect_metric_anomaly(
        self,
        metric_name: str,
        current_value: float,
        historical_values: List[float],
        window_size: int = 100
    ) -> Tuple[bool, float]:
        """Detect anomalies in time series metrics"""
        if metric_name not in self.models:
            await self._train_metric_model(metric_name, historical_values)

        # Prepare features
        features = self._extract_time_series_features(
            current_value,
            historical_values,
            window_size
        )

        # Scale features
        if metric_name in self.scalers:
            features = self.scalers[metric_name].transform([features])

        # Predict anomaly
        model = self.models[metric_name]
        anomaly_score = model.decision_function([features])[0]
        is_anomaly = model.predict([features])[0] == -1

        return is_anomaly, float(anomaly_score)

    async def detect_log_anomaly(
        self,
        log_entries: List[Dict[str, Any]]
    ) -> List[Dict[str, Any]]:
        """Detect anomalies in log patterns"""
        # Extract log features
        features = []
        for entry in log_entries:
            features.append([
                len(entry.get("message", "")),
                entry.get("level_numeric", 0),
                entry.get("hour_of_day", 0),
                entry.get("day_of_week", 0)
            ])

        features = np.array(features)

        # Train or use existing model
        model_name = "log_patterns"
        if model_name not in self.models:
            self.models[model_name] = IsolationForest(
                contamination=0.1,
                random_state=42
            )
            self.models[model_name].fit(features)

        # Predict anomalies
        predictions = self.models[model_name].predict(features)
        scores = self.models[model_name].decision_function(features)

        # Return anomalous logs
        anomalies = []
        for i, (pred, score) in enumerate(zip(predictions, scores)):
            if pred == -1:  # Anomaly
                anomalies.append({
                    "log_index": i,
                    "log_entry": log_entries[i],
                    "anomaly_score": float(score),
                    "reason": "Unusual log pattern detected"
                })

        return anomalies

    async def detect_performance_anomaly(
        self,
        metrics: Dict[str, List[float]],
        threshold: float = 0.1
    ) -> Dict[str, Dict[str, Any]]:
        """Detect performance anomalies across multiple metrics"""
        results = {}

        for metric_name, values in metrics.items():
            if len(values) < 10:
                continue

            # Calculate statistical anomalies
            mean_val = np.mean(values)
            std_val = np.std(values)
            recent_values = values[-5:]

            for i, value in enumerate(recent_values):
                z_score = abs(value - mean_val) / (std_val + 1e-8)

                if z_score > 3:  # Statistical outlier
                    if metric_name not in results:
                        results[metric_name] = {
                            "anomalies": [],
                            "severity": "medium"
                        }

                    results[metric_name]["anomalies"].append({
                        "value": value,
                        "z_score": z_score,
                        "index": i,
                        "type": "statistical"
                    })

        return results

    async def _train_metric_model(
        self,
        metric_name: str,
        values: List[float]
    ) -> None:
        """Train anomaly detection model for a metric"""
        if len(values) < 30:
            # Not enough data for training
            return

        # Prepare training data
        training_features = []
        for i in range(10, len(values)):
            # Create window of previous values
            window = values[i-10:i]
            features = [
                np.mean(window),
                np.std(window),
                values[i],  # Current value
                (values[i] - np.mean(window)) / (np.std(window) + 1e-8),  # Z-score
                max(window) - min(window),  # Range
            ]
            training_features.append(features)

        training_features = np.array(training_features)

        # Scale features
        scaler = StandardScaler()
        scaled_features = scaler.fit_transform(training_features)
        self.scalers[metric_name] = scaler

        # Train model
        if self.model_type == "isolation_forest":
            model = IsolationForest(
                contamination=0.05,
                random_state=42
            )
            model.fit(scaled_features)
        else:
            # Add other model types as needed
            model = IsolationForest(contamination=0.05, random_state=42)
            model.fit(scaled_features)

        self.models[metric_name] = model
        self.is_trained[metric_name] = True

    def _extract_time_series_features(
        self,
        current_value: float,
        historical_values: List[float],
        window_size: int
    ) -> List[float]:
        """Extract features from time series data"""
        if len(historical_values) < window_size:
            window = historical_values
        else:
            window = historical_values[-window_size:]

        features = [
            np.mean(window),
            np.std(window),
            min(window),
            max(window),
            current_value,
            (current_value - np.mean(window)) / (np.std(window) + 1e-8),
            len([v for v in window if v > np.mean(window)]) / len(window),  # Percentage above mean
        ]

        return features

class PredictiveAnalyzer:
    """Predictive analytics for capacity planning and incident prediction"""

    def __init__(self):
        self.models: Dict[str, Any] = {}

    async def predict_capacity_needs(
        self,
        metric_history: Dict[str, List[float]],
        forecast_horizon: int = 7
    ) -> Dict[str, Dict[str, Any]]:
        """Predict future resource needs"""
        predictions = {}

        for metric_name, values in metric_history.items():
            if len(values) < 30:
                continue

            # Simple linear trend prediction (use more sophisticated models in production)
            x = np.arange(len(values))
            coeffs = np.polyfit(x, values, 1)

            # Predict future values
            future_x = np.arange(len(values), len(values) + forecast_horizon * 24)
            future_values = np.polyval(coeffs, future_x)

            # Calculate growth rate
            growth_rate = (future_values[-1] - values[-1]) / values[-1] * 100

            predictions[metric_name] = {
                "current": values[-1],
                "predicted_peak": float(max(future_values)),
                "predicted_average": float(np.mean(future_values)),
                "growth_rate": growth_rate,
                "recommendation": self._get_capacity_recommendation(growth_rate),
                "confidence": 0.75  # Should be calculated based on model accuracy
            }

        return predictions

    async def predict_incidents(
        self,
        alert_history: List[Dict[str, Any]],
        system_metrics: Dict[str, List[float]]
    ) -> Dict[str, Any]:
        """Predict likelihood of future incidents"""
        # Feature engineering
        features = self._extract_incident_features(alert_history, system_metrics)

        # Simplified risk score calculation
        risk_factors = {
            "error_rate_trend": self._calculate_trend(
                system_metrics.get("error_rate", [])
            ),
            "cpu_utilization_avg": np.mean(
                system_metrics.get("cpu_utilization", [0])
            ),
            "memory_utilization_avg": np.mean(
                system_metrics.get("memory_utilization", [0])
            ),
            "recent_incident_count": len([
                a for a in alert_history[-24:]
                if a.get("severity") == "critical"
            ]),
        }

        # Calculate risk score (0-100)
        risk_score = min(100, max(0, (
            risk_factors["error_rate_trend"] * 30 +
            risk_factors["cpu_utilization_avg"] * 25 +
            risk_factors["memory_utilization_avg"] * 25 +
            risk_factors["recent_incident_count"] * 20
        )))

        return {
            "risk_score": risk_score,
            "risk_level": self._get_risk_level(risk_score),
            "contributing_factors": risk_factors,
            "recommendations": self._get_mitigation_recommendations(risk_factors)
        }

    def _extract_incident_features(
        self,
        alert_history: List[Dict[str, Any]],
        system_metrics: Dict[str, List[float]]
    ) -> List[float]:
        """Extract features for incident prediction"""
        features = []

        # Alert patterns
        recent_alerts = alert_history[-24:] if len(alert_history) >= 24 else alert_history
        features.append(len(recent_alerts))
        features.append(len([a for a in recent_alerts if a.get("severity") == "critical"]))

        # System metrics
        for metric_name in ["cpu", "memory", "disk", "network"]:
            values = system_metrics.get(f"{metric_name}_utilization", [0])
            features.append(np.mean(values[-12:]) if values else 0)
            features.append(np.std(values[-12:]) if values else 0)

        return features

    def _calculate_trend(self, values: List[float]) -> float:
        """Calculate trend coefficient (positive = increasing)"""
        if len(values) < 2:
            return 0.0

        x = np.arange(len(values))
        coeffs = np.polyfit(x, values, 1)
        return coeffs[0]

    def _get_capacity_recommendation(self, growth_rate: float) -> str:
        """Get capacity recommendation based on growth rate"""
        if growth_rate > 50:
            return "Immediate scaling required - consider 2x current capacity"
        elif growth_rate > 20:
            return "Plan for scaling within next week - recommend 1.5x capacity"
        elif growth_rate > 0:
            return "Monitor growth - consider 25% buffer"
        else:
            return "Capacity adequate - no immediate action needed"

    def _get_risk_level(self, risk_score: float) -> str:
        """Get risk level from risk score"""
        if risk_score >= 80:
            return "critical"
        elif risk_score >= 60:
            return "high"
        elif risk_score >= 40:
            return "medium"
        else:
            return "low"

    def _get_mitigation_recommendations(
        self,
        risk_factors: Dict[str, float]
    ) -> List[str]:
        """Get mitigation recommendations based on risk factors"""
        recommendations = []

        if risk_factors["cpu_utilization_avg"] > 80:
            recommendations.append("Scale CPU resources or optimize CPU usage")

        if risk_factors["memory_utilization_avg"] > 80:
            recommendations.append("Increase memory allocation or check for memory leaks")

        if risk_factors["error_rate_trend"] > 0:
            recommendations.append("Investigate and fix increasing error rate")

        if risk_factors["recent_incident_count"] > 3:
            recommendations.append("Review recent incidents for root cause analysis")

        return recommendations

class AILogAnalyzer:
    """AI-powered log analysis and pattern detection"""

    def __init__(self):
        self.patterns: Dict[str, List[str]] = {}
        self.embeddings: Dict[str, np.ndarray] = {}

    async def analyze_logs(
        self,
        logs: List[LogEntry]
    ) -> Dict[str, Any]:
        """Analyze logs for patterns and anomalies"""
        analysis = {
            "total_logs": len(logs),
            "error_rate": 0.0,
            "patterns": [],
            "anomalies": [],
            "trending_topics": [],
            "recommendations": []
        }

        # Calculate error rate
        error_count = len([l for l in logs if l.level.lower() in ["error", "critical"]])
        analysis["error_rate"] = error_count / len(logs) * 100

        # Extract patterns using regex and ML
        analysis["patterns"] = await self._extract_patterns(logs)

        # Find anomalies
        analysis["anomalies"] = await self._detect_log_anomalies(logs)

        # Identify trending topics/errors
        analysis["trending_topics"] = await self._identify_trending_topics(logs)

        # Generate recommendations
        analysis["recommendations"] = self._generate_log_recommendations(analysis)

        return analysis

    async def _extract_patterns(self, logs: List[LogEntry]) -> List[Dict[str, Any]]:
        """Extract common patterns from logs"""
        # Simplified pattern extraction
        # In production, use more sophisticated NLP techniques
        patterns = []

        # Group logs by service
        service_logs = {}
        for log in logs:
            if log.service not in service_logs:
                service_logs[log.service] = []
            service_logs[log.service].append(log)

        # Find error patterns
        for service, service_log_list in service_logs.items():
            error_logs = [l for l in service_log_list if l.level.lower() == "error"]
            if len(error_logs) > 0:
                # Find common substrings in error messages
                common_errors = {}
                for log in error_logs:
                    # Simplified - use actual pattern matching in production
                    for word in log.message.split():
                        if len(word) > 5:  # Only consider longer words
                            common_errors[word] = common_errors.get(word, 0) + 1

                if common_errors:
                    most_common = max(common_errors, key=common_errors.get)
                    patterns.append({
                        "service": service,
                        "type": "error_pattern",
                        "pattern": most_common,
                        "count": common_errors[most_common],
                        "percentage": common_errors[most_common] / len(error_logs) * 100
                    })

        return patterns

    async def _detect_log_anomalies(self, logs: List[LogEntry]) -> List[Dict[str, Any]]:
        """Detect anomalies in logs"""
        anomalies = []

        # Check for unusual error spikes
        error_by_minute = {}
        for log in logs:
            if log.level.lower() in ["error", "critical"]:
                minute = log.timestamp.replace(second=0, microsecond=0)
                error_by_minute[minute] = error_by_minute.get(minute, 0) + 1

        if error_by_minute:
            avg_errors = np.mean(list(error_by_minute.values()))
            for minute, count in error_by_minute.items():
                if count > avg_errors * 3:  # 3x spike
                    anomalies.append({
                        "type": "error_spike",
                        "timestamp": minute.isoformat(),
                        "value": count,
                        "baseline": avg_errors,
                        "severity": "high"
                    })

        return anomalies

    async def _identify_trending_topics(self, logs: List[LogEntry]) -> List[Dict[str, Any]]:
        """Identify trending topics in logs"""
        # Simplified trending detection
        word_count = {}
        for log in logs[-100:]:  # Last 100 logs
            for word in log.message.lower().split():
                if len(word) > 5 and word.isalpha():
                    word_count[word] = word_count.get(word, 0) + 1

        # Get top trending words
        trending = sorted(word_count.items(), key=lambda x: x[1], reverse=True)[:10]

        return [
            {"topic": word, "count": count}
            for word, count in trending
        ]

    def _generate_log_recommendations(self, analysis: Dict[str, Any]) -> List[str]:
        """Generate recommendations based on log analysis"""
        recommendations = []

        if analysis["error_rate"] > 5:
            recommendations.append(
                f"High error rate ({analysis['error_rate']:.1f}%) detected - investigate immediately"
            )

        if len(analysis["anomalies"]) > 0:
            recommendations.append(
                f"Found {len(analysis['anomalies'])} anomalies - review and address"
            )

        for pattern in analysis["patterns"][:3]:
            if pattern["percentage"] > 30:
                recommendations.append(
                    f"Common error pattern in {pattern['service']}: {pattern['pattern']}"
                )

        return recommendations

3. Automation and Self-Healing

Automation Engine

# aiops/automation/engine.py
import asyncio
import subprocess
from typing import Dict, List, Any, Optional, Callable
from dataclasses import dataclass
from enum import Enum
import logging
from datetime import datetime, timedelta

logger = logging.getLogger(__name__)

class ActionType(str, Enum):
    """Types of automation actions"""
    SCALE = "scale"
    RESTART = "restart"
    PATCH = "patch"
    NOTIFY = "notify"
    SUPPRESS = "suppress"
    ISOLATE = "isolate"
    FAILBACK = "failback"

@dataclass
class Action:
    """Automation action definition"""
    type: ActionType
    target: str
    params: Dict[str, Any]
    conditions: List[str] = None
    timeout: int = 300
    retries: int = 3

class AutomationEngine:
    """Automation engine for executing remediation actions"""

    def __init__(self):
        self.actions: Dict[str, Action] = {}
        self.execution_history: List[Dict[str, Any]] = []
        self.action_handlers: Dict[ActionType, Callable] = {
            ActionType.SCALE: self._handle_scale,
            ActionType.RESTART: self._handle_restart,
            ActionType.PATCH: self._handle_patch,
            ActionType.NOTIFY: self._handle_notify,
            ActionType.SUPPRESS: self._handle_suppress,
            ActionType.ISOLATE: self._handle_isolate,
            ActionType.FAILBACK: self._handle_failback
        }

    def register_action(self, action: Action) -> None:
        """Register an automation action"""
        self.actions[action.target] = action
        logger.info(f"Registered action: {action.type} for {action.target}")

    async def execute(self, action: Action) -> Dict[str, A

…(truncated)