demo/scenarios.py

"""
Incident scenarios for the demo - EXPANDED VERSION
"""

INCIDENT_SCENARIOS = {
    "Cache Miss Storm": {
        "description": "Redis cluster experiencing 80% cache miss rate causing database overload",
        "severity": "CRITICAL",
        "component": "redis_cache",
        "metrics": {
            "cache_hit_rate": 18.5,
            "database_load": 92,
            "response_time_ms": 1850,
            "affected_users": 45000,
            "eviction_rate_per_sec": 125
        },
        "business_impact": {
            "revenue_loss_per_hour": 8500,
            "sla_violation": True,
            "customer_sat_change": -40,
            "affected_services": ["API Gateway", "User Service", "Payment"]
        },
        "roi_data": {
            "hourly_revenue_loss": 8500,
            "manual_recovery_hours": 1.0,
            "enterprise_recovery_hours": 0.2,
            "engineers_required": 4,
            "engineer_hourly_rate": 150,
            "estimated_monthly_occurrences": 2,
            "enterprise_savings_percentage": 0.85
        }
    },
    
    "Database Connection Pool Exhaustion": {
        "description": "PostgreSQL connection pool exhausted causing API timeouts",
        "severity": "HIGH",
        "component": "postgresql_database",
        "metrics": {
            "active_connections": 98,
            "max_connections": 100,
            "api_latency_ms": 2450,
            "error_rate": 15.2,
            "queue_depth": 1250,
            "connection_wait_seconds": 45
        },
        "business_impact": {
            "revenue_loss_per_hour": 4200,
            "affected_services": ["API Gateway", "User Service", "Payment Service"],
            "sla_violation": True,
            "partner_api_impact": 3
        },
        "roi_data": {
            "hourly_revenue_loss": 4200,
            "manual_recovery_hours": 0.75,
            "enterprise_recovery_hours": 0.13,
            "engineers_required": 2,
            "engineer_hourly_rate": 150,
            "estimated_monthly_occurrences": 3,
            "enterprise_savings_percentage": 0.82
        }
    },
    
    "Kubernetes Memory Leak": {
        "description": "Java microservice memory leak causing pod restarts",
        "severity": "HIGH",
        "component": "java_payment_service",
        "metrics": {
            "memory_usage": 96,
            "gc_pause_time_ms": 4500,
            "error_rate": 28.5,
            "restart_frequency_per_hour": 12,
            "heap_fragmentation": 42
        },
        "business_impact": {
            "revenue_loss_per_hour": 5500,
            "session_loss": 8500,
            "payment_failures_percentage": 3.2,
            "support_tickets_increase": 300
        },
        "roi_data": {
            "hourly_revenue_loss": 5500,
            "manual_recovery_hours": 1.5,
            "enterprise_recovery_hours": 0.25,
            "engineers_required": 3,
            "engineer_hourly_rate": 150,
            "estimated_monthly_occurrences": 1,
            "enterprise_savings_percentage": 0.79
        }
    },
    
    "API Rate Limit Storm": {
        "description": "Third-party API rate limiting causing cascading failures",
        "severity": "MEDIUM",
        "component": "external_api_gateway",
        "metrics": {
            "rate_limit_hits_percentage": 95,
            "error_rate": 42.8,
            "retry_storm": True,
            "cascade_effect_services": 3,
            "queue_backlog": 8500
        },
        "business_impact": {
            "revenue_loss_per_hour": 3800,
            "partner_sla_breach": True,
            "data_sync_delay_hours": 4,
            "customer_reports_delay_hours": 6
        },
        "roi_data": {
            "hourly_revenue_loss": 3800,
            "manual_recovery_hours": 1.25,
            "enterprise_recovery_hours": 0.17,
            "engineers_required": 3,
            "engineer_hourly_rate": 150,
            "estimated_monthly_occurrences": 4,
            "enterprise_savings_percentage": 0.85
        }
    },
    
    "Network Partition": {
        "description": "Network partition causing split-brain in distributed database",
        "severity": "CRITICAL",
        "component": "distributed_database",
        "metrics": {
            "partition_detected": True,
            "write_conflicts": 1250,
            "data_inconsistency_percentage": 8.5,
            "replication_lag_seconds": 45,
            "quorum_lost": True
        },
        "business_impact": {
            "revenue_loss_per_hour": 12000,
            "data_corruption_risk": True,
            "recovery_complexity": "HIGH",
            "compliance_violation": True
        },
        "roi_data": {
            "hourly_revenue_loss": 12000,
            "manual_recovery_hours": 2.0,
            "enterprise_recovery_hours": 0.3,
            "engineers_required": 5,
            "engineer_hourly_rate": 150,
            "estimated_monthly_occurrences": 0.5,
            "enterprise_savings_percentage": 0.88
        }
    },
    
    "Storage I/O Saturation": {
        "description": "Storage system I/O saturation causing application timeouts",
        "severity": "HIGH",
        "component": "storage_cluster",
        "metrics": {
            "io_utilization": 98,
            "latency_ms": 450,
            "throughput_mbps": 1250,
            "queue_depth": 850,
            "error_rate": 8.5
        },
        "business_impact": {
            "revenue_loss_per_hour": 6800,
            "data_processing_delay_hours": 3,
            "analytics_backlog": True,
            "reporting_failure": True
        },
        "roi_data": {
            "hourly_revenue_loss": 6800,
            "manual_recovery_hours": 1.75,
            "enterprise_recovery_hours": 0.22,
            "engineers_required": 3,
            "engineer_hourly_rate": 150,
            "estimated_monthly_occurrences": 1.5,
            "enterprise_savings_percentage": 0.83
        }
    }
}