foxhunt/monitoring/metrics.rs

use prometheus::{
    Counter, Histogram, Gauge, IntCounter, IntGauge, 
    register_counter, register_histogram, register_gauge, 
    register_int_counter, register_int_gauge,
    Opts, HistogramOpts, Registry, Encoder, TextEncoder
};
use std::collections::HashMap;
use std::sync::Arc;
use lazy_static::lazy_static;
use tracing::{error, info, warn};

/// Core HFT trading metrics for Foxhunt system
/// Optimized for high-frequency data collection with minimal latency impact

lazy_static! {
    // Order processing metrics
    static ref ORDER_COUNTER: Counter = register_counter!(
        "foxhunt_orders_total",
        "Total orders processed by the trading system"
    ).expect("Failed to register orders counter");
    
    static ref ORDER_FILL_COUNTER: Counter = register_counter!(
        "foxhunt_order_fills_total", 
        "Total order fills executed"
    ).expect("Failed to register order fills counter");
    
    static ref ORDER_REJECTION_COUNTER: Counter = register_counter!(
        "foxhunt_order_rejections_total",
        "Total order rejections"
    ).expect("Failed to register order rejections counter");
    
    // Latency metrics - critical for HFT performance
    static ref LATENCY_HISTOGRAM: Histogram = register_histogram!(
        HistogramOpts::new(
            "foxhunt_latency_microseconds",
            "Latency distribution in microseconds"
        ).buckets(vec![1.0, 5.0, 10.0, 25.0, 50.0, 100.0, 250.0, 500.0, 1000.0])
    ).expect("Failed to register latency histogram");
    
    static ref ORDER_PROCESSING_LATENCY: Histogram = register_histogram!(
        HistogramOpts::new(
            "foxhunt_order_processing_latency_microseconds",
            "Order processing latency from receipt to exchange submission"
        ).buckets(vec![1.0, 5.0, 10.0, 25.0, 50.0, 100.0, 250.0, 500.0])
    ).expect("Failed to register order processing latency histogram");
    
    static ref MARKET_DATA_LATENCY: Histogram = register_histogram!(
        HistogramOpts::new(
            "foxhunt_market_data_latency_microseconds", 
            "Market data processing latency"
        ).buckets(vec![0.1, 0.5, 1.0, 5.0, 10.0, 25.0, 50.0, 100.0])
    ).expect("Failed to register market data latency histogram");
    
    // Risk metrics
    static ref RISK_BREACH_COUNTER: Counter = register_counter!(
        "foxhunt_risk_breaches_total",
        "Total risk limit breaches"
    ).expect("Failed to register risk breaches counter");
    
    static ref POSITION_VALUE_GAUGE: Gauge = register_gauge!(
        "foxhunt_position_value_usd",
        "Current total position value in USD"
    ).expect("Failed to register position value gauge");
    
    static ref VAR_GAUGE: Gauge = register_gauge!(
        "foxhunt_var_usd",
        "Current Value at Risk in USD"
    ).expect("Failed to register VaR gauge");
    
    // Trading engine metrics
    static ref ACTIVE_ORDERS_GAUGE: IntGauge = register_int_gauge!(
        "foxhunt_active_orders",
        "Number of currently active orders"
    ).expect("Failed to register active orders gauge");
    
    static ref TRADING_SESSIONS_GAUGE: IntGauge = register_int_gauge!(
        "foxhunt_trading_sessions_active",
        "Number of active trading sessions"
    ).expect("Failed to register trading sessions gauge");
    
    // Market data metrics
    static ref MARKET_DATA_MESSAGES_COUNTER: Counter = register_counter!(
        "foxhunt_market_data_messages_total",
        "Total market data messages received"
    ).expect("Failed to register market data messages counter");
    
    static ref MARKET_DATA_DROPS_COUNTER: Counter = register_counter!(
        "foxhunt_market_data_drops_total",
        "Total market data messages dropped"
    ).expect("Failed to register market data drops counter");
    
    // AI/ML metrics
    static ref ML_PREDICTIONS_COUNTER: Counter = register_counter!(
        "foxhunt_ml_predictions_total",
        "Total ML model predictions generated"
    ).expect("Failed to register ML predictions counter");
    
    static ref ML_MODEL_ACCURACY_GAUGE: Gauge = register_gauge!(
        "foxhunt_ml_model_accuracy",
        "Current ML model accuracy percentage"
    ).expect("Failed to register ML model accuracy gauge");
    
    // System health metrics
    static ref CPU_USAGE_GAUGE: Gauge = register_gauge!(
        "foxhunt_cpu_usage_percent",
        "Current CPU usage percentage"
    ).expect("Failed to register CPU usage gauge");
    
    static ref MEMORY_USAGE_GAUGE: Gauge = register_gauge!(
        "foxhunt_memory_usage_bytes",
        "Current memory usage in bytes"
    ).expect("Failed to register memory usage gauge");
    
    // Broker connectivity metrics
    static ref BROKER_CONNECTIONS_GAUGE: IntGauge = register_int_gauge!(
        "foxhunt_broker_connections",
        "Number of active broker connections"
    ).expect("Failed to register broker connections gauge");
    
    static ref BROKER_DISCONNECTS_COUNTER: Counter = register_counter!(
        "foxhunt_broker_disconnects_total",
        "Total broker disconnection events"
    ).expect("Failed to register broker disconnects counter");
    
    // Performance metrics
    static ref THROUGHPUT_GAUGE: Gauge = register_gauge!(
        "foxhunt_throughput_ops_per_second",
        "Current system throughput in operations per second"
    ).expect("Failed to register throughput gauge");
}

/// Metrics collector for the Foxhunt HFT system
#[derive(Debug)]
pub struct FoxhuntMetrics {
    registry: Arc<Registry>,
    custom_counters: HashMap<String, Counter>,
    custom_histograms: HashMap<String, Histogram>,
    custom_gauges: HashMap<String, Gauge>,
}

impl Default for FoxhuntMetrics {
    fn default() -> Self {
        Self::new()
    }
}

impl FoxhuntMetrics {
    /// Create a new metrics collector instance
    pub fn new() -> Self {
        Self {
            registry: Arc::new(Registry::new()),
            custom_counters: HashMap::new(),
            custom_histograms: HashMap::new(),
            custom_gauges: HashMap::new(),
        }
    }

    /// Record an order being processed
    #[inline(always)]
    pub fn record_order() {
        ORDER_COUNTER.inc();
    }

    /// Record an order fill
    #[inline(always)]
    pub fn record_order_fill() {
        ORDER_FILL_COUNTER.inc();
    }

    /// Record an order rejection
    #[inline(always)]
    pub fn record_order_rejection() {
        ORDER_REJECTION_COUNTER.inc();
    }

    /// Record latency measurement in microseconds
    #[inline(always)]
    pub fn record_latency(latency_us: f64) {
        LATENCY_HISTOGRAM.observe(latency_us);
    }

    /// Record order processing latency in microseconds
    #[inline(always)]
    pub fn record_order_processing_latency(latency_us: f64) {
        ORDER_PROCESSING_LATENCY.observe(latency_us);
    }

    /// Record market data latency in microseconds
    #[inline(always)]
    pub fn record_market_data_latency(latency_us: f64) {
        MARKET_DATA_LATENCY.observe(latency_us);
    }

    /// Record a risk breach event
    #[inline(always)]
    pub fn record_risk_breach() {
        RISK_BREACH_COUNTER.inc();
        warn!("Risk breach recorded in metrics");
    }

    /// Update current position value
    #[inline(always)]
    pub fn update_position_value(value_usd: f64) {
        POSITION_VALUE_GAUGE.set(value_usd);
    }

    /// Update Value at Risk
    #[inline(always)]
    pub fn update_var(var_usd: f64) {
        VAR_GAUGE.set(var_usd);
    }

    /// Update active orders count
    #[inline(always)]
    pub fn update_active_orders(count: i64) {
        ACTIVE_ORDERS_GAUGE.set(count);
    }

    /// Update trading sessions count
    #[inline(always)]
    pub fn update_trading_sessions(count: i64) {
        TRADING_SESSIONS_GAUGE.set(count);
    }

    /// Record market data message received
    #[inline(always)]
    pub fn record_market_data_message() {
        MARKET_DATA_MESSAGES_COUNTER.inc();
    }

    /// Record market data message dropped
    #[inline(always)]
    pub fn record_market_data_drop() {
        MARKET_DATA_DROPS_COUNTER.inc();
    }

    /// Record ML prediction generated
    #[inline(always)]
    pub fn record_ml_prediction() {
        ML_PREDICTIONS_COUNTER.inc();
    }

    /// Update ML model accuracy
    #[inline(always)]
    pub fn update_ml_accuracy(accuracy: f64) {
        ML_MODEL_ACCURACY_GAUGE.set(accuracy);
    }

    /// Update CPU usage percentage
    #[inline(always)]
    pub fn update_cpu_usage(percentage: f64) {
        CPU_USAGE_GAUGE.set(percentage);
    }

    /// Update memory usage in bytes
    #[inline(always)]
    pub fn update_memory_usage(bytes: f64) {
        MEMORY_USAGE_GAUGE.set(bytes);
    }

    /// Update broker connections count
    #[inline(always)]
    pub fn update_broker_connections(count: i64) {
        BROKER_CONNECTIONS_GAUGE.set(count);
    }

    /// Record broker disconnection
    #[inline(always)]
    pub fn record_broker_disconnect() {
        BROKER_DISCONNECTS_COUNTER.inc();
    }

    /// Update system throughput
    #[inline(always)]
    pub fn update_throughput(ops_per_second: f64) {
        THROUGHPUT_GAUGE.set(ops_per_second);
    }

    /// Get metrics in Prometheus text format
    pub fn export_metrics() -> Result<String, Box<dyn std::error::Error + Send + Sync>> {
        let encoder = TextEncoder::new();
        let metric_families = prometheus::gather();
        let mut buffer = Vec::new();
        encoder.encode(&metric_families, &mut buffer)?;
        Ok(String::from_utf8(buffer)?)
    }

    /// Create a custom counter metric
    pub fn create_custom_counter(&mut self, name: &str, help: &str) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
        let counter = Counter::new(name, help)?;
        self.registry.register(Box::new(counter.clone()))?;
        self.custom_counters.insert(name.to_string(), counter);
        Ok(())
    }

    /// Increment a custom counter
    pub fn increment_custom_counter(&self, name: &str) {
        if let Some(counter) = self.custom_counters.get(name) {
            counter.inc();
        } else {
            error!("Custom counter '{}' not found", name);
        }
    }

    /// Reset all metrics (for testing purposes)
    #[cfg(test)]
    pub fn reset_all() {
        // Reset all static metrics to zero
        ORDER_COUNTER.reset();
        ORDER_FILL_COUNTER.reset();
        ORDER_REJECTION_COUNTER.reset();
        RISK_BREACH_COUNTER.reset();
        MARKET_DATA_MESSAGES_COUNTER.reset();
        MARKET_DATA_DROPS_COUNTER.reset();
        ML_PREDICTIONS_COUNTER.reset();
        BROKER_DISCONNECTS_COUNTER.reset();
        
        // Reset gauges to zero
        POSITION_VALUE_GAUGE.set(0.0);
        VAR_GAUGE.set(0.0);
        ACTIVE_ORDERS_GAUGE.set(0);
        TRADING_SESSIONS_GAUGE.set(0);
        ML_MODEL_ACCURACY_GAUGE.set(0.0);
        CPU_USAGE_GAUGE.set(0.0);
        MEMORY_USAGE_GAUGE.set(0.0);
        BROKER_CONNECTIONS_GAUGE.set(0);
        THROUGHPUT_GAUGE.set(0.0);
    }

    /// Log current metrics summary
    pub fn log_metrics_summary() {
        info!(
            "Metrics Summary - Orders: {}, Fills: {}, Rejections: {}, Active Orders: {}, Risk Breaches: {}",
            ORDER_COUNTER.get(),
            ORDER_FILL_COUNTER.get(),
            ORDER_REJECTION_COUNTER.get(),
            ACTIVE_ORDERS_GAUGE.get(),
            RISK_BREACH_COUNTER.get()
        );
    }
}

/// Convenience functions for direct metric recording
/// These are optimized for hot path usage with minimal overhead

/// Record order with timing
#[inline(always)]
pub fn record_order() {
    FoxhuntMetrics::record_order();
}

/// Record order fill
#[inline(always)]
pub fn record_order_fill() {
    FoxhuntMetrics::record_order_fill();
}

/// Record order rejection
#[inline(always)]
pub fn record_order_rejection() {
    FoxhuntMetrics::record_order_rejection();
}

/// Record latency measurement
#[inline(always)]
pub fn record_latency(latency_us: f64) {
    FoxhuntMetrics::record_latency(latency_us);
}

/// Record order processing latency
#[inline(always)]
pub fn record_order_processing_latency(latency_us: f64) {
    FoxhuntMetrics::record_order_processing_latency(latency_us);
}

/// Record market data latency
#[inline(always)]
pub fn record_market_data_latency(latency_us: f64) {
    FoxhuntMetrics::record_market_data_latency(latency_us);
}

/// Record risk breach
#[inline(always)]
pub fn record_risk_breach() {
    FoxhuntMetrics::record_risk_breach();
}

/// Update position value
#[inline(always)]
pub fn update_position_value(value_usd: f64) {
    FoxhuntMetrics::update_position_value(value_usd);
}

/// Update VaR
#[inline(always)]
pub fn update_var(var_usd: f64) {
    FoxhuntMetrics::update_var(var_usd);
}

/// Update active orders count
#[inline(always)]
pub fn update_active_orders(count: i64) {
    FoxhuntMetrics::update_active_orders(count);
}

/// Record market data message
#[inline(always)]
pub fn record_market_data_message() {
    FoxhuntMetrics::record_market_data_message();
}

/// Record ML prediction
#[inline(always)]
pub fn record_ml_prediction() {
    FoxhuntMetrics::record_ml_prediction();
}

/// Update system throughput
#[inline(always)]
pub fn update_throughput(ops_per_second: f64) {
    FoxhuntMetrics::update_throughput(ops_per_second);
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn test_order_metrics() {
        FoxhuntMetrics::reset_all();
        
        record_order();
        record_order();
        record_order_fill();
        
        assert_eq!(ORDER_COUNTER.get(), 2.0);
        assert_eq!(ORDER_FILL_COUNTER.get(), 1.0);
    }

    #[test]
    fn test_latency_metrics() {
        record_latency(25.5);
        record_order_processing_latency(15.2);
        record_market_data_latency(2.1);
        
        // Histograms don't have direct getters, but we can verify they accept values
        // In a real environment, these would be scraped by Prometheus
    }

    #[test]
    fn test_risk_metrics() {
        FoxhuntMetrics::reset_all();
        
        record_risk_breach();
        update_position_value(150000.0);
        update_var(5000.0);
        
        assert_eq!(RISK_BREACH_COUNTER.get(), 1.0);
        assert_eq!(POSITION_VALUE_GAUGE.get(), 150000.0);
        assert_eq!(VAR_GAUGE.get(), 5000.0);
    }

    #[test]
    fn test_custom_metrics() {
        let mut metrics = FoxhuntMetrics::new();
        
        metrics.create_custom_counter("test_counter", "Test counter").unwrap();
        metrics.increment_custom_counter("test_counter");
        
        // Custom counter incremented successfully
    }

    #[test]
    fn test_metrics_export() {
        record_order();
        
        let exported = FoxhuntMetrics::export_metrics().unwrap();
        assert!(exported.contains("foxhunt_orders_total"));
    }
}