deepwiki-rs 1.1.5

use serde::{Deserialize, Serialize};
use std::collections::HashMap;
use std::sync::Arc;
use std::sync::atomic::{AtomicU64, AtomicUsize, Ordering};
use std::time::Duration;

use crate::llm::client::types::TokenUsage;

/// 缓存性能监控器
#[derive(Clone)]
pub struct CachePerformanceMonitor {
    metrics: Arc<CacheMetrics>,
}

/// 缓存指标
#[derive(Default)]
pub struct CacheMetrics {
    /// 缓存命中次数
    pub cache_hits: AtomicUsize,
    /// 缓存未命中次数
    pub cache_misses: AtomicUsize,
    /// 缓存写入次数
    pub cache_writes: AtomicUsize,
    /// 缓存错误次数
    pub cache_errors: AtomicUsize,
    /// 总节省的推理时间（秒）
    pub total_inference_time_saved: AtomicU64,
    /// 总节省的推理成本（估算）
    pub total_cost_saved: AtomicUsize,
    /// 总节省的输入token数量
    pub total_input_tokens_saved: AtomicUsize,
    /// 总节省的输出token数量
    pub total_output_tokens_saved: AtomicUsize,
}

/// 缓存性能报告
#[derive(Debug, Serialize, Deserialize)]
pub struct CachePerformanceReport {
    /// 缓存命中率
    pub hit_rate: f64,
    /// 总缓存操作次数
    pub total_operations: usize,
    /// 缓存命中次数
    pub cache_hits: usize,
    /// 缓存未命中次数
    pub cache_misses: usize,
    /// 缓存写入次数
    pub cache_writes: usize,
    /// 缓存错误次数
    pub cache_errors: usize,
    /// 节省的推理时间（秒）
    pub inference_time_saved: f64,
    /// 节省的推理成本（美元，估算）
    pub cost_saved: f64,
    /// 性能提升百分比
    pub performance_improvement: f64,
    /// 节省的输入token数量
    pub input_tokens_saved: usize,
    /// 节省的输出token数量
    pub output_tokens_saved: usize,
    /// 分类统计
    pub category_stats: HashMap<String, CategoryPerformanceStats>,
}

/// 分类性能统计
#[derive(Debug, Serialize, Deserialize)]
pub struct CategoryPerformanceStats {
    pub hits: u64,
    pub misses: u64,
    pub hit_rate: f64,
    pub time_saved: f64,
    pub cost_saved: f64,
}

impl CachePerformanceMonitor {
    pub fn new() -> Self {
        Self {
            metrics: Arc::new(CacheMetrics::default()),
        }
    }

    /// 记录缓存命中
    pub fn record_cache_hit(
        &self,
        category: &str,
        inference_time_saved: Duration,
        token_usage: TokenUsage,
        model_name: &str,
    ) {
        self.metrics.cache_hits.fetch_add(1, Ordering::Relaxed);
        self.metrics
            .total_inference_time_saved
            .fetch_add(inference_time_saved.as_millis() as u64, Ordering::Relaxed);

        // 记录节省的token数量
        self.metrics
            .total_input_tokens_saved
            .fetch_add(token_usage.input_tokens, Ordering::Relaxed);
        self.metrics
            .total_output_tokens_saved
            .fetch_add(token_usage.output_tokens, Ordering::Relaxed);

        // 基于实际token使用情况计算节省的成本
        let estimated_cost_saved = token_usage.estimate_cost(model_name);
        self.metrics.total_cost_saved.fetch_add(
            (estimated_cost_saved * 1000.0) as usize, // 存储为毫美元
            Ordering::Relaxed,
        );

        println!(
            "   💰 缓存命中 [{}] - 节省推理时间: {:.2}秒, 节省tokens: {}输入+{}输出, 估算节省成本: ${:.4}",
            category,
            inference_time_saved.as_secs_f64(),
            token_usage.input_tokens,
            token_usage.output_tokens,
            estimated_cost_saved
        );
    }

    /// 记录缓存未命中
    pub fn record_cache_miss(&self, category: &str) {
        self.metrics.cache_misses.fetch_add(1, Ordering::Relaxed);
        println!("   ⌛ 缓存未命中 [{}] - 需要进行AI推理", category);
    }

    /// 记录缓存写入
    pub fn record_cache_write(&self, category: &str) {
        self.metrics.cache_writes.fetch_add(1, Ordering::Relaxed);
        println!("   💾 缓存写入 [{}] - 结果已缓存", category);
    }

    /// 记录缓存错误
    pub fn record_cache_error(&self, category: &str, error: &str) {
        self.metrics.cache_errors.fetch_add(1, Ordering::Relaxed);
        eprintln!("   ❌ 缓存错误 [{}]: {}", category, error);
    }

    /// 生成性能报告
    pub fn generate_report(&self) -> CachePerformanceReport {
        let hits = self.metrics.cache_hits.load(Ordering::Relaxed);
        let misses = self.metrics.cache_misses.load(Ordering::Relaxed);
        let writes = self.metrics.cache_writes.load(Ordering::Relaxed);
        let errors = self.metrics.cache_errors.load(Ordering::Relaxed);
        let total_operations = hits + misses;

        let hit_rate = if total_operations > 0 {
            hits as f64 / total_operations as f64
        } else {
            0.0
        };

        let inference_time_saved = self
            .metrics
            .total_inference_time_saved
            .load(Ordering::Relaxed) as f64
            / 1000.0; // 转换为秒
        let cost_saved = self.metrics.total_cost_saved.load(Ordering::Relaxed) as f64 / 1000.0; // 转换为美元

        let input_tokens_saved = self
            .metrics
            .total_input_tokens_saved
            .load(Ordering::Relaxed);
        let output_tokens_saved = self
            .metrics
            .total_output_tokens_saved
            .load(Ordering::Relaxed);

        let performance_improvement = if misses > 0 {
            (hits as f64 / (hits + misses) as f64) * 100.0
        } else {
            0.0
        };

        CachePerformanceReport {
            hit_rate,
            total_operations,
            cache_hits: hits,
            cache_misses: misses,
            cache_writes: writes,
            cache_errors: errors,
            inference_time_saved,
            cost_saved,
            performance_improvement,
            input_tokens_saved,
            output_tokens_saved,
            category_stats: HashMap::new(), // TODO: 实现分类统计
        }
    }
}

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