@@ -0,0 +1,301 @@ |
| 1 | +use serde::{Deserialize, Serialize}; |
| 2 | +use std::collections::HashMap; |
| 3 | +use std::time::{Duration, Instant, SystemTime, UNIX_EPOCH}; |
| 4 | +use tokio::sync::RwLock; |
| 5 | +use std::sync::Arc; |
| 6 | + |
| 7 | +#[derive(Debug, Clone, Serialize, Deserialize)] |
| 8 | +pub struct Metrics { |
| 9 | + pub commands: CommandMetrics, |
| 10 | + pub network: NetworkMetrics, |
| 11 | + pub storage: StorageMetrics, |
| 12 | + pub performance: PerformanceMetrics, |
| 13 | +} |
| 14 | + |
| 15 | +#[derive(Debug, Clone, Serialize, Deserialize)] |
| 16 | +pub struct CommandMetrics { |
| 17 | + pub total_commands: u64, |
| 18 | + pub commands_by_type: HashMap<String, u64>, |
| 19 | + pub command_durations: HashMap<String, Vec<f64>>, // in milliseconds |
| 20 | + pub failed_commands: u64, |
| 21 | + pub last_command_time: Option<u64>, // unix timestamp |
| 22 | +} |
| 23 | + |
| 24 | +#[derive(Debug, Clone, Serialize, Deserialize)] |
| 25 | +pub struct NetworkMetrics { |
| 26 | + pub total_requests: u64, |
| 27 | + pub failed_requests: u64, |
| 28 | + pub average_response_time: f64, // milliseconds |
| 29 | + pub timeouts: u64, |
| 30 | + pub bytes_uploaded: u64, |
| 31 | + pub bytes_downloaded: u64, |
| 32 | + pub peer_connections: u32, |
| 33 | +} |
| 34 | + |
| 35 | +#[derive(Debug, Clone, Serialize, Deserialize)] |
| 36 | +pub struct StorageMetrics { |
| 37 | + pub files_uploaded: u64, |
| 38 | + pub files_downloaded: u64, |
| 39 | + pub total_bytes_processed: u64, |
| 40 | + pub cache_hits: u64, |
| 41 | + pub cache_misses: u64, |
| 42 | + pub disk_operations: u64, |
| 43 | +} |
| 44 | + |
| 45 | +#[derive(Debug, Clone, Serialize, Deserialize)] |
| 46 | +pub struct PerformanceMetrics { |
| 47 | + pub memory_usage_mb: f64, |
| 48 | + pub cpu_usage_percent: f64, |
| 49 | + pub startup_time_ms: f64, |
| 50 | + pub gc_collections: u32, |
| 51 | + pub peak_memory_mb: f64, |
| 52 | +} |
| 53 | + |
| 54 | +#[derive(Debug)] |
| 55 | +pub struct MetricsCollector { |
| 56 | + metrics: Arc<RwLock<Metrics>>, |
| 57 | + start_time: Instant, |
| 58 | +} |
| 59 | + |
| 60 | +impl Default for Metrics { |
| 61 | + fn default() -> Self { |
| 62 | + Self { |
| 63 | + commands: CommandMetrics { |
| 64 | + total_commands: 0, |
| 65 | + commands_by_type: HashMap::new(), |
| 66 | + command_durations: HashMap::new(), |
| 67 | + failed_commands: 0, |
| 68 | + last_command_time: None, |
| 69 | + }, |
| 70 | + network: NetworkMetrics { |
| 71 | + total_requests: 0, |
| 72 | + failed_requests: 0, |
| 73 | + average_response_time: 0.0, |
| 74 | + timeouts: 0, |
| 75 | + bytes_uploaded: 0, |
| 76 | + bytes_downloaded: 0, |
| 77 | + peer_connections: 0, |
| 78 | + }, |
| 79 | + storage: StorageMetrics { |
| 80 | + files_uploaded: 0, |
| 81 | + files_downloaded: 0, |
| 82 | + total_bytes_processed: 0, |
| 83 | + cache_hits: 0, |
| 84 | + cache_misses: 0, |
| 85 | + disk_operations: 0, |
| 86 | + }, |
| 87 | + performance: PerformanceMetrics { |
| 88 | + memory_usage_mb: 0.0, |
| 89 | + cpu_usage_percent: 0.0, |
| 90 | + startup_time_ms: 0.0, |
| 91 | + gc_collections: 0, |
| 92 | + peak_memory_mb: 0.0, |
| 93 | + }, |
| 94 | + } |
| 95 | + } |
| 96 | +} |
| 97 | + |
| 98 | +impl MetricsCollector { |
| 99 | + pub fn new() -> Self { |
| 100 | + Self { |
| 101 | + metrics: Arc::new(RwLock::new(Metrics::default())), |
| 102 | + start_time: Instant::now(), |
| 103 | + } |
| 104 | + } |
| 105 | + |
| 106 | + pub async fn record_command_start(&self, command: &str) -> CommandTimer { |
| 107 | + let mut metrics = self.metrics.write().await; |
| 108 | + metrics.commands.total_commands += 1; |
| 109 | + *metrics.commands.commands_by_type.entry(command.to_string()).or_insert(0) += 1; |
| 110 | + metrics.commands.last_command_time = Some( |
| 111 | + SystemTime::now() |
| 112 | + .duration_since(UNIX_EPOCH) |
| 113 | + .unwrap_or_default() |
| 114 | + .as_secs() |
| 115 | + ); |
| 116 | + |
| 117 | + CommandTimer { |
| 118 | + command: command.to_string(), |
| 119 | + start_time: Instant::now(), |
| 120 | + collector: self.metrics.clone(), |
| 121 | + } |
| 122 | + } |
| 123 | + |
| 124 | + pub async fn record_network_request(&self, duration: Duration, success: bool, bytes_transferred: u64, upload: bool) { |
| 125 | + let mut metrics = self.metrics.write().await; |
| 126 | + metrics.network.total_requests += 1; |
| 127 | + |
| 128 | + if !success { |
| 129 | + metrics.network.failed_requests += 1; |
| 130 | + } |
| 131 | + |
| 132 | + // Update average response time (simple moving average) |
| 133 | + let duration_ms = duration.as_secs_f64() * 1000.0; |
| 134 | + let total_requests = metrics.network.total_requests as f64; |
| 135 | + metrics.network.average_response_time = |
| 136 | + (metrics.network.average_response_time * (total_requests - 1.0) + duration_ms) / total_requests; |
| 137 | + |
| 138 | + if upload { |
| 139 | + metrics.network.bytes_uploaded += bytes_transferred; |
| 140 | + } else { |
| 141 | + metrics.network.bytes_downloaded += bytes_transferred; |
| 142 | + } |
| 143 | + } |
| 144 | + |
| 145 | + pub async fn record_timeout(&self) { |
| 146 | + let mut metrics = self.metrics.write().await; |
| 147 | + metrics.network.timeouts += 1; |
| 148 | + metrics.network.failed_requests += 1; |
| 149 | + } |
| 150 | + |
| 151 | + pub async fn record_file_upload(&self, file_size: u64) { |
| 152 | + let mut metrics = self.metrics.write().await; |
| 153 | + metrics.storage.files_uploaded += 1; |
| 154 | + metrics.storage.total_bytes_processed += file_size; |
| 155 | + metrics.storage.disk_operations += 1; |
| 156 | + } |
| 157 | + |
| 158 | + pub async fn record_file_download(&self, file_size: u64) { |
| 159 | + let mut metrics = self.metrics.write().await; |
| 160 | + metrics.storage.files_downloaded += 1; |
| 161 | + metrics.storage.total_bytes_processed += file_size; |
| 162 | + metrics.storage.disk_operations += 1; |
| 163 | + } |
| 164 | + |
| 165 | + pub async fn record_cache_hit(&self) { |
| 166 | + let mut metrics = self.metrics.write().await; |
| 167 | + metrics.storage.cache_hits += 1; |
| 168 | + } |
| 169 | + |
| 170 | + pub async fn record_cache_miss(&self) { |
| 171 | + let mut metrics = self.metrics.write().await; |
| 172 | + metrics.storage.cache_misses += 1; |
| 173 | + } |
| 174 | + |
| 175 | + pub async fn update_peer_count(&self, peer_count: u32) { |
| 176 | + let mut metrics = self.metrics.write().await; |
| 177 | + metrics.network.peer_connections = peer_count; |
| 178 | + } |
| 179 | + |
| 180 | + pub async fn record_memory_usage(&self, memory_mb: f64) { |
| 181 | + let mut metrics = self.metrics.write().await; |
| 182 | + metrics.performance.memory_usage_mb = memory_mb; |
| 183 | + if memory_mb > metrics.performance.peak_memory_mb { |
| 184 | + metrics.performance.peak_memory_mb = memory_mb; |
| 185 | + } |
| 186 | + } |
| 187 | + |
| 188 | + pub async fn record_cpu_usage(&self, cpu_percent: f64) { |
| 189 | + let mut metrics = self.metrics.write().await; |
| 190 | + metrics.performance.cpu_usage_percent = cpu_percent; |
| 191 | + } |
| 192 | + |
| 193 | + pub async fn get_metrics(&self) -> Metrics { |
| 194 | + let metrics = self.metrics.read().await; |
| 195 | + let mut result = metrics.clone(); |
| 196 | + result.performance.startup_time_ms = self.start_time.elapsed().as_secs_f64() * 1000.0; |
| 197 | + result |
| 198 | + } |
| 199 | + |
| 200 | + pub async fn get_summary(&self) -> MetricsSummary { |
| 201 | + let metrics = self.get_metrics().await; |
| 202 | + |
| 203 | + MetricsSummary { |
| 204 | + uptime_seconds: self.start_time.elapsed().as_secs(), |
| 205 | + total_commands: metrics.commands.total_commands, |
| 206 | + success_rate: if metrics.commands.total_commands > 0 { |
| 207 | + ((metrics.commands.total_commands - metrics.commands.failed_commands) as f64 |
| 208 | + / metrics.commands.total_commands as f64) * 100.0 |
| 209 | + } else { 0.0 }, |
| 210 | + network_success_rate: if metrics.network.total_requests > 0 { |
| 211 | + ((metrics.network.total_requests - metrics.network.failed_requests) as f64 |
| 212 | + / metrics.network.total_requests as f64) * 100.0 |
| 213 | + } else { 0.0 }, |
| 214 | + average_response_time_ms: metrics.network.average_response_time, |
| 215 | + total_bytes_transferred: metrics.network.bytes_uploaded + metrics.network.bytes_downloaded, |
| 216 | + files_processed: metrics.storage.files_uploaded + metrics.storage.files_downloaded, |
| 217 | + cache_hit_rate: if metrics.storage.cache_hits + metrics.storage.cache_misses > 0 { |
| 218 | + (metrics.storage.cache_hits as f64 / |
| 219 | + (metrics.storage.cache_hits + metrics.storage.cache_misses) as f64) * 100.0 |
| 220 | + } else { 0.0 }, |
| 221 | + memory_usage_mb: metrics.performance.memory_usage_mb, |
| 222 | + peak_memory_mb: metrics.performance.peak_memory_mb, |
| 223 | + } |
| 224 | + } |
| 225 | + |
| 226 | + pub async fn export_metrics(&self) -> Result<String, serde_json::Error> { |
| 227 | + let metrics = self.get_metrics().await; |
| 228 | + serde_json::to_string_pretty(&metrics) |
| 229 | + } |
| 230 | + |
| 231 | + pub async fn reset_metrics(&self) { |
| 232 | + let mut metrics = self.metrics.write().await; |
| 233 | + *metrics = Metrics::default(); |
| 234 | + } |
| 235 | +} |
| 236 | + |
| 237 | +#[derive(Debug)] |
| 238 | +pub struct CommandTimer { |
| 239 | + command: String, |
| 240 | + start_time: Instant, |
| 241 | + collector: Arc<RwLock<Metrics>>, |
| 242 | +} |
| 243 | + |
| 244 | +impl Drop for CommandTimer { |
| 245 | + fn drop(&mut self) { |
| 246 | + let duration = self.start_time.elapsed(); |
| 247 | + let duration_ms = duration.as_secs_f64() * 1000.0; |
| 248 | + |
| 249 | + // Use a blocking task to update metrics since we're in Drop |
| 250 | + tokio::task::block_in_place(|| { |
| 251 | + tokio::runtime::Handle::current().block_on(async { |
| 252 | + let mut metrics = self.collector.write().await; |
| 253 | + metrics.commands.command_durations |
| 254 | + .entry(self.command.clone()) |
| 255 | + .or_insert_with(Vec::new) |
| 256 | + .push(duration_ms); |
| 257 | + }); |
| 258 | + }); |
| 259 | + } |
| 260 | +} |
| 261 | + |
| 262 | +#[derive(Debug, Serialize, Deserialize)] |
| 263 | +pub struct MetricsSummary { |
| 264 | + pub uptime_seconds: u64, |
| 265 | + pub total_commands: u64, |
| 266 | + pub success_rate: f64, |
| 267 | + pub network_success_rate: f64, |
| 268 | + pub average_response_time_ms: f64, |
| 269 | + pub total_bytes_transferred: u64, |
| 270 | + pub files_processed: u64, |
| 271 | + pub cache_hit_rate: f64, |
| 272 | + pub memory_usage_mb: f64, |
| 273 | + pub peak_memory_mb: f64, |
| 274 | +} |
| 275 | + |
| 276 | +// Global metrics collector instance |
| 277 | +lazy_static::lazy_static! { |
| 278 | + pub static ref METRICS: MetricsCollector = MetricsCollector::new(); |
| 279 | +} |
| 280 | + |
| 281 | +// Convenience macros for metrics collection |
| 282 | +#[macro_export] |
| 283 | +macro_rules! time_command { |
| 284 | + ($command:expr) => { |
| 285 | + crate::metrics::METRICS.record_command_start($command).await |
| 286 | + }; |
| 287 | +} |
| 288 | + |
| 289 | +#[macro_export] |
| 290 | +macro_rules! record_network_success { |
| 291 | + ($duration:expr, $bytes:expr, $upload:expr) => { |
| 292 | + crate::metrics::METRICS.record_network_request($duration, true, $bytes, $upload).await |
| 293 | + }; |
| 294 | +} |
| 295 | + |
| 296 | +#[macro_export] |
| 297 | +macro_rules! record_network_failure { |
| 298 | + ($duration:expr, $bytes:expr, $upload:expr) => { |
| 299 | + crate::metrics::METRICS.record_network_request($duration, false, $bytes, $upload).await |
| 300 | + }; |
| 301 | +} |