use async_trait::async_trait; use cm_dashboard_shared::{Metric, MetricValue, Status, registry}; use std::time::Duration; use tracing::debug; use super::{Collector, CollectorError, utils}; use crate::config::CpuConfig; /// Extremely efficient CPU metrics collector /// /// EFFICIENCY OPTIMIZATIONS: /// - Single /proc/loadavg read for all load metrics /// - Single /proc/stat read for CPU usage /// - Minimal string allocations /// - No process spawning /// - <0.1ms collection time target pub struct CpuCollector { config: CpuConfig, name: String, } impl CpuCollector { pub fn new(config: CpuConfig) -> Self { Self { config, name: "cpu".to_string(), } } /// Calculate CPU load status using configured thresholds fn calculate_load_status(&self, load: f32) -> Status { if load >= self.config.load_critical_threshold { Status::Critical } else if load >= self.config.load_warning_threshold { Status::Warning } else { Status::Ok } } /// Calculate CPU temperature status using configured thresholds fn calculate_temperature_status(&self, temp: f32) -> Status { if temp >= self.config.temperature_critical_threshold { Status::Critical } else if temp >= self.config.temperature_warning_threshold { Status::Warning } else { Status::Ok } } /// Collect CPU load averages from /proc/loadavg /// Format: "0.52 0.58 0.59 1/257 12345" async fn collect_load_averages(&self) -> Result, CollectorError> { let content = utils::read_proc_file("/proc/loadavg")?; let parts: Vec<&str> = content.trim().split_whitespace().collect(); if parts.len() < 3 { return Err(CollectorError::Parse { value: content, error: "Expected at least 3 values in /proc/loadavg".to_string(), }); } let load_1min = utils::parse_f32(parts[0])?; let load_5min = utils::parse_f32(parts[1])?; let load_15min = utils::parse_f32(parts[2])?; // Calculate status for each load average (use 1min for primary status) let load_1min_status = self.calculate_load_status(load_1min); let load_5min_status = self.calculate_load_status(load_5min); let load_15min_status = self.calculate_load_status(load_15min); Ok(vec![ Metric::new( registry::CPU_LOAD_1MIN.to_string(), MetricValue::Float(load_1min), load_1min_status, ).with_description("CPU load average over 1 minute".to_string()), Metric::new( registry::CPU_LOAD_5MIN.to_string(), MetricValue::Float(load_5min), load_5min_status, ).with_description("CPU load average over 5 minutes".to_string()), Metric::new( registry::CPU_LOAD_15MIN.to_string(), MetricValue::Float(load_15min), load_15min_status, ).with_description("CPU load average over 15 minutes".to_string()), ]) } /// Collect CPU temperature from thermal zones /// Prioritizes x86_pkg_temp over generic thermal zones (legacy behavior) async fn collect_temperature(&self) -> Result, CollectorError> { // Try x86_pkg_temp first (Intel CPU package temperature) if let Ok(temp) = self.read_thermal_zone("/sys/class/thermal/thermal_zone0/temp").await { let temp_celsius = temp as f32 / 1000.0; let status = self.calculate_temperature_status(temp_celsius); return Ok(Some(Metric::new( registry::CPU_TEMPERATURE_CELSIUS.to_string(), MetricValue::Float(temp_celsius), status, ).with_description("CPU package temperature".to_string()) .with_unit("°C".to_string()))); } // Fallback: try other thermal zones for zone_id in 0..10 { let path = format!("/sys/class/thermal/thermal_zone{}/temp", zone_id); if let Ok(temp) = self.read_thermal_zone(&path).await { let temp_celsius = temp as f32 / 1000.0; let status = self.calculate_temperature_status(temp_celsius); return Ok(Some(Metric::new( registry::CPU_TEMPERATURE_CELSIUS.to_string(), MetricValue::Float(temp_celsius), status, ).with_description(format!("CPU temperature from thermal_zone{}", zone_id)) .with_unit("°C".to_string()))); } } debug!("No CPU temperature sensors found"); Ok(None) } /// Read temperature from thermal zone efficiently async fn read_thermal_zone(&self, path: &str) -> Result { let content = utils::read_proc_file(path)?; utils::parse_u64(content.trim()) } /// Collect CPU frequency from /proc/cpuinfo or scaling governor async fn collect_frequency(&self) -> Result, CollectorError> { // Try scaling frequency first (more accurate for current frequency) if let Ok(freq) = utils::read_proc_file("/sys/devices/system/cpu/cpu0/cpufreq/scaling_cur_freq") { if let Ok(freq_khz) = utils::parse_u64(freq.trim()) { let freq_mhz = freq_khz as f32 / 1000.0; return Ok(Some(Metric::new( registry::CPU_FREQUENCY_MHZ.to_string(), MetricValue::Float(freq_mhz), Status::Ok, // Frequency doesn't have status thresholds ).with_description("Current CPU frequency".to_string()) .with_unit("MHz".to_string()))); } } // Fallback: parse /proc/cpuinfo for base frequency if let Ok(content) = utils::read_proc_file("/proc/cpuinfo") { for line in content.lines() { if line.starts_with("cpu MHz") { if let Some(freq_str) = line.split(':').nth(1) { if let Ok(freq_mhz) = utils::parse_f32(freq_str) { return Ok(Some(Metric::new( registry::CPU_FREQUENCY_MHZ.to_string(), MetricValue::Float(freq_mhz), Status::Ok, ).with_description("CPU base frequency from /proc/cpuinfo".to_string()) .with_unit("MHz".to_string()))); } } break; // Only need first CPU entry } } } debug!("CPU frequency not available"); Ok(None) } } #[async_trait] impl Collector for CpuCollector { fn name(&self) -> &str { &self.name } async fn collect(&self) -> Result, CollectorError> { debug!("Collecting CPU metrics"); let start = std::time::Instant::now(); let mut metrics = Vec::with_capacity(5); // Pre-allocate for efficiency // Collect load averages (always available) metrics.extend(self.collect_load_averages().await?); // Collect temperature (optional) if let Some(temp_metric) = self.collect_temperature().await? { metrics.push(temp_metric); } // Collect frequency (optional) if let Some(freq_metric) = self.collect_frequency().await? { metrics.push(freq_metric); } let duration = start.elapsed(); debug!("CPU collection completed in {:?} with {} metrics", duration, metrics.len()); // Efficiency check: warn if collection takes too long if duration.as_millis() > 1 { debug!("CPU collection took {}ms - consider optimization", duration.as_millis()); } // Store performance metrics // Performance tracking handled by cache system Ok(metrics) } fn get_performance_metrics(&self) -> Option { None // Performance tracking handled by cache system } }