lesavka/common/src/process_metrics.rs

#![forbid(unsafe_code)]

use std::time::Instant;
use std::{process::Command, sync::OnceLock};

/// Sample per-process CPU pressure from Linux procfs.
///
/// Inputs: none when constructed; call `sample_percent()` over time.
/// Outputs: percentage of one CPU core used by the current process over the
/// elapsed sampling window.
/// Why: this gives the launcher and server lightweight pressure telemetry
/// without adding a heavyweight system-information dependency.
#[derive(Debug, Clone, Default)]
pub struct ProcessCpuSampler {
    last: Option<(Instant, u64)>,
}

impl ProcessCpuSampler {
    #[must_use]
    pub fn new() -> Self {
        Self::default()
    }

    pub fn sample_percent(&mut self) -> Option<f32> {
        let now = Instant::now();
        let runtime_ns = read_process_runtime_ns()?;
        self.sample_percent_at(now, runtime_ns)
    }

    pub fn sample_tenths_percent(&mut self) -> Option<u32> {
        self.sample_percent()
            .map(|pct| (pct * 10.0).clamp(0.0, u32::MAX as f32) as u32)
    }

    fn sample_percent_at(&mut self, now: Instant, runtime_ns: u64) -> Option<f32> {
        let previous = self.last.replace((now, runtime_ns))?;
        let elapsed_ns = now
            .saturating_duration_since(previous.0)
            .as_nanos()
            .min(u128::from(u64::MAX)) as u64;
        if elapsed_ns == 0 || runtime_ns < previous.1 {
            return None;
        }
        Some(runtime_ns.saturating_sub(previous.1) as f32 * 100.0 / elapsed_ns as f32)
    }
}

fn read_process_runtime_ns() -> Option<u64> {
    read_process_runtime_from_stat().or_else(read_process_runtime_from_schedstat)
}

fn read_process_runtime_from_stat() -> Option<u64> {
    let text = std::fs::read_to_string("/proc/self/stat").ok()?;
    let ticks_per_second = *clock_ticks_per_second()?;
    parse_stat_runtime_ns(&text, ticks_per_second)
}

fn parse_stat_runtime_ns(text: &str, ticks_per_second: u64) -> Option<u64> {
    let close = text.rfind(')')?;
    let rest = text.get(close + 2..)?;
    let fields: Vec<&str> = rest.split_whitespace().collect();
    let utime_ticks = fields.get(11)?.parse::<u64>().ok()?;
    let stime_ticks = fields.get(12)?.parse::<u64>().ok()?;
    let total_ticks = utime_ticks.saturating_add(stime_ticks);
    Some(total_ticks.saturating_mul(1_000_000_000) / ticks_per_second.max(1))
}

fn read_process_runtime_from_schedstat() -> Option<u64> {
    let text = std::fs::read_to_string("/proc/self/schedstat").ok()?;
    parse_schedstat_runtime_ns(&text)
}

fn parse_schedstat_runtime_ns(text: &str) -> Option<u64> {
    text.split_whitespace().next()?.parse::<u64>().ok()
}

fn clock_ticks_per_second() -> Option<&'static u64> {
    static CLOCK_TICKS_PER_SECOND: OnceLock<Option<u64>> = OnceLock::new();
    CLOCK_TICKS_PER_SECOND
        .get_or_init(|| {
            let output = Command::new("getconf").arg("CLK_TCK").output().ok()?;
            if !output.status.success() {
                return None;
            }
            String::from_utf8(output.stdout)
                .ok()?
                .trim()
                .parse::<u64>()
                .ok()
        })
        .as_ref()
}

#[cfg(test)]
mod tests {
    use super::*;
    use std::thread;
    use std::time::Duration;

    #[test]
    fn schedstat_runtime_reads() {
        assert!(read_process_runtime_ns().is_some());
    }

    #[test]
    fn schedstat_fallback_reader_is_non_panicking() {
        let _ = read_process_runtime_from_schedstat();
    }

    #[test]
    fn clock_tick_lookup_reads() {
        assert!(clock_ticks_per_second().is_some());
    }

    #[test]
    fn sampler_returns_percentage_after_two_samples() {
        let mut sampler = ProcessCpuSampler::new();
        assert!(sampler.sample_percent().is_none());
        thread::sleep(Duration::from_millis(10));
        let _ = sampler.sample_percent();
    }

    #[test]
    fn sampler_handles_manual_runtime_edges() {
        let mut sampler = ProcessCpuSampler::new();
        let now = Instant::now();
        assert!(sampler.sample_percent_at(now, 1_000).is_none());
        assert_eq!(
            sampler
                .sample_percent_at(now + Duration::from_millis(10), 2_000)
                .expect("percentage"),
            0.01
        );
        assert!(
            sampler
                .sample_percent_at(now + Duration::from_millis(20), 1_500)
                .is_none()
        );
        assert!(
            sampler
                .sample_percent_at(now + Duration::from_millis(20), 1_500)
                .is_none()
        );
    }

    #[test]
    fn procfs_parsers_handle_valid_and_malformed_samples() {
        let stat = "123 (lesavka worker) S 1 2 3 4 5 6 7 8 9 10 11 12 13 21 17";
        assert_eq!(parse_stat_runtime_ns(stat, 100), Some(230_000_000));
        assert_eq!(parse_stat_runtime_ns("missing close paren", 100), None);
        assert_eq!(parse_stat_runtime_ns("123 (name) S too short", 100), None);
        assert_eq!(parse_schedstat_runtime_ns("12345 678 9"), Some(12345));
        assert_eq!(parse_schedstat_runtime_ns("not-a-number 678"), None);
        assert_eq!(parse_schedstat_runtime_ns(""), None);
    }

    #[test]
    fn tenths_percent_is_clamped_to_integer_range() {
        let mut sampler = ProcessCpuSampler::new();
        let now = Instant::now();
        assert!(sampler.sample_percent_at(now, 0).is_none());
        assert_eq!(
            sampler
                .sample_percent_at(now + Duration::from_millis(1), u64::MAX)
                .map(|pct| (pct * 10.0).clamp(0.0, u32::MAX as f32) as u32),
            Some(u32::MAX)
        );
    }
}