quick_search/crates/quicksearch-core/examples/common/mod.rs

139 lines
4.8 KiB
Rust
Raw Normal View History

//! Shared measurement plumbing for the `examples/*probe.rs` binaries.
//!
//! A `mod.rs` one directory down is not an auto-discovered target, so this
//! compiles only as a module of whichever probe writes `mod common;`.
//! Everything here reads the machine, not the library; none of it ships.
// The unused half is not dead code; it is another binary's.
#![allow(dead_code)]
use std::path::{Path, PathBuf};
/// Drop a tree — and optionally an index — from the page cache: without
/// this every read figure is `0.0 MiB` and anything measured about
/// traversal or read cost is measuring memory. `fadvise(DONTNEED)` needs
/// no root and evicts only this tree, but leaves dentry/inode caches warm;
/// on virtiofs/NFS the host cache survives, so "cold" there is a lower bound.
#[cfg(unix)]
pub fn evict(tree: &Path, db: Option<&Path>) -> (usize, u64) {
use std::os::unix::io::AsRawFd;
// Dirty pages cannot be dropped; flush before asking.
unsafe { libc::sync() };
fn drop_one(path: &Path) -> Option<u64> {
let file = std::fs::File::open(path).ok()?;
let len = file.metadata().ok()?.len();
// (0, 0) means "to end of file".
let rc = unsafe { libc::posix_fadvise(file.as_raw_fd(), 0, 0, libc::POSIX_FADV_DONTNEED) };
(rc == 0).then_some(len)
}
fn walk(dir: &Path, files: &mut usize, bytes: &mut u64) {
let Ok(entries) = std::fs::read_dir(dir) else {
return;
};
for entry in entries.flatten() {
let path = entry.path();
match entry.file_type() {
Ok(ft) if ft.is_dir() => walk(&path, files, bytes),
Ok(ft) if ft.is_file() => {
if let Some(len) = drop_one(&path) {
*files += 1;
*bytes += len;
}
}
_ => {}
}
}
}
let (mut files, mut bytes) = (0usize, 0u64);
walk(tree, &mut files, &mut bytes);
if let Some(db) = db {
for suffix in ["", "-wal", "-shm"] {
let p = PathBuf::from(format!("{}{}", db.display(), suffix));
if let Some(len) = drop_one(&p) {
files += 1;
bytes += len;
}
}
}
(files, bytes)
}
/// No equivalent worth having off Unix.
#[cfg(not(unix))]
pub fn evict(_tree: &Path, _db: Option<&Path>) -> (usize, u64) {
eprintln!("evict: not supported on this platform");
(0, 0)
}
/// The kernel's accounting from `/proc/self/io`: `read_bytes` is what
/// reached the block layer (the disk, not the page cache), and `syscr`
/// separates "read a lot" from "read a little, many times". Zero on
/// filesystems that don't report it; the caller says so.
#[derive(Default, Clone, Copy)]
pub struct Io {
pub rchar: u64,
pub wchar: u64,
pub syscr: u64,
pub syscw: u64,
pub read_bytes: u64,
pub write_bytes: u64,
pub cancelled: u64,
}
impl Io {
pub fn read() -> Io {
let mut io = Io::default();
let Ok(text) = std::fs::read_to_string("/proc/self/io") else {
return io;
};
for line in text.lines() {
let Some((key, value)) = line.split_once(':') else {
continue;
};
let Ok(value) = value.trim().parse::<u64>() else {
continue;
};
match key {
"rchar" => io.rchar = value,
"wchar" => io.wchar = value,
"syscr" => io.syscr = value,
"syscw" => io.syscw = value,
"read_bytes" => io.read_bytes = value,
"write_bytes" => io.write_bytes = value,
"cancelled_write_bytes" => io.cancelled = value,
_ => {}
}
}
io
}
pub fn since(&self, start: &Io) -> Io {
Io {
rchar: self.rchar.saturating_sub(start.rchar),
wchar: self.wchar.saturating_sub(start.wchar),
syscr: self.syscr.saturating_sub(start.syscr),
syscw: self.syscw.saturating_sub(start.syscw),
read_bytes: self.read_bytes.saturating_sub(start.read_bytes),
write_bytes: self.write_bytes.saturating_sub(start.write_bytes),
cancelled: self.cancelled.saturating_sub(start.cancelled),
}
}
}
pub fn mib(bytes: u64) -> String {
format!("{:.1} MiB", bytes as f64 / (1024.0 * 1024.0))
}
/// Peak RSS from the kernel's own `VmHWM` — unlike a sampled figure it
/// cannot miss a spike; `None` where `/proc` is absent.
pub fn vm_hwm() -> Option<u64> {
let status = std::fs::read_to_string("/proc/self/status").ok()?;
let line = status.lines().find(|l| l.starts_with("VmHWM:"))?;
let kib: u64 = line.split_whitespace().nth(1)?.parse().ok()?;
Some(kib * 1024)
}