mirror of
https://github.com/bootandy/dust.git
synced 2026-06-08 11:29:05 +03:00
Large refactor. Use rayon, 10X performance boost
Code changes: Removed ignore & channel crates. Using a single reciever thread to build a hashmap to prevend duplicate inodes being reported gave a severe performance penalty Using rayon crate with some hand crafted file traversal has improved performance aprox 10X Behaviour changes: Removed parameter 'limit by filesystem' - don't think this is used, and I only added it as it was easy to add with the ignore crate. Sym links will now not appear in the output tree unless using '-s' 'apparent-size' flag Change behaviour of multiple args so that it unifies them and compares them under one tree instead of treating them individually: https://github.com/bootandy/dust/issues/136
This commit is contained in:
+128
@@ -0,0 +1,128 @@
|
||||
#[allow(unused_imports)]
|
||||
use std::fs;
|
||||
|
||||
use std::path::Path;
|
||||
|
||||
#[cfg(target_family = "unix")]
|
||||
fn get_block_size() -> u64 {
|
||||
// All os specific implementations of MetatdataExt seem to define a block as 512 bytes
|
||||
// https://doc.rust-lang.org/std/os/linux/fs/trait.MetadataExt.html#tymethod.st_blocks
|
||||
512
|
||||
}
|
||||
|
||||
#[cfg(target_family = "unix")]
|
||||
pub fn get_metadata(d: &Path, use_apparent_size: bool) -> Option<(u64, Option<(u64, u64)>)> {
|
||||
use std::os::unix::fs::MetadataExt;
|
||||
match d.metadata() {
|
||||
Ok(md) => {
|
||||
if use_apparent_size {
|
||||
Some((md.len(), Some((md.ino(), md.dev()))))
|
||||
} else {
|
||||
Some((md.blocks() * get_block_size(), Some((md.ino(), md.dev()))))
|
||||
}
|
||||
}
|
||||
Err(_e) => None,
|
||||
}
|
||||
}
|
||||
|
||||
#[cfg(target_family = "windows")]
|
||||
pub fn get_metadata(d: &Path, _use_apparent_size: bool) -> Option<(u64, Option<(u64, u64)>)> {
|
||||
// On windows opening the file to get size, file ID and volume can be very
|
||||
// expensive because 1) it causes a few system calls, and more importantly 2) it can cause
|
||||
// windows defender to scan the file.
|
||||
// Therefore we try to avoid doing that for common cases, mainly those of
|
||||
// plain files:
|
||||
|
||||
// The idea is to make do with the file size that we get from the OS for
|
||||
// free as part of iterating a folder. Therefore we want to make sure that
|
||||
// it makes sense to use that free size information:
|
||||
|
||||
// Volume boundaries:
|
||||
// The user can ask us not to cross volume boundaries. If the DirEntry is a
|
||||
// plain file and not a reparse point or other non-trivial stuff, we assume
|
||||
// that the file is located on the same volume as the directory that
|
||||
// contains it.
|
||||
|
||||
// File ID:
|
||||
// This optimization does deprive us of access to a file ID. As a
|
||||
// workaround, we just make one up that hopefully does not collide with real
|
||||
// file IDs.
|
||||
// Hard links: Unresolved. We don't get inode/file index, so hard links
|
||||
// count once for each link. Hopefully they are not too commonly in use on
|
||||
// windows.
|
||||
|
||||
// Size:
|
||||
// We assume (naively?) that for the common cases the free size info is the
|
||||
// same as one would get by doing the expensive thing. Sparse, encrypted and
|
||||
// compressed files are not included in the common cases, as one can image
|
||||
// there being more than view on their size.
|
||||
|
||||
// Savings in orders of magnitude in terms of time, io and cpu have been
|
||||
// observed on hdd, windows 10, some 100Ks files taking up some hundreds of
|
||||
// GBs:
|
||||
// Consistently opening the file: 30 minutes.
|
||||
// With this optimization: 8 sec.
|
||||
|
||||
use std::io;
|
||||
use winapi_util::Handle;
|
||||
fn handle_from_path_limited<P: AsRef<Path>>(path: P) -> io::Result<Handle> {
|
||||
use std::fs::OpenOptions;
|
||||
use std::os::windows::fs::OpenOptionsExt;
|
||||
const FILE_READ_ATTRIBUTES: u32 = 0x0080;
|
||||
|
||||
// So, it seems that it does does have to be that expensive to open
|
||||
// files to get their info: Avoiding opening the file with the full
|
||||
// GENERIC_READ is key:
|
||||
|
||||
// https://docs.microsoft.com/en-us/windows/win32/secauthz/generic-access-rights:
|
||||
// "For example, a Windows file object maps the GENERIC_READ bit to the
|
||||
// READ_CONTROL and SYNCHRONIZE standard access rights and to the
|
||||
// FILE_READ_DATA, FILE_READ_EA, and FILE_READ_ATTRIBUTES
|
||||
// object-specific access rights"
|
||||
|
||||
// The flag FILE_READ_DATA seems to be the expensive one, so we'll avoid
|
||||
// that, and a most of the other ones. Simply because it seems that we
|
||||
// don't need them.
|
||||
|
||||
let file = OpenOptions::new()
|
||||
.access_mode(FILE_READ_ATTRIBUTES)
|
||||
.open(path)?;
|
||||
Ok(Handle::from_file(file))
|
||||
}
|
||||
|
||||
fn get_metadata_expensive(d: &Path) -> Option<(u64, Option<(u64, u64)>)> {
|
||||
use winapi_util::file::information;
|
||||
|
||||
let h = handle_from_path_limited(d).ok()?;
|
||||
let info = information(&h).ok()?;
|
||||
|
||||
Some((
|
||||
info.file_size(),
|
||||
Some((info.file_index(), info.volume_serial_number())),
|
||||
))
|
||||
}
|
||||
|
||||
use std::os::windows::fs::MetadataExt;
|
||||
match d.metadata() {
|
||||
Ok(ref md) => {
|
||||
const FILE_ATTRIBUTE_ARCHIVE: u32 = 0x20u32;
|
||||
const FILE_ATTRIBUTE_READONLY: u32 = 0x1u32;
|
||||
const FILE_ATTRIBUTE_HIDDEN: u32 = 0x2u32;
|
||||
const FILE_ATTRIBUTE_SYSTEM: u32 = 0x4u32;
|
||||
const FILE_ATTRIBUTE_NORMAL: u32 = 0x80u32;
|
||||
const FILE_ATTRIBUTE_DIRECTORY: u32 = 0x10u32;
|
||||
|
||||
let attr_filtered = md.file_attributes()
|
||||
& !(FILE_ATTRIBUTE_HIDDEN | FILE_ATTRIBUTE_READONLY | FILE_ATTRIBUTE_SYSTEM);
|
||||
if attr_filtered == FILE_ATTRIBUTE_ARCHIVE
|
||||
|| attr_filtered == FILE_ATTRIBUTE_DIRECTORY
|
||||
|| md.file_attributes() == FILE_ATTRIBUTE_NORMAL
|
||||
{
|
||||
Some((md.len(), None))
|
||||
} else {
|
||||
get_metadata_expensive(&d)
|
||||
}
|
||||
}
|
||||
_ => get_metadata_expensive(&d),
|
||||
}
|
||||
}
|
||||
Reference in New Issue
Block a user