LSD is very slow compared to ls when listing large directories

[JamsNJellies]

Listing the Void Linux repo srcpkgs directory (about 12055 directories) takes far longer with lsd than it does with GNU ls.

lsd --color never 1.39s user 0.12s system 99% cpu 1.521 total lsd 1.42s user 0.12s system 98% cpu 1.559 total ls 0.01s user 0.02s system 91% cpu 0.034 total

[ftpd]

+1 to this. I've just did the other test: compared lsd and exa when listing long output and tree view: exa -l -T 2.54s user 6.14s system 47% cpu 18.298 total lsd -l --tree 332.53s user 96.61s system 50% cpu 14:13.04 total (obviously no info from vanilla ls, as it doesn't have tree functionality).

I did it on my 13GB home directory (macOS, so many small files in ~/Library).

[ben2talk]

Yes. Also using in .zshrc or .bashrc

Automatically ls when changing directory

cd() { builtin cd "$@" && lsd

lsd comes up with Permission denied (oserror 13). errors when opening Dolphin terminal...

Compare with...

# Automatically ls when changing directory
cd() {
  builtin cd "$@" && exa --icons --group-directories-first  
}

[meain]

@ben2talk Permission denied is probably a different issues: https://github.com/Peltoche/lsd/issues/79

[hedonihilist]

Same here. The time of listing 287188 files:

❯ ls | wc -l     
287188
❯ time lsd 1>/dev/null
lsd > /dev/null  6.50s user 2.60s system 99% cpu 9.131 total
❯ time ls 1>/dev/null
ls > /dev/null  0.42s user 0.10s system 99% cpu 0.525 total

Whyt lsd is much slower than ls? Is there anything we can do?

[zwpaper]

there is a PR for speed up, but not yet finished. https://github.com/Peltoche/lsd/pull/441.

[arctic-penguin]

Since #441 was closed, is there any indication whether this issue will be fixed?

[arctic-penguin]

Here is how lsd compares to exa and ls at the moment. Test cases is a directory with 10,376 files (a fuzzing corpus).

$ hyperfine ls exa lsd
Benchmark 1: ls
  Time (mean ± σ):      16.6 ms ±   1.9 ms    [User: 12.4 ms, System: 4.1 ms]
  Range (min … max):    13.1 ms …  20.3 ms    144 runs

Benchmark 2: exa
  Time (mean ± σ):      47.9 ms ±   5.7 ms    [User: 22.5 ms, System: 25.0 ms]
  Range (min … max):    38.8 ms …  61.5 ms    57 runs

Benchmark 3: lsd
  Time (mean ± σ):      7.213 s ±  0.310 s    [User: 0.897 s, System: 1.699 s]
  Range (min … max):    7.043 s …  8.060 s    10 runs

  Warning: Statistical outliers were detected. Consider re-running this benchmark on a quiet system without any interferences from other programs. It might help to use the '--warmup' or '--prepare' options.

Summary
  'ls' ran
    2.89 ± 0.47 times faster than 'exa'
  434.83 ± 52.45 times faster than 'lsd'

So plain-old ls is 400 times faster than lsd.

$ ls --version
ls (GNU coreutils) 9.2
Copyright (C) 2023 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <https://gnu.org/licenses/gpl.html>.
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.

Written by Richard M. Stallman and David MacKenzie.
$ lsd --version
lsd 0.23.1
$ exa --version
exa - list files on the command-line
v0.10.1 [+git]
https://the.exa.website/

[planet36]

Use strace with ls, exa, and lsd in a large directory, and notice that lsd is making many more system calls that the other 2.

strace ls --color=never -l --sort=none 2> ~/strace.ls.txt strace exa --color=never --no-icons --long --sort=none 2> ~/strace.exa.txt strace lsd --color=never --icon=never --long --sort=none 2> ~/strace.lsd.txt

wc --lines --total=never strace.*.txt

3183 strace.exa.txt 4900 strace.ls.txt 408402 strace.lsd.txt

[planet36]

You can compare the number of particular system calls with these commands.

sed -E -e 's/\(.*//g' ~/strace.ls.txt | sort | uniq -c sed -E -e 's/\(.*//g' ~/strace.exa.txt | sort | uniq -c sed -E -e 's/\(.*//g' ~/strace.lsd.txt | sort | uniq -c

[planet36]

This is how I looked into the issue of lsd's bad performance.

I created a directory in /tmp with 10,000 empty files. On my Linux system, /tmp is a tmpfs, and bash was the shell I used.

mkdir /tmp/1E4
cd /tmp/1E4
touch $(seq 1E4)

Then I compared the execution time of exa, lsd, ls, and uu-ls on that directory.

The options used were intended to disable features that differ between programs (such as icons and colors) but still display "long" output.

printf '\n\nexa'   ; time LC_ALL=C exa --color=never --no-icons --long --sort=none > /dev/null
printf '\n\nlsd'   ; time LC_ALL=C lsd --color=never --icon=never --long --sort=none --ignore-config > /dev/null
printf '\n\nls'    ; time LC_ALL=C ls --color=never -l --sort=none > /dev/null
printf '\n\nuu-ls' ; time LC_ALL=C uu-ls --color=never -l --sort=none > /dev/null

exa
real    0m0.082s
user    0m0.064s
sys     0m0.045s


lsd
real    0m4.467s
user    0m1.506s
sys     0m1.913s


ls
real    0m0.028s
user    0m0.010s
sys     0m0.016s


uu-ls
real    0m0.036s
user    0m0.025s
sys     0m0.010s

Next I logged the system calls made by each program.

LC_ALL=C strace exa --color=never --no-icons --long --sort=none > /dev/null 2> ~/strace.exa.txt
LC_ALL=C strace lsd --color=never --icon=never --long --sort=none --ignore-config > /dev/null 2> ~/strace.lsd.txt
LC_ALL=C strace ls --color=never -l --sort=none > /dev/null 2> ~/strace.ls.txt
LC_ALL=C strace uu-ls --color=never -l --sort=none > /dev/null 2> ~/strace.uu-ls.txt

cd
wc --lines --total=never strace.*.txt

    21320 strace.exa.txt
  2777707 strace.lsd.txt
    20512 strace.ls.txt
    10499 strace.uu-ls.txt

You can see that lsd had many more system calls than the others. It's strace output was about 176M, so keep that in mind if you test on a directory with more files.

Here's a count of how many times each system call was made for each program.

printf '\n\nexa\n'   ; sed -n -E -e 's/\(.*//gp' strace.exa.txt   | sort | uniq -c
printf '\n\nlsd\n'   ; sed -n -E -e 's/\(.*//gp' strace.lsd.txt   | sort | uniq -c
printf '\n\nls\n'    ; sed -n -E -e 's/\(.*//gp' strace.ls.txt    | sort | uniq -c
printf '\n\nuu-ls\n' ; sed -n -E -e 's/\(.*//gp' strace.uu-ls.txt | sort | uniq -c

exa
      1 access
      2 arch_prctl
     34 brk
      4 clone3
     17 close
      1 execve
      1 exit_group
   1059 futex
     11 getdents64
      2 getrandom
      1 ioctl
     56 mmap
     19 mprotect
      5 mremap
      5 munmap
     16 newfstatat
     19 openat
      1 poll
      2 pread64
      2 prlimit64
     36 read
      1 rseq
      6 rt_sigaction
      9 rt_sigprocmask
      2 sched_getaffinity
      1 set_robust_list
      1 set_tid_address
      3 sigaltstack
  10001 statx
  10001 write


lsd
      1 access
      2 arch_prctl
  34008 brk
 800395 close
  20306 connect
  20002 epoll_create1
 120012 epoll_ctl
  80008 epoll_pwait2
    989 epoll_wait
      1 execve
      1 exit_group
      2 futex
  40013 getdents64
      1 getpid
      3 getrandom
      2 ioctl
  30003 lgetxattr
  30003 lseek
     30 mmap
      8 mprotect
      9 mremap
      7 munmap
 470059 newfstatat
 940106 openat
      1 poll
      6 prctl
      2 pread64
      2 prlimit64
  30015 read
  10001 readlink
  20991 recvfrom
      1 rseq
      5 rt_sigaction
  40004 rt_sigprocmask
      1 sched_getaffinity
  20002 sendto
      1 set_robust_list
      1 set_tid_address
      3 sigaltstack
  20306 socket
  10001 statx
  20002 timerfd_create
  20389 timerfd_settime
      1 write


ls
      1 access
      2 arch_prctl
      5 brk
     96 close
      6 connect
      2 epoll_create1
     12 epoll_ctl
      8 epoll_pwait2
      1 execve
      1 exit_group
      2 futex
     13 getdents64
      1 getpid
      2 getrandom
  10000 getxattr
      2 ioctl
      3 lseek
     28 mmap
      7 mprotect
      4 mremap
      3 munmap
     59 newfstatat
    105 openat
      6 prctl
      2 pread64
      1 prlimit64
     12 read
      2 recvfrom
      1 rseq
      4 rt_sigprocmask
      2 sendto
      1 set_robust_list
      1 set_tid_address
      6 socket
  10000 statx
      2 timerfd_create
      3 timerfd_settime
    105 write


uu-ls
      1 access
      2 arch_prctl
     11 brk
     95 close
      6 connect
      2 epoll_create1
     12 epoll_ctl
      8 epoll_pwait2
      1 execve
      1 exit_group
      2 futex
     13 getdents64
      1 getpid
      3 getrandom
      2 ioctl
      3 lseek
     28 mmap
      8 mprotect
      4 mremap
      5 munmap
     60 newfstatat
    108 openat
      1 poll
      6 prctl
      2 pread64
      2 prlimit64
     15 read
      2 recvfrom
      1 rseq
      5 rt_sigaction
      4 rt_sigprocmask
      1 sched_getaffinity
      2 sendto
      1 set_robust_list
      1 set_tid_address
      3 sigaltstack
      6 socket
  10001 statx
      2 timerfd_create
      2 timerfd_settime
     65 write

I used a spreadsheet to put the strace results in a table for easier comparison.

System call	exa	lsd	ls	uu-ls
access	1	1	1	1
arch_prctl	2	2	2	2
brk	34	34008	5	11
clone3	4
close	17	800395	96	95
connect		20306	6	6
epoll_create1		20002	2	2
epoll_ctl		120012	12	12
epoll_pwait2		80008	8	8
epoll_wait		989
execve	1	1	1	1
exit_group	1	1	1	1
futex	1059	2	2	2
getdents64	11	40013	13	13
getpid		1	1	1
getrandom	2	3	2	3
getxattr			10000
ioctl	1	2	2	2
lgetxattr		30003
lseek		30003	3	3
mmap	56	30	28	28
mprotect	19	8	7	8
mremap	5	9	4	4
munmap	5	7	3	5
newfstatat	16	470059	59	60
openat	19	940106	105	108
poll	1	1	6	1
prctl		6		6
pread64	2	2	2	2
prlimit64	2	2	1	2
read	36	30015	12	15
readlink		10001
recvfrom		20991	2	2
rseq	1	1	1	1
rt_sigaction	6	5		5
rt_sigprocmask	9	40004	4	4
sched_getaffinity	2	1		1
sendto		20002	2	2
set_robust_list	1	1	1	1
set_tid_address	1	1	1	1
sigaltstack	3	3		3
socket		20306	6	6
statx	10001	10001	10000	10001
timerfd_create		20002	2	2
timerfd_settime		20389	3	2
write	10001	1	105	65

Here are some of the calls that stood out to me.

grep -F 'readlink(' strace.lsd.txt

• readlink was called for every file, even though they aren't symbolic links.

grep -E '^(socket|connect)\(' strace.lsd.txt | sort | uniq -c | sort -n | tail

• 20,000 sockets were created and connected to "/run/systemd/userdb/io.systemd.DynamicUser".

grep -F '/etc/nsswitch.conf' strace.lsd.txt | sort | uniq -c | sort -n | tail

• "/etc/nsswitch.conf" was stat'd 30,000 times.

grep -E '^openat\(.+"/' strace.lsd.txt | sort | uniq -c | sort -n | tail

• "/etc/passwd" was opened and read 10,000 times.
• "/etc/group" was opened and read 20,000 times.
• "/run/systemd/userdb/" was opened 20,000 times.
• "/" was opened 150,000 times!

tl;dr: WTH?!

[CamJN]

as a workaround for now i turned off total-size which resulted in a 3632x speedup.

[planet36]

as a workaround for now i turned off total-size which resulted in a 3632x speedup.

The --total-size option isn't enabled by default. Not giving that option doesn't fix the systemic performance problems of lsd that I've shown above.

[CamJN]

It was enabled in my config file, I probably turned it on at some point, but it seems pretty much mandatory to have it disabled. Like I said the difference was massive, even if it doesn't address the overall slowness.

[planet36]

I added total-size: false to my config.yaml, but I didn't see a difference in performance.

Commands:

mkdir -p /tmp/1E4 ; cd /tmp/1E4 ; touch $(seq 1E4)
printf '\neza'          ; time LC_ALL=C command eza   --color=never --sort=none -l --no-icons --group > /dev/null
printf '\nlsd'          ; time LC_ALL=C command lsd   --color=never --sort=none -l --icon=never --ignore-config > /dev/null
printf '\nlsd (config)' ; time LC_ALL=C command lsd   --color=never --sort=none -l --icon=never > /dev/null
printf '\nls'           ; time LC_ALL=C command ls    --color=never --sort=none -l > /dev/null
printf '\nuu-ls'        ; time LC_ALL=C command uu-ls --color=never --sort=none -l > /dev/null

Output:

eza real 0m0.093s user 0m0.060s sys 0m0.072s

lsd real 0m5.200s user 0m1.325s sys 0m1.688s

lsd (config) real 0m5.974s user 0m1.165s sys 0m1.659s

ls real 0m0.064s user 0m0.034s sys 0m0.029s

uu-ls real 0m0.056s user 0m0.041s sys 0m0.014s

[tjex]

I just set total-size: false and it performs fast and snappy again. Maybe it's a clearer test to remove --ignore-config from the lsd command?

I added total-size: false to my config.yaml, but I didn't see a difference in performance.

Commands:

printf '\nlsd'          ; time LC_ALL=C command lsd   --color=never --sort=none -l --icon=never --ignore-config > /dev/null

[planet36]

I just set total-size: false and it performs fast and snappy again. Maybe it's a clearer test to remove --ignore-config from the lsd command?

No. In my example there were two calls of lsd: one that uses the config file, and one that ignores it. The intent was to show that neither method has acceptable performance.

[tjex]

ahhhhh. missed that (obviously).. Sorry!