perf(dracut-functions.sh): optimise get_persistent_dev() #2003
Conversation
nabijaczleweli
requested review from
haraldh,
danimo and
johannbg
as code owners
nabijaczleweli
force-pushed
the
120ms
branch
2 times, most recently
from
4f1262b to
f9f4ab1
Compare
Head branch was pushed to by a user without write access
|
Admittedly, it only happened when I extended the stat cmdline to 175x its original size on my mid-sized system (lsblk is 31 devices with grep -v p and 92 without), but I had been worried about just blowing the stat environment block size. The contenders are: sget_persistent_dev() {
local _dev _pol
_dev=$(stat -L -c $'%t:%T' "$1" 2> /dev/null)
[ -z "$_dev" ] && return
if [[ -n $persistent_policy ]]; then
_pol="/dev/disk/${persistent_policy}/*"
else
_pol=
fi
# shellcheck disable=SC2086
stat -L -c $'%n\t%t:%T' \
$_pol \
/dev/mapper/* \
/dev/disk/by-uuid/* \
/dev/disk/by-label/* \
/dev/disk/by-partuuid/* \
/dev/disk/by-partlabel/* \
/dev/disk/by-id/* \
/dev/disk/by-path/* 2> /dev/null \
| awk -v_dev="$_dev" -vdfl="$1" '$0 ~ ("\t" _dev "$") {--NF; print; dfl=0; exit} END {if(dfl) print dfl}'
}
pget_persistent_dev() {
local _dev _pol
_dev=$(stat -L -c $'%t:%T' "$1" 2> /dev/null)
[ -z "$_dev" ] && return
if [[ -n $persistent_policy ]]; then
_pol="/dev/disk/${persistent_policy}/*"
else
_pol=
fi
# shellcheck disable=SC2086
printf '%s\0' \
$_pol \
/dev/mapper/* \
/dev/disk/by-uuid/* \
/dev/disk/by-label/* \
/dev/disk/by-partuuid/* \
/dev/disk/by-partlabel/* \
/dev/disk/by-id/* \
/dev/disk/by-path/* \
| xargs -0 stat -L -c $'%n\t%t:%T' 2> /dev/null \
| awk -v_dev="$_dev" -vdfl="$1" '$0 ~ ("\t" _dev "$") {--NF; print; dfl=0; exit} END {if(dfl) print dfl}'
}
fget_persistent_dev() {
local _dev _pol
_dev=$(stat -L -c $'%t:%T' "$1" 2> /dev/null)
[ -z "$_dev" ] && return
if [[ -n $persistent_policy ]]; then
_pol="/dev/disk/${persistent_policy}"
else
_pol=
fi
# shellcheck disable=SC2086
find \
$_pol \
/dev/mapper \
/dev/disk/by-uuid \
/dev/disk/by-label \
/dev/disk/by-partuuid \
/dev/disk/by-partlabel \
/dev/disk/by-id \
/dev/disk/by-path \
-print0 \
| xargs -0 stat -L -c $'%n\t%t:%T' 2> /dev/null \
| awk -v_dev="$_dev" -vdfl="$1" '$0 ~ ("\t" _dev "$") {--NF; print; dfl=0; exit} END {if(dfl) print dfl}'
}And the results: Which I'd classify as "who the hell knows" because, really, who does, it's all roughly within jitter, so I've rebased and updated to use printf because I like it more and it avoids an exec. |
This shaves ~40ms (from ~50ms; average case, huge variance up to 700ms on the old one), and makes 99base go from real 0m0.350s user 0m0.250s sys 0m0.119s to real 0m0.216s user 0m0.162s sys 0m0.076s
|
This issue is being marked as stale because it has not had any recent activity. It will be closed if no further activity occurs. If this is still an issue in the latest release of Dracut and you would like to keep it open please comment on this issue within the next 7 days. Thank you for your contributions. |
|
world's stinkiest bot 2022 (real) |
|
This issue is being marked as stale because it has not had any recent activity. It will be closed if no further activity occurs. If this is still an issue in the latest release of Dracut and you would like to keep it open please comment on this issue within the next 7 days. Thank you for your contributions. |
|
so true botstie |
|
This issue is being marked as stale because it has not had any recent activity. It will be closed if no further activity occurs. If this is still an issue in the latest release of Dracut and you would like to keep it open please comment on this issue within the next 7 days. Thank you for your contributions. |
|
oh no, cringe (dutch) |
LaszloGombos
added
enhancement
99base install() goes from like
to like
on my system.
Microbenchmarks show (appx. an easy-average case):
For best case (this is the first device in the glob):
And for worst case (
/dev/zerodoesn't have a corresponding device blockdev in the glob):Or, in short:
In the cases that are run in the 99base setup, I saw baseline speeds of 40-50ms, and it's run like 3 times there for me. If you have a less weird (non-root-on-ZFS) or more weird (wide btrfs) setup you'll likely see a better improvement.
get_persistent_dev()is only used inmodule-setup.shsChecklist