Bug 1447512

As I mentioned in

https://bugzilla.redhat.com/show_bug.cgi?id=1446698

Disable list_modules and list_files in lsinitrd will significantly reduce dracut build time (in my case ~30% reduce in dracut and ~15% reduce in kdumpctl start with/without rebuild). There are mainly 4 sources of overhead caused by lsinitrd:



1. In dracut script (with rebuild)

if (( maxloglvl >= 5 )); then
    if [[ $allowlocal ]]; then
	"$dracutbasedir/lsinitrd.sh" "${DRACUT_TMPDIR}/initramfs.img"| ddebug
    else
        lsinitrd "${DRACUT_TMPDIR}/initramfs.img"| ddebug
    fi
fi

These verbose statements will be triggered by default, by kdumpctl calling dracut. Removing these leads to a reduce from 18s to 13s in my QEMU machine.



2. In kdumpctl script,  (with/without rebuild)

_dracut_args=$(lsinitrd $TARGET_INITRD | grep "^Arguments:" | head -1)
if [[ -z "$_dracut_args" ]];then
	echo "Warning: No dracut arguments found in initrd"
	return 0
fi

This lsinitrd call needs only Arguments lines but all the information are dumped by default.



3. In kdumpctl script, check_wdt_modified() (with/without rebuild)

wdtdrv=$(< "$dir/device/modalias")
wdtdrv=$(modinfo $wdtdrv | grep filename | awk -F"kernel/" '{print $2}')
active=$(< "$dir/state")
# rebuild when:
# module for this watchdog is not found and watchdog is active
# module for this watchdog is found and watchdog is inactive
lsinitrd $TARGET_INITRD | grep $wdtdrv &> /dev/null

This call of lsinitrd needs only module info.
BTW, the wdtdrv variable seems not getting a valid module's name but its alias instead, and at least on my PC they are not the same thing.



4. In kdumpctl script, check_wdt_modified() (with/without rebuild)

# check if watchdog kernel module unloaded.
loaded_mods=$(lsinitrd $TARGET_INITRD -f etc/cmdline.d/00-watchdog.conf)

This call of lsinitrd seems to be dead code because the 00-watchdog.conf file semms not an always valid path.

(In reply to Ziyue Yang from comment #3)

> 1. In dracut script (with rebuild)
> 
> if (( maxloglvl >= 5 )); then
>     if [[ $allowlocal ]]; then
> 	"$dracutbasedir/lsinitrd.sh" "${DRACUT_TMPDIR}/initramfs.img"| ddebug
>     else
>         lsinitrd "${DRACUT_TMPDIR}/initramfs.img"| ddebug
>     fi
> fi
> 
> These verbose statements will be triggered by default, by kdumpctl calling
> dracut. Removing these leads to a reduce from 18s to 13s in my QEMU machine.
> 

Confirmed that maxloglvl will be 5, set by sysloglvl, in root mode, causing lsinitrd overhead.

*** Bug 1459622 has been marked as a duplicate of this bug. ***

Hi Ziyue, you should create a dedicated bug for RHEL7 dracut, rather than depending on a Fedora bug.

After the backport of bz1484701, an estimation of time consume of each component in my VM is as below:
It's a rough classification of each stage but could be a good reference.

Kdumpclt Check feasibility:	14ms
Kdumpclt Check Config:		15ms
Kdumpclt Check Status:		17ms
Mkdumprd Prepare:		254ms
Mkdumprd parse config:		415ms
Dracut Parse Args:		361ms
Dracut Select Modules:		454ms
Dracut Prepare Install Modules:	29ms
Dracut Install Modules:		4694ms
Dracut Prepare Compress Image:	748ms
Dracut Compress Image:		4479ms
Dracut Cleanup:			12ms

kexec load and unload in kernel is not counted, took around 0.3s. 

11492 (11.5s) in total.

Before backport of bz1484701 it will use about 14s in my VM.

Most time is now consumed by install dracut modules and compress image.
Mkdumprd parse config could be optimzied to cost only ~10ms by backporting a not merged upstream patch:
[PATCH] mkdumprd: Improve the config reading logic

Dracut module installation related part should be optimized to speed up about ~1000ms in RHEL-7 by backport:
https://github.com/dracutdevs/dracut/pull/545

But it's a bit late for dracut backport, and will cause conflict, and also changes too many codes. So would be better to only keep optimizing the time consumption in RHEL-8, and leave RHEL-7.7 with current time consumption for now.

As commented in comment 14, closing this bug.

*** This bug has been marked as a duplicate of bug 1484701 ***