[PATCH] e2scrub_all: refactor device probe loop

[PATCH] e2scrub_all: refactor device probe loop

[Darrick J. Wong]

From: Darrick J. Wong <darrick.wong@oracle.com>

Paul Menzel reported that the e2scrub_all reaper service that runs at
startup takes a long time to run, and Ted T'so pointed out that we could
do a lot less work by using lvs as the outer loop in the ext4 filesystem
probe function so that we only have to lsblk the lvm devices containing
ext4 filesystems.

Therefore, refactor the loops to put lvs first, which should boost speed
a bit.

Reported-by: Paul Menzel <pmenzel@molgen.mpg.de>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
---
 scrub/e2scrub_all.in |   51 ++++++++++++++++++++++++++++++++++++--------------
 1 file changed, 37 insertions(+), 14 deletions(-)

diff --git a/scrub/e2scrub_all.in b/scrub/e2scrub_all.in
index 23d122d25..41420d03d 100644
--- a/scrub/e2scrub_all.in
+++ b/scrub/e2scrub_all.in
@@ -19,6 +19,7 @@
 #  Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301, USA.
 
 scrub_all=0
+reap=0
 conffile="@root_sysconfdir@/e2scrub.conf"
 
 test -f "${conffile}" && . "${conffile}"
@@ -61,7 +62,7 @@ exitcode() {
 while getopts "ArV" opt; do
 	case "${opt}" in
 	"A") scrub_all=1;;
-	"r") scrub_args="${scrub_args} -r";;
+	"r") scrub_args="${scrub_args} -r"; reap=1;;
 	"V") print_version; exitcode 0;;
 	*) print_help; exitcode 2;;
 	esac
@@ -69,27 +70,27 @@ done
 shift "$((OPTIND - 1))"
 
 # Find scrub targets, make sure we only do this once.
-ls_scrub_targets() {
-	lsblk -o NAME,FSTYPE,MOUNTPOINT -p -P -n | while read vars; do
+ls_scan_targets() {
+	lvs --name-prefixes -o vg_name,lv_name,lv_path \
+			-S lv_active=active,lv_role=public --noheadings | \
+	while read vars; do
 		eval "${vars}"
+		eval "$(lsblk -o NAME,FSTYPE,MOUNTPOINT -p -P -n "${LVM2_LV_PATH}")"
 
-		# Skip non-ext[234]
+		# Skip unless ext*
 		case "${FSTYPE}" in
-		ext[234])	;;
-		*)		continue;;
+		ext*) ;;
+		*) continue;;
 		esac
 
+		# Don't run against a snapshot ever
+		echo "${LVM2_LV_ROLE}" | grep -q "snapshot" && continue
+
 		# Skip unmounted filesystems unless -A
 		if [ "${scrub_all}" -eq 0 ] && [ -z "${MOUNTPOINT}" ]; then
 			continue;
 		fi
 
-		# Skip non-lvm devices and lvm snapshots
-		lvm_vars="$(lvs --nameprefixes -o vg_name,lv_name,lv_role --noheadings "${NAME}" 2> /dev/null)"
-		test $? -ne 0 && continue
-		eval "${lvm_vars}"
-		echo "${LVM2_LV_ROLE}" | grep -q "snapshot" && continue
-
 		if [ -n "${MOUNTPOINT}" ]; then
 			echo "${MOUNTPOINT}"
 		else
@@ -98,6 +99,28 @@ ls_scrub_targets() {
 	done | sort | uniq
 }
 
+# Find leftover scrub snapshots
+ls_reap_targets() {
+	lvs --name-prefixes -o vg_name,lv_name,lv_path,origin \
+		-S lv_role=snapshot --noheadings | while read vars; do
+		eval "${vars}"
+
+		# Filter out anything except our snapshots
+		case "${LVM2_LV_NAME}" in
+		*.e2scrub) echo "${LVM2_LV_PATH}";;
+		esac
+	done | sort | uniq
+}
+
+# Figure out what we're targeting
+ls_targets() {
+	if [ "${reap}" -eq 1 ]; then
+		ls_reap_targets
+	else
+		ls_scan_targets
+	fi
+}
+
 # systemd doesn't know to do path escaping on the instance variable we pass
 # to the e2scrub service, which breaks things if there is a dash in the path
 # name.  Therefore, do the path escaping ourselves if needed.
@@ -118,10 +141,10 @@ escape_path_for_systemd() {
 
 # Scrub any mounted fs on lvm by creating a snapshot and fscking that.
 stdin="$(realpath /dev/stdin)"
-ls_scrub_targets | while read tgt; do
+ls_targets | while read tgt; do
 	# If we're not reaping and systemd is present, try invoking the
 	# systemd service.
-	if [ -z "${scrub_args}" ] && type systemctl > /dev/null 2>&1; then
+	if [ "${reap}" -ne 1 ] && type systemctl > /dev/null 2>&1; then
 		tgt_esc="$(escape_path_for_systemd "${tgt}")"
 		${DBG} systemctl start "e2scrub@${tgt_esc}" 2> /dev/null < "${stdin}"
 		res=$?

Re: [PATCH] e2scrub_all: refactor device probe loop

[Lukas Czerner]

On Mon, Mar 18, 2019 at 05:17:32PM -0700, Darrick J. Wong wrote:
> From: Darrick J. Wong <darrick.wong@oracle.com>
> 
> Paul Menzel reported that the e2scrub_all reaper service that runs at
> startup takes a long time to run, and Ted T'so pointed out that we could
> do a lot less work by using lvs as the outer loop in the ext4 filesystem
> probe function so that we only have to lsblk the lvm devices containing
> ext4 filesystems.
> 
> Therefore, refactor the loops to put lvs first, which should boost speed
> a bit.
> 
> Reported-by: Paul Menzel <pmenzel@molgen.mpg.de>
> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> ---
>  scrub/e2scrub_all.in |   51 ++++++++++++++++++++++++++++++++++++--------------
>  1 file changed, 37 insertions(+), 14 deletions(-)
> 
> diff --git a/scrub/e2scrub_all.in b/scrub/e2scrub_all.in
> index 23d122d25..41420d03d 100644
> --- a/scrub/e2scrub_all.in
> +++ b/scrub/e2scrub_all.in
> @@ -19,6 +19,7 @@
>  #  Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301, USA.
>  
>  scrub_all=0
> +reap=0
>  conffile="@root_sysconfdir@/e2scrub.conf"
>  
>  test -f "${conffile}" && . "${conffile}"
> @@ -61,7 +62,7 @@ exitcode() {
>  while getopts "ArV" opt; do
>  	case "${opt}" in
>  	"A") scrub_all=1;;
> -	"r") scrub_args="${scrub_args} -r";;
> +	"r") scrub_args="${scrub_args} -r"; reap=1;;
>  	"V") print_version; exitcode 0;;
>  	*) print_help; exitcode 2;;
>  	esac
> @@ -69,27 +70,27 @@ done
>  shift "$((OPTIND - 1))"
>  
>  # Find scrub targets, make sure we only do this once.
> -ls_scrub_targets() {
> -	lsblk -o NAME,FSTYPE,MOUNTPOINT -p -P -n | while read vars; do
> +ls_scan_targets() {
> +	lvs --name-prefixes -o vg_name,lv_name,lv_path \
> +			-S lv_active=active,lv_role=public --noheadings | \

You're not using vg_name, nor lv_name so you can drop it maybe ? Also
you're missing lv_role since you're checking it later, however you can
do this instead

	-S lv_active=active,lv_role=public -S lv_role!=snapshot

so you only need to ask for lv_path

> +	while read vars; do
>  		eval "${vars}"
> +		eval "$(lsblk -o NAME,FSTYPE,MOUNTPOINT -p -P -n "${LVM2_LV_PATH}")"

Now that you only have one lv filed you can maybe pass this directly to
the lsblk ?


This looks funny but works fine for me :)

lsblk -o MOUNTPOINT,NAME,FSTYPE -p -n `lvs -o lv_path -S lv_active=active,lv_role=public -S lv_role!=snapshot --noheadings` | grep 'ext[234]' | awk '{print $1}'

if you only want mounted file systems than you just add

	| grep -v -E '^/dev/'

>  
> -		# Skip non-ext[234]
> +		# Skip unless ext*
>  		case "${FSTYPE}" in
> -		ext[234])	;;
> -		*)		continue;;
> +		ext*) ;;
> +		*) continue;;
>  		esac
>  
> +		# Don't run against a snapshot ever
> +		echo "${LVM2_LV_ROLE}" | grep -q "snapshot" && continue
> +
>  		# Skip unmounted filesystems unless -A
>  		if [ "${scrub_all}" -eq 0 ] && [ -z "${MOUNTPOINT}" ]; then
>  			continue;
>  		fi
>  
> -		# Skip non-lvm devices and lvm snapshots
> -		lvm_vars="$(lvs --nameprefixes -o vg_name,lv_name,lv_role --noheadings "${NAME}" 2> /dev/null)"
> -		test $? -ne 0 && continue
> -		eval "${lvm_vars}"
> -		echo "${LVM2_LV_ROLE}" | grep -q "snapshot" && continue
> -
>  		if [ -n "${MOUNTPOINT}" ]; then
>  			echo "${MOUNTPOINT}"
>  		else
> @@ -98,6 +99,28 @@ ls_scrub_targets() {
>  	done | sort | uniq
>  }
>  
> +# Find leftover scrub snapshots
> +ls_reap_targets() {
> +	lvs --name-prefixes -o vg_name,lv_name,lv_path,origin \

You only seem to be using lv_name and lv_path ?

also I think you can do this:

lvs -o lv_path -S lv_role=snapshot -S lv_name=~\(e2scrub$\) --noheadings

Also since you only ask for one field I tihnk it's already sorted so no
need to sort, or add -O

-Lukas

> +		-S lv_role=snapshot --noheadings | while read vars; do
> +		eval "${vars}"
> +
> +		# Filter out anything except our snapshots
> +		case "${LVM2_LV_NAME}" in
> +		*.e2scrub) echo "${LVM2_LV_PATH}";;
> +		esac
> +	done | sort | uniq
> +}
> +
> +# Figure out what we're targeting
> +ls_targets() {
> +	if [ "${reap}" -eq 1 ]; then
> +		ls_reap_targets
> +	else
> +		ls_scan_targets
> +	fi
> +}
> +
>  # systemd doesn't know to do path escaping on the instance variable we pass
>  # to the e2scrub service, which breaks things if there is a dash in the path
>  # name.  Therefore, do the path escaping ourselves if needed.
> @@ -118,10 +141,10 @@ escape_path_for_systemd() {
>  
>  # Scrub any mounted fs on lvm by creating a snapshot and fscking that.
>  stdin="$(realpath /dev/stdin)"
> -ls_scrub_targets | while read tgt; do
> +ls_targets | while read tgt; do
>  	# If we're not reaping and systemd is present, try invoking the
>  	# systemd service.
> -	if [ -z "${scrub_args}" ] && type systemctl > /dev/null 2>&1; then
> +	if [ "${reap}" -ne 1 ] && type systemctl > /dev/null 2>&1; then
>  		tgt_esc="$(escape_path_for_systemd "${tgt}")"
>  		${DBG} systemctl start "e2scrub@${tgt_esc}" 2> /dev/null < "${stdin}"
>  		res=$?

Re: [PATCH] e2scrub_all: refactor device probe loop

[Theodore Ts'o]

On Tue, Mar 19, 2019 at 11:58:24AM +0100, Lukas Czerner wrote:
> >  # Find scrub targets, make sure we only do this once.
> > -ls_scrub_targets() {
> > -	lsblk -o NAME,FSTYPE,MOUNTPOINT -p -P -n | while read vars; do
> > +ls_scan_targets() {
> > +	lvs --name-prefixes -o vg_name,lv_name,lv_path \
> > +			-S lv_active=active,lv_role=public --noheadings | \
> 
> You're not using vg_name, nor lv_name so you can drop it maybe ? Also
> you're missing lv_role since you're checking it later, however you can
> do this instead
> 
> 	-S lv_active=active,lv_role=public -S lv_role!=snapshot

We don't need to check lv_role at all, since the command already
included:

	-S lv_active=active,lv_role=public

And we need to do lv_role=public and not lv_role!=snapshot, since we
need to also exclude thinpools.

Speaking of thinpools, although we *can* take a thick snapshot of a
thinp volume, in the future we might want to add support for taking a
thinp snapshot.  (Although that will make the check which I've been
working on to make sure there's enough free space a bit more
complicated.)

I have some other patches queued for e2scrub, so I'll take a whack and
trying to make a reivison of this patch, and then send out a stack of
patches against e2scrub for folks to look at.

			    	     	  - Ted

Re: [PATCH] e2scrub_all: refactor device probe loop

[Lukas Czerner]

On Wed, Mar 20, 2019 at 12:17:08PM -0400, Theodore Ts'o wrote:
> On Tue, Mar 19, 2019 at 11:58:24AM +0100, Lukas Czerner wrote:
> > >  # Find scrub targets, make sure we only do this once.
> > > -ls_scrub_targets() {
> > > -	lsblk -o NAME,FSTYPE,MOUNTPOINT -p -P -n | while read vars; do
> > > +ls_scan_targets() {
> > > +	lvs --name-prefixes -o vg_name,lv_name,lv_path \
> > > +			-S lv_active=active,lv_role=public --noheadings | \
> > 
> > You're not using vg_name, nor lv_name so you can drop it maybe ? Also
> > you're missing lv_role since you're checking it later, however you can
> > do this instead
> > 
> > 	-S lv_active=active,lv_role=public -S lv_role!=snapshot
> 
> We don't need to check lv_role at all, since the command already
> included:
> 
> 	-S lv_active=active,lv_role=public

You actually need to check the lv_role for snapshot, because snapshot can
be "public" as well.

> 
> And we need to do lv_role=public and not lv_role!=snapshot, since we
> need to also exclude thinpools.

No, you need to check both if you want to exclude snapshots.


I think that specifying -S twice tricked you, sorry about that. But
again you can do:

	-S lv_active=active,lv_role=public,lv_role!=snapshot

Thanks!
-Lukas