Skip to content

Instantly share code, notes, and snippets.

@cabal95
Created July 25, 2015 17:53
Show Gist options
  • Save cabal95/e36c06e716d3328b512b to your computer and use it in GitHub Desktop.
Save cabal95/e36c06e716d3328b512b to your computer and use it in GitHub Desktop.
I use this script to backup my QEMU/KVM/libVirt virtual machines. The script requires KVM 2.1+ since it uses the live blockcommit mode. This means the data in the snapshot disk is rolled back into the original instead of the other way around. Script does NOT handle spaces in paths.
#!/bin/bash
#
BACKUPDEST="$1"
DOMAIN="$2"
MAXBACKUPS="$3"
if [ -z "$BACKUPDEST" -o -z "$DOMAIN" ]; then
echo "Usage: ./vm-backup <backup-folder> <domain> [max-backups]"
exit 1
fi
if [ -z "$MAXBACKUPS" ]; then
MAXBACKUPS=6
fi
echo "Beginning backup for $DOMAIN"
#
# Generate the backup path
#
BACKUPDATE=`date "+%Y-%m-%d.%H%M%S"`
BACKUPDOMAIN="$BACKUPDEST/$DOMAIN"
BACKUP="$BACKUPDOMAIN/$BACKUPDATE"
mkdir -p "$BACKUP"
#
# Get the list of targets (disks) and the image paths.
#
TARGETS=`virsh domblklist "$DOMAIN" --details | grep ^file | awk '{print $3}'`
IMAGES=`virsh domblklist "$DOMAIN" --details | grep ^file | awk '{print $4}'`
#
# Create the snapshot.
#
DISKSPEC=""
for t in $TARGETS; do
DISKSPEC="$DISKSPEC --diskspec $t,snapshot=external"
done
virsh snapshot-create-as --domain "$DOMAIN" --name backup --no-metadata \
--atomic --disk-only $DISKSPEC >/dev/null
if [ $? -ne 0 ]; then
echo "Failed to create snapshot for $DOMAIN"
exit 1
fi
#
# Copy disk images
#
for t in $IMAGES; do
NAME=`basename "$t"`
cp "$t" "$BACKUP"/"$NAME"
done
#
# Merge changes back.
#
BACKUPIMAGES=`virsh domblklist "$DOMAIN" --details | grep ^file | awk '{print $4}'`
for t in $TARGETS; do
virsh blockcommit "$DOMAIN" "$t" --active --pivot >/dev/null
if [ $? -ne 0 ]; then
echo "Could not merge changes for disk $t of $DOMAIN. VM may be in invalid state."
exit 1
fi
done
#
# Cleanup left over backup images.
#
for t in $BACKUPIMAGES; do
rm -f "$t"
done
#
# Dump the configuration information.
#
virsh dumpxml "$DOMAIN" >"$BACKUP/$DOMAIN.xml"
#
# Cleanup older backups.
#
LIST=`ls -r1 "$BACKUPDOMAIN" | grep -E '^[0-9]{4}-[0-9]{2}-[0-9]{2}\.[0-9]+$'`
i=1
for b in $LIST; do
if [ $i -gt "$MAXBACKUPS" ]; then
echo "Removing old backup "`basename $b`
rm -rf "$b"
fi
i=$[$i+1]
done
echo "Finished backup"
echo ""
@Marc176
Copy link

Marc176 commented Mar 24, 2020

@Ryushin: Thanks! :)

@brentl99
Copy link

brentl99 commented Apr 16, 2020

I have posted an update to the script I posted some time ago. There was a bug in the file retention logic (the last loop of the script) that assumed the name of the image matched the VM name. Also added to this script is the option to stream encrypt the images using gpg.

@tuomotalvitie
Copy link

tuomotalvitie commented Apr 21, 2020

Thank you for sharing these, and thank you @brentl99 for your variation. I added backing up the offline (well... not running) instances as well using the logic @panciom used for it, and created a check for skipping the backup if the modified timestamp of all the disks is older than the newest backup. Otherwise the logic is untouched (pivoting, retention, gpg).
It is quite possible that the check for not running should be explicitly for specific states (Shut off - Crashed? ...). Panciom checks for "in" which should cover "in shutdown".
Two disclaimers: 1. I don't speak .sh natively 2. I've been only running the script on dev servers.

#!/bin/bash
#
# [This script is slightly modified version of brentl99's script at
# https://gist.github.com/cabal95/e36c06e716d3328b512b - non-running instances 
# are also backed up, as implemented by panciom - additionally if the instance is
# offline, and the most recent backup is newer than disk modification times, the backup
# is skipped. Note: .xml is not checked for modification.]
#
# This script backs up a list of VMs.
# An overview of the process is as follows:
# * invokes a "snapshot" which transfers VM disk I/O to new "snapshot" image file(s).
# * copy (and encrypt if applicable) the VM's image filbrentl99e(s) to a backup
# * invoke a "blockcommit" which merges (or "pivots") the snapshot image back
#   to the VM's primary image file(s)
# * delete the snapshot image file(s)
# * make a copy of the VM define/XML file
# * delete old images and XMLs, retaining images for x dates where x = IMAGECNT
#
# Note: On CentOS 7 snapshotting requires the "-ev" version of qemu.
#       yum install centos-release-qemu-ev qemu-kvm-ev libvirt
#
# The script uses gzip to compress the source image (e.g. qcow2) on the fly 
# to the destination backup image. bzip2 was also tested, but bzip2 (and 
# other compression utilities) provide better compression (15-20%) but gzip
# is 7-10 times faster.
# 
# The script uses gpg symmetric encryption to encrypt the source image. The 
# encryption password is set using the ENCPASS field, and items must be decrypted
# before they are unzipped. Encrypted files will have an extension of .gz.gpg. By 
# leaving the ENCPASS field blank you can disable this feature.
#
# If the process fails part way through the snapshot, copy, or blockcommit, 
# the VM may be left running on the snapshot file which is Not desirable.
#
# TODO: optionally check the xml for modiication as well (and perhaps only 
# back it up, if only it has changes)
# TODO: parametrize Domains for clarity?
# TODO: perhaps check the permissions before starting, or run in test mode
# if no parameters, including sending mail

# define an emergency mail recipient
EMR=servermessages@mydomain.com
# encryption password. if left blank, files are not encrypted
#### gpg logic untouched from brentl99 version ####
#ENCPASS="MyPassworD"
HOST="$(hostname)"
SHCMD="$(basename -- $0)"
BACKUPROOT=/mydata01/qemu_backups
IMAGECNT=3
[ ! -f $BACKUPROOT/logs ] && mkdir -p $BACKUPROOT/logs
DATE="$(date +%Y-%m-%d.%H%M%S)"
LOG="$BACKUPROOT/logs/qemu-backup.$(date +%Y-%m-%d).log"
ERRORED=0
BREAK=false
SNAPPREFIX=snaptemp-

#Uncomment to backup domains myVMa and myVMb
#DOMAINS="myVMa myVMb"

#Alternatively list all VMs and back them all up
#DOMAINS=$(virsh list --all | tail -n +3 | awk '{print $2}')

# Path to XML files for stopped VM...
PATH_LIBVIRT_QEMU="/etc/libvirt/qemu"

# extract the date coding in filename (note: filename format must be YYYY-MM-DD)
dtmatch () { sed -n -e 's/.*\(2[0-1][0-9][0-9]-[0-1][0-9]-[0-3][0-9]\).*/\1/p'; }

echo "$SHCMD: Starting backups on $(date +'%d-%m-%Y %H:%M:%S')"  >> $LOG
for DOMAIN in $DOMAINS; do
	BREAK=false

	echo "---- VM Backup start $DOMAIN ---- $(date +'%d-%m-%Y %H:%M:%S')"  >> $LOG

	VM_RUNNING=1
	VMSTATE=$(virsh list --all | grep [[:space:]]$DOMAIN[[:space:]] | awk '{print $3}')
	if [[ $VMSTATE != "running" ]]; then
		echo "-> VM $DOMAIN not running. No snapshot and blockcommit. Only copy." >> $LOG
    		VM_RUNNING=0
	fi

	BACKUPFOLDER=$BACKUPROOT/$DOMAIN
	[ ! -d $BACKUPFOLDER ] && mkdir -p $BACKUPFOLDER
	TARGETS=$(virsh domblklist $DOMAIN --details | grep disk | awk '{print $3}')
	IMAGES=$(virsh domblklist $DOMAIN --details | grep disk | awk '{print $4}')

	# check to make sure the VM is running on a standard image, not
	# a snapshot that may be from a backup that previously failed
	for IMAGE in $IMAGES; do
		set -o noglob
		if [[ $IMAGE == *${SNAPPREFIX}* ]]; then
			set +o noglob
			ERR="$SHCMD: Error VM $DOMAIN is running on a snapshot disk image: $IMAGE"
			echo $ERR >> $LOG
			echo "$ERR
Host:       $HOST
Disk Image: $IMAGE
Domain:     $DOMAIN
Command:    virsh domblklist $DOMAIN --details" | mail -s "$SHCMD snapshot Exception for $DOMAIN" $EMR
			BREAK=true
			ERRORED=$(($ERRORED+1))
			break
		fi
		set +o noglob
	done
	[ $BREAK == true ] && continue

	if [[ $VM_RUNNING -eq 1 ]]; then
	# gather all the disks being used by the VM so they can be collectively snapshotted
	# TODO: Does this prefix check make sense for the targets (images certainly)
	DISKSPEC=""
	for TARGET in $TARGETS; do
		set -o noglob
		if [[ $TARGET == *${SNAPPREFIX}* ]]; then
			set +o noglob
			ERR="$SHCMD: Error VM $DOMAIN is running on a snapshot disk image: $TARGET"
			echo $ERR >> $LOG
			echo "$ERR
Host:       $HOST
Disk Image: $IMAGE
Domain:     $DOMAIN
Command:    $CMD" | mail -s "$SHCMD snapshot Exception for $DOMAIN" $EMR
			BREAK=true
			break
		fi
		set +o noglob
		DISKSPEC="$DISKSPEC --diskspec $TARGET,snapshot=external"
	done
		[ $BREAK == true ] && continue
		# transfer the VM to snapshot disk image(s)
			CMD="virsh snapshot-create-as --domain $DOMAIN --name ${SNAPPREFIX}$DOMAIN-$DATE --no-metadata --atomic --disk-only $DISKSPEC >> $LOG 2>&1"
			echo "Command: $CMD" >> $LOG 2>&1
			eval "$CMD"
			if [ $? -ne 0 ]; then
				ERR="Failed to create snapshot for $DOMAIN"
				echo $ERR >> $LOG
				echo "$ERR
Host:    $HOST
Domain:  $DOMAIN
Command: $CMD" | mail -s "$SHCMD snapshot Exception for $DOMAIN" $EMR
				ERRORED=$(($ERRORED+1))
				continue
			fi
	else
		unset -v LATEST
		for FILE in "$BACKUPFOLDER"/$(basename -- $IMAGE)-*.gz*; do
			if [[ ! -e "$FILE" ]]; then continue; fi
			[[ $FILE -nt $LATEST ]] && LATEST=$FILE
		done
		BACKUP=false
		for IMAGE in $IMAGES; do
			if [[ $IMAGE -nt $LATEST ]]; then
				BACKUP=true 
				break
			fi
		done
		if [ $BACKUP == false ]; then
			echo "-> Images of VM $DOMAIN older than the last backup (based on Modified)." >> $LOG
			echo "-> NOTE: .xml NOT checked" >> $LOG
			continue;
		fi
	fi

	# copy/back/compress the VM's disk image(s)
	for IMAGE in $IMAGES; do
		echo "Copying $IMAGE to $BACKUPFOLDER" >> $LOG
		ZFILE="$BACKUPFOLDER/$(basename -- $IMAGE)-$DATE.gz"
		# determine whether the gzip is to be encrypted or not
		if [ -z "${ENCPASS}" ]; then 
			CMD="gzip < $IMAGE > $ZFILE 2>> $LOG"
		else
			exec {pwout}> /tmp/agspw.$$
			exec {pwin}< /tmp/agspw.$$
			rm /tmp/agspw.$$
			echo $ENCPASS >&$pwout
			ZFILE="$ZFILE.gpg"
			CMD="gzip < $IMAGE --to-stdout | gpg --batch --yes -o $ZFILE --passphrase-fd $pwin -c >> $LOG"
		fi
		echo "Command: $CMD" >> $LOG
		SECS=$(printf "%.0f" $(/usr/bin/time -f %e sh -c "$CMD" 2>&1))
		printf '%s%dh:%dm:%ds\n' "Duration: " $(($SECS/3600)) $(($SECS%3600/60)) $(($SECS%60)) >> $LOG
		# clear fds if necessary
		if [ -n "${ENCPASS}" ]; then 
			exec {pwout}>&-
			exec {pwin}<&-
			unset pwout pwin
		fi
		BYTES=$(stat -c %s $IMAGE)
		printf "%s%'d\n" "Source MB: " $(($BYTES/1024/1024)) >> $LOG
                # Commented out below, as seed image operation takes 0s
		# printf "%s%'d\n" "kB/Second: " $(($BYTES/$SECS/1024)) >> $LOG
		ZBYTES=$(stat -c %s $ZFILE)
		printf "%s%'d\n" "Destination MB: " $(($ZBYTES/1024/1024)) >> $LOG
		printf "%s%d%s\n" "Compression: " $((($BYTES-$ZBYTES)*100/$BYTES)) "%" >> $LOG
	done

	if [[ $VM_RUNNING -eq 1 ]]; then
		# Update the VM's disk image(s) with any changes recorded in the snapshot
		# while the copy process was running.  In qemu lingo this is called a "pivot"
			BACKUPIMAGES=$(virsh domblklist $DOMAIN --details | grep disk | awk '{print $4}')
			for TARGET in $TARGETS; do
				CMD="virsh blockcommit $DOMAIN $TARGET --active --pivot >> $LOG 2>&1"
				echo "Command: $CMD" >> $LOG
				eval "$CMD"

				if [ $? -ne 0 ]; then
					ERR="Could not merge changes for disk of $TARGET of $DOMAIN. VM may be in an invalid state."
					echo $ERR >> $LOG
					echo "$ERR
Host:    $HOST
Domain:  $DOMAIN
Command: $CMD" | mail -s "$SHCMD blockcommit Exception for $DOMAIN" $EMR
					BREAK=true
					ERRORORED=$(($ERRORED+1))
					break
				fi
			done
		[ $BREAK == true ] && continue

		# Now that the VM's disk image(s) have been successfully committed/pivoted to
		# back to the main disk image, remove the temporary snapshot image file(s)
			for BACKUP in $BACKUPIMAGES; do
				set -o noglob
				if [[ $BACKUP == *${SNAPPREFIX}* ]]; then
					set +o noglob
					CMD="rm -f $BACKUP >> $LOG 2>&1"
					echo " Deleting temporary image $BACKUP" >> $LOG
					echo "Command: $CMD" >> $LOG
					eval "$CMD"
				fi
				set +o noglob
			done

		# capture the VM's definition in use at the time the backup was done
			CMD="virsh dumpxml $DOMAIN > $BACKUPFOLDER/$DOMAIN-$DATE.xml 2>> $LOG"
			echo "Command: $CMD" >> $LOG
			eval "$CMD"
	else
		# copy the VM's definition
		CMD="cp $PATH_LIBVIRT_QEMU/$DOMAIN.xml $BACKUPFOLDER/$DOMAIN-$DATE.xml 2>> $LOG"
		echo "Command: $CMD" >> $LOG
		eval "$CMD"
	fi

	# Tracks whether xmls have been cleared
	DDEL='no'
	# check image retention count
	for IMAGE in $IMAGES; do
		COUNT=`find $BACKUPFOLDER -type f -name $(basename -- $IMAGE)-'*'.gz'*' -print | dtmatch | sort -u | wc -l`
		if [ $COUNT -gt $IMAGECNT  ]; then
			echo "$SHCMD: Count for BACKUPFOLDER ($BACKUPFOLDER) for image ($(basename -- $IMAGE)) too high ($COUNT), deleting historical files over $IMAGECNT..."
			LIST=`find $BACKUPFOLDER -type f -name $(basename -- $IMAGE)-'*'.gz'*' -print | dtmatch | sort -ur | sed -e "1,$IMAGECNT"d`
			# make sure LIST has a value otherwise fgrep will allow the entire find
			# result to be passed to xarg rm
			if [ -n "$LIST" ]; then
				# Delete the specific images in the dates list
				find $BACKUPFOLDER -type f -name $(basename -- $IMAGE)-'*' | fgrep "$LIST" | xargs rm
				# Only delete old xmls once
				if [[ $DDEL == 'no' ]]; then
					# Delete the xmls in the dates list
					find $BACKUPFOLDER -type f -name $DOMAIN-'*' | fgrep "$LIST" | xargs rm
					DDEL='yes'
				fi
			fi
		fi
	done

	echo "---- Backup done $DOMAIN ---- $(date +'%d-%m-%Y %H:%M:%S') ----" >> $LOG
done
echo "$SHCMD: Finished backups at $(date +'%d-%m-%Y %H:%M:%S')
====================" >> $LOG

exit $ERRORED

@lightninjay
Copy link

lightninjay commented Apr 23, 2020

@tuomotalvitie
A small portion of your script has a spelling error. "SNAPPEFIX" should be "SNAPPREFIX"

  # Now that the VM's disk image(s) have been successfully committed/pivoted to
  # back to the main disk image, remove the temporary snapshot image file(s)
  	for BACKUP in $BACKUPIMAGES; do
  		set -o noglob
  		if [[ $BACKUP == *${SNAPPEFIX}* ]]; then
  			set +o noglob
  			CMD="rm -f $BACKUP >> $LOG 2>&1"
  			echo " Deleting temporary image $BACKUP" >> $LOG
  			echo "Command: $CMD" >> $LOG
  			eval "$CMD"
  		fi
  		set +o noglob
  	done

I took some time modifying the script for my own use, since I run libvirt through the QEMU:///system hypervisor rather than the QEMU:///session one. Had to change the virsh commands to include "-c QEMU:///system" to redirect calls to that hypervisor.

@tuomotalvitie
Copy link

@lightninjay
Thank you, I updated the script to fix the spelling error.

( Ping @brentl99 )

@reinistihovs
Copy link

reinistihovs commented Jun 25, 2020

Hi!

I ran in an interesting problem,
I wasn't able to blockcommit the backup image in to the base image,
I got this error: failed to pivot job for disk vda error: block copy still active: disk 'vda' not ready for pivot yet
Next i found some similar cases online, and found out that its probably a libvirt bug, that is fixed in a newer release, i wasnt able to update libvirt, because this host had more live machines on it, that couldnt be stopped.

I tried to restart the blockcommit many times, but without result, same error every time.
In the end I solved the problem by turning off the VM and do the blockcommit offline by using qemu-img commit /path/to/uncommited/external/image/hostname.backup
I had to use qemu-img, because virsh blockcommit doesnt work, while the VM is offline :(
then i edited VM configuration, virsh edit VMNAME and changed the storage from .backup image to the original .qcow2 image.

Here are some handy commands to solve a similar case:

By running this command, you will see all the VM disks, and where their data is stored physically.

virsh domblklist VMNAME --details

Output:

Type      Device     Target     Source
------------------------------------------------
file       disk       vda        /./.VMNAME.backup

if you see (VMNAME.backup), that means that data is still stored on external image.

By running this command you can see the progress of blockcommit, in my case it was stuck at 100%, you have to run this for each disk.
virsh blockjob VMNAME vda --info

To manually start a blockcommit:
virsh blockcommit VMNAME vda --active --pivot

To abort the blockcommit:
virsh blockjob VMNAME vda --abort

Here is a nice article, that helped me:
https://kashyapc.fedorapeople.org/virt/lc-2012/snapshots-handout.html

@Marc176
Copy link

Marc176 commented Jun 25, 2020

Interesting - which version of libivirt had this issue? Which OS ?

@reinistihovs
Copy link

Hi!

Ubuntu 16.04.2 LTS
libvirt version: 1.3.1

the primary Image is 950GB large.

@tmuhr85111
Copy link

Hey folks,
thanks for this discussion! I did a bit of fine tuning on @tuomotalvitie 's script and all the input I got from here certainly got me
going forward. I do like the concept this is following and together with borg/borgmatic it meanwhile makes a pretty high performing backup solution for our KVM hypervisors:

Among other things

echo "$SHCMD: Count for BACKUPFOLDER ($BACKUPFOLDER) for image ($(basename -- $IMAGE)) too high ($COUNT), deleting historical files over $IMAGECNT..."

was not going to the logfile, but instead triggered a cron mail when run automatically - which actually is really a very small flaw only...

Working on CentOS I also spent some time in figuring out how to optimize email for spam-proof transmission of logfiles (using msmtp for this), make the subject lines of those a bit more redable and informative plus send the mail body in monospaced html for better readability.

Currently I'm running this in a test environment for the next couple of weeks and will post the findings / modifications if they should prove to be working stably over time.

Thanks, meanwhile for the effort you spent on this!

@Ryushin
Copy link

Ryushin commented Jul 14, 2020

Not sure if you saw above, but I spent a couple of weeks making a version that uses borg:
https://gist.github.com/cabal95/e36c06e716d3328b512b#gistcomment-3181333

@tmuhr85111
Copy link

@Ryushin:
thanks for the hint! Yes, I actually used your work, too, which made a great reference. Sorry for not having mentioned you and @brentl99 as well as all the other contributors in my previous post.

@Marc176
Copy link

Marc176 commented Jul 18, 2020

@Ryushin: i am also using your script since a few months now on production and works very well (Centos 8 with libvirt)

I have one question.. for performance reasons i did not use compression yet for the backups (to have a fast as possible backup) but since we keep one week for every VM, it takes a big of space...
Can i enable compression now afterwards? Or do i have to creat new borg repo?

@Ryushin
Copy link

Ryushin commented Jul 19, 2020

@Ryushin: i am also using your script since a few months now on production and works very well (Centos 8 with libvirt)

I have one question.. for performance reasons i did not use compression yet for the backups (to have a fast as possible backup) but since we keep one week for every VM, it takes a big of space...
Can i enable compression now afterwards? Or do i have to creat new borg repo?

Good to hear my script it working good for someone else. I spent more time on it than I thought I would. But I have it running in four other installations and it's working well.

I would google the Borg compression question. I would think you could enable compression though. From what I remember though, borg is single threaded. So the compression by borg might greatly add to your time to backup a job. I highly recommend using a file system that supports compression such as ZFS. That is what I do for all my backups. Let ZFS handle the compression. You can change the type and level of compression in ZFS. Plus you know your data won't be affected by bit rot. All my standalone servers run ZFS. The VMs will use ext4 for Linux VMs and NTFS for Windows and since they reside on top of ZFS, all is good.

@brentl99
Copy link

brentl99 commented Jul 19, 2020

Hello. You may be able to retrofit my compression method https://gist.github.com/cabal95/e36c06e716d3328b512b#gistcomment-3132817 back into the borg backup script. Although borg's differential backup strategy may not be effective when it doesn’t manage compression.

I used streamed compression specifically so that it would not impact backup duration. Most multiple cpu and/or multicore systems will be io bound during the backup not cpu bound and therefore able to compress the backup without affecting backup duration. Writing compressed data means you are writing fewer bytes to disk which may make the backup faster. Also, if you are later transferring this data across a network it is precompressed saving possible further compression cycles or network bandwidth.

Using a ZFS target, as suggested above, is a good solution assuming that your backups do not then need to be transferred across a network to another storage device.

@flexoid
Copy link

flexoid commented Jul 20, 2020

@Ryushin
Do you think it's worth publishing your script as a separate gist?
I would be glad to fork it to enhance it a bit (create a more robust snapshot with support of QEMU guest agent if possible).

@Ryushin
Copy link

Ryushin commented Jul 20, 2020

@Ryushin
Do you think it's worth publishing your script as a separate gist?
I would be glad to fork it to enhance it a bit (create a more robust snapshot with support of QEMU guest agent if possible).

I was thinking of doing that for a long time. There are links on the Internet that come here though. Even though this thread is getting very long and detailed. Almost all the scripts are derivatives of the original though. Let me think on it a couple of days. it would also be nice if cabal95 if he thinks these separate scripts should be in their own github/gist site. I also don't want to take away from his work.

@brentl99
Copy link

By all means, my git knowledge and skills are modest. It has got a lot more mileage than I ever expected. I was lazy in not forking it. Certainly happy to see all the interest. :)

@brentl99
Copy link

If it is forked it would be good to add a comment here that points to the fork as this thread comes up readily via Google search as Ryushin has elluded.

@cabal95
Copy link
Author

cabal95 commented Jul 20, 2020

I am fine either way. To be honest, I've been amazed at how much work people have done on what I originally posted. I no longer using the script myself as I am now using a NAS that has all that VM + backup features built-in. I'm okay with either choice, if you want to fork and create a new gist I can edit the original gist and add a link to the new one so that people don't have to dig down into the comments to find where the latest info is.

I also know QEMU, virsh etc. have come a long way since I originally wrote the script so it might make sense to "start clean" in that sense too.

@flexoid
Copy link

flexoid commented Jul 20, 2020

Yep, sorry for missing some credits, it is a really great example of collaboration.

I spent quite a lot of time trying to find simple and ready to use solution, as it's only for my home NAS and my job and professional skills are in a bit different area. And looks like there's not much in google except for this gist & thread.

What would be great is to create a repo with a set of scripts based on this approach, that can be more extendable and configurable. For example, VM preparing (snapshot creation etc.) and actual backup (cp, rsync, borg, whatever...) can be implemented as separate modules, with the ability to add alternative ones. Having it as a repo would also mean that we can go forward with proper contribution, which is impossible with gist.

I know how open source works. Unfortunately, my bash skills are weak so most likely I won't be that person. Posting it just in case anyone will think it's a good idea to spend some time on it.

@Marc176
Copy link

Marc176 commented Sep 8, 2020

@Ryushin: noticed that with the options:

How often to perform full check on borg repositories.

Day of the week to perform the full checK. Use full weekday name (date %A).

CHECK_DOW="Friday"

Which week in the month to perform the full check. Put any number(s) 1-5.

CHECK_WEEKS="12345"

a check will never be performed... any idea why?

@Marc176
Copy link

Marc176 commented Sep 8, 2020

Ok... Weekday name must match language ;)

@tuomotalvitie
Copy link

On related note,
SECS=$(printf "%.0f" $(/usr/bin/time -f %e sh -c "$CMD" 2>&1))
ended up creating a bit of manual work on a machine due to language differences:
printf: 891.16: invalid number
which will result in a division by 0 error.
There's probably a smart way to do this, but as a quick fix I'm going with
LC_NUMERIC="en_US.UTF-8" SECS=$(printf "%.0f" $(/usr/bin/time -f %e sh -c "$CMD" 2>&1))

@norik83
Copy link

norik83 commented Dec 8, 2020

#!/bin/bash

#################### config variables########################

#Uncomment to backup domains myVMa and myVMb
DOMAINS="myVMa myVMb"

#Alternatively list all VMs and back them all up
#DOMAINS=$(virsh list --all | tail -n +3 | awk '{print $2}')

BACKUPROOT=/mnt/remote/backup_vms/$(hostname)

LOG="$BACKUPROOT/logs/qemu-backup.$(date +%Y-%m-%d).log"
SNAPPREFIX=snaptemp-

# Path to XML files for stopped VM...
PATH_LIBVIRT_QEMU="/etc/libvirt/qemu"

#pause before blockcommit
pause_domain_bc=false
#pause after blockcommit fail. if pause_domain_bc is set to true, this option is skiped
pause_domain_after_bc_fail=true
#on fail blockcommit try blockjob
enable_bj_attemts=true
blockjob_retrycount=1
blockjob_delay="30s"
pause_while_bj_attemtps=true
#if blockjob attempts fail, try pause domain? (if pause_while_bj_attemtps is set to true, script ignore this option)
pause_domain_after_bj_fail=true

EMR="email@example.com"
#****************************
#you can create conf file with variables which overwrite those above
#fo example if name of yours script is backup_vms.sh,create file in location wher is your script and name him->  backup_vms.conf
#if you want other name and path, you need to find and change value for variable below (outsite config variables block) named-> confFile
#it is usefull if we work with many hypervisiors and make some changes in script
#****************************

#################### end of config variables#################

SHCMD="$(basename -- $0)"
SHCMD_path=$(dirname $(readlink -f $0))
confFile="$SHCMD_path/${SHCMD%.*}.conf"
if [ -f "$confFile" ]; then
    source "$confFile"
    echo "overwrite vars from config file"
else
    echo "no config file, but if you set variables on begining it is not necessary"
fi

[ ! -f $BACKUPROOT/logs ] && mkdir -p $BACKUPROOT/logs
DATE="$(date +%Y-%m-%d.%H%M%S)"

ERRORED=0
COMMITSNAPTAB[0]=false
BREAK=false






# extract the date coding in FILEFULLNAME (note: filename format must be YYYY-MM-DD)
dtmatch () { sed -n -e 's/.*\(2[0-1][0-9][0-9]-[0-1][0-9]-[0-3][0-9]\).*/\1/p'; }
#check if previous instance of script is not running, it's mandatory for blockcommit after interput script from some reason
if pidof -x "$SHCMD" -o $$ >/dev/null;then
	ERR="An another instance of this script is already running, please clear all the sessions of this script before starting a new one
	If you are sure to stop previous instance, try command: kill -9 \$(pidof -x \"$SHCMD\")"
	echo "$ERR" >> $LOG
	echo "$ERR"
	exit 1
fi
echo "$SHCMD: Starting backups on $(date +'%d-%m-%Y %H:%M:%S')"  >> $LOG
for DOMAIN in $DOMAINS; do
	BREAK=false
	#check if domain exists
	VMSRC=$(virsh list --all | grep [[:space:]]$DOMAIN[[:space:]] | awk '{print $3}')
	VMSRC=${#VMSRC}
	if [[ $VMSRC -eq 0 ]]; then
		ERR="Domain $DOMAIN on this hypervisior not exists"
		echo "$ERR" >> $LOG
		echo "$ERR"	
		continue
	fi
	
	echo "---- VM Backup start $DOMAIN ---- $(date +'%d-%m-%Y %H:%M:%S')"  >> $LOG

	VM_RUNNING=1
	
	VMSTATE=$(LC_ALL=en_EN virsh list --all | grep [[:space:]]$DOMAIN[[:space:]] | awk '{print $3}')
	echo "$VMSTATE"
	if [[ $VMSTATE == "running" || $VMSTATE == "paused" ]]; then
		MSG="-> VM $DOMAIN running or paused."
		echo "$MSG" >> $LOG
		echo "$MSG"		
	else

		MSG="-> VM $DOMAIN not running. No snapshot and blockcommit. Only copy."
		echo "$MSG" >> $LOG
		echo "$MSG"
    	VM_RUNNING=0
	fi

	BACKUPFOLDER=$BACKUPROOT/$DOMAIN
	[ ! -d $BACKUPFOLDER ] && mkdir -p $BACKUPFOLDER
	TARGETS=$(virsh domblklist $DOMAIN --details | grep disk | awk '{print $3}')
	IMAGES=$(virsh domblklist $DOMAIN --details | grep disk | awk '{print $4}')

	# check to make sure the VM is running on a standard image, not
	# a snapshot that may be from a backup that previously failed
	unset COMMITSNAPTAB[*]
	i=0
	COMMITSNAP=false
	for IMAGE in $IMAGES; do
		set -o noglob
		if [[ $IMAGE == *${SNAPPREFIX}* ]]; then
			set +o noglob
			if [[ $VM_RUNNING -eq 0 ]]; then
				ERR="$SHCMD: Error VM $DOMAIN is not running but is on a snapshot disk image: $IMAGE"
				echo $ERR >> $LOG
				echo "$ERR"
				echo "$ERR
Host:       $HOST
Disk Image: $IMAGE
Domain:     $DOMAIN
Command:    virsh domblklist $DOMAIN --details" | mail -s "$SHCMD snapshot Exception for $DOMAIN" $EMR
				BREAK=true
				ERRORED=$(($ERRORED+1))
				break
			else
				COMMITSNAPTAB[$i]=true
				COMMITSNAP=true
				MSG="VM $DOMAIN is running but is on snapshot image $IMAGE so mark him for blockcommit first"
				echo "$MSG" >> $LOG
				echo "$MSG"
			fi
		else
			# if not running on snapshot still must check if ther is no tmp file, if is presents, must be deleted before start new snapshot
			FILEPATH="${IMAGE%/*}"
			FILEFULLNAME=`basename "$IMAGE"`
			FILENAME="${FILEFULLNAME%.*}"
			#FILEEXT="${FILEFULLNAME#*.}
			TMPFILE="$FILEPATH/$FILENAME.$SNAPPREFIX$DOMAIN"
			if [ -f "$TMPFILE" ]; then
				CMD="rm -f $TMPFILE >> $LOG 2>&1"
				MSG=" Deleting temporary image $TMPFILE after making snapshot"
				echo "$MSG" >>$LOG
				echo "$MSG"
				echo "Command: $CMD" >> $LOG
				echo "Command: $CMD"
				eval "$CMD"
			fi
		fi
		set +o noglob
		i=$i+1
	done
	[ $BREAK == true ] && continue

	if [[ $VM_RUNNING -eq 1 ]]; then
		#if vm running on snap, first i need to merge
		#i=0
		if [[ $COMMITSNAP == true ]]; then
			i=0
			for TARGET in $TARGETS; do
				if [[ ${COMMITSNAPTAB[$i]} == true ]]; then
					if [[ $VMSTATE == "paused" ]]; then
						pause_enabled=true
					else
						pause_enabled=false
					fi
					if [[ $pause_domain_bc == true && $pause_enabled == false ]]; then
						MSG="Try to pause domain $DOMAIN first"
						echo "$MSG" >> $LOG
						echo "$MSG"
						CMDst="virsh suspend $DOMAIN"
						echo "Command: $CMDst" >> $LOG
						echo "Command: $CMDst"
						eval "$CMDst"
						if [ $? -ne 0 ]; then
							ERR="can't pause domain $DOMAIN, continue anyway"
							echo "$ERR" >> $LOG
							echo "$ERR"
						else
							pause_enabled=true
						fi
					fi
					MSG="$TARGET BLOCKCOMMIT previous snapshot "
					echo "$MSG" >> $LOG
					echo "$MSG"
					CMD="virsh blockcommit $DOMAIN $TARGET --active --pivot >> $LOG 2>&1"
					echo "Command: $CMD" >> $LOG
					echo "Command: $CMD"
					eval "$CMD"
					if [ $? -ne 0 ]; then
						ERR="Could not merge changes for disk $TARGET of $DOMAIN with blockcommit. VM may be in an invalid state."
						echo "$ERR" >> $LOG
						echo "$ERR"
						echo "$ERR
Host:    $HOST
Domain:  $DOMAIN
Command: $CMD" | mail -s "$SHCMD blockcommit Exception for $DOMAIN" $EMR
						if [[ $pause_domain_after_bc_fail == true && $pause_enabled == false ]]; then
							MSG="Try to pause domain $DOMAIN and retry bc"
							echo "$MSG" >> $LOG
							echo "$MSG"
							CMDst="virsh suspend $DOMAIN"
							echo "Command: $CMDst" >> $LOG
							echo "Command: $CMDst"
							eval "$CMDst"
							if [ $? -ne 0 ]; then
								ERR="can't pause domain $DOMAIN, continue anyway"
								echo "$ERR" >> $LOG
								echo "$ERR"
							else
								pause_enabled=true
							fi
							MSG="$TARGET BLOCKCOMMIT previous snapshot "
							echo "$MSG" >> $LOG
							echo "$MSG"
							CMD="virsh blockcommit $DOMAIN $TARGET --active --pivot >> $LOG 2>&1"
							echo "Command: $CMD" >> $LOG
							echo "Command: $CMD"
							eval "$CMD"
							if [ $? -ne 0 ]; then
								ERR="Could not merge changes for disk $TARGET of $DOMAIN (paused) with blockcommit. VM may be in an invalid state."
								echo "$ERR" >> $LOG
								echo "$ERR"
								echo "$ERR
Host:    $HOST
Domain:  $DOMAIN
Command: $CMD" | mail -s "$SHCMD blockcommit Exception for $DOMAIN" $EMR
							else
								#if bc succes need to disable bj, so set to false
								enable_bj_attemts=false
							fi
						fi	
						if [ $enable_bj_attemts == true ]; then
							if [[ $pause_while_bj_attemtps == true ]]; then
								MSG="Try to pause domain $DOMAIN first"
								echo "$MSG" >> $LOG
								echo "$MSG"
								CMDst="virsh suspend $DOMAIN"
								echo "Command: $CMDst" >> $LOG
								echo "Command: $CMDst"
								eval "$CMDst"
								if [ $? -ne 0 ]; then
									ERR="can't pause domain $DOMAIN, continue anyway"
									echo "$ERR" >> $LOG
									echo "$ERR"
								else
									pause_enabled=true
								fi
							else
								if [[ $pause_enabled == true && $VMSTATE == "running" ]]; then
									MSG="Resume domain $DOMAIN"
									echo "$MSG" >> $LOG
									echo "$MSG"
									CMDst="virsh resume $DOMAIN"
									echo "Command: $CMDst" >> $LOG
									echo "Command: $CMDst"
									eval "$CMDst"
									if [ $? -ne 0 ]; then
										ERR="can't resume domain $DOMAIN, continue anyway"
										echo "$ERR" >> $LOG
										echo "$ERR"
									else
										pause_enabled=false
									fi
								fi
							fi
						
							for (( j=1; j<=$blockjob_retrycount; j++ )) do
								CMD="virsh blockjob --domain $DOMAIN --pivot $TARGET"
								echo "Command: $CMD" >> $LOG
								echo "Command: $CMD"
								eval "$CMD"
								if [ $? -ne 0 ]; then
									ERR="virsh blockjob attempt no $j --domain $DOMAIN --pivot $TARGET failed"
									echo "$ERR" >> $LOG
									echo "$ERR"
									if [[ $j -eq $blockjob_retrycount ]]; then
										ERR="all blockjob attempts --domain $DOMAIN --pivot $TARGET failed."
										echo "$ERR" >> $LOG
										echo "$ERR"
										echo "$ERR
Host:    $HOST
Domain:  $DOMAIN
Command: $CMD" | mail -s "$SHCMD blockjob  Exception for $DOMAIN" $EMR
										if [[ $pause_domain_after_bj_fail == true && $pause_enabled == false && $VMSTATE == "running" ]]; then
											MSG="Try to pause domain $DOMAIN first"
											echo "$MSG" >> $LOG
											echo "$MSG"
											CMDst="virsh suspend $DOMAIN"
											echo "Command: $CMDst" >> $LOG
											echo "Command: $CMDst"
											eval "$CMDst"
											if [ $? -ne 0 ]; then
												ERR="can't pause domain $DOMAIN"
												echo "$ERR" >> $LOG
												echo "$ERR"
											else
												echo "Command: $CMD" >> $LOG
												echo "Command: $CMD"
												eval "$CMD"
													if [ $? -ne 0 ]; then
														ERR="can't blockjob --piovot $TARGET anyway, I'm giving up"
														echo "$ERR" >> $LOG
														echo "$ERR"
														echo "$ERR
Host:    $HOST
Domain:  $DOMAIN
Command: $CMD" | mail -s "$SHCMD blockjob Exception for $DOMAIN" $EMR
														echo "$ERR" >> $LOG
														echo "$ERR"
														BREAK=true
														ERRORORED=$(($ERRORED+1))
													fi
												MSG="Resume domain $DOMAIN"
												echo "$MSG" >> $LOG
												echo "$MSG"
												CMDst="virsh resume $DOMAIN"
												echo "Command: $CMDst" >> $LOG
												echo "Command: $CMDst"
												eval "$CMDst"
												if [ $? -ne 0 ]; then
													ERR="can't resume domain $DOMAIN, continue anyway"
													echo "$ERR" >> $LOG
													echo "$ERR"
												else
													pause_enabled=false
												fi
											fi
										else
											BREAK=true
											ERRORORED=$(($ERRORED+1))
											break
										fi
									fi
									sleep $blockjob_delay
								else
									MSG="virsh blockjob attempt no $j --domain $DOMAIN --pivot $TARGET success"
									echo "$MSG" >> $LOG
									echo "$MSG"
									break
								fi
							done
							[ $BREAK == true ] && break
						fi
					fi
					if [[ $pause_enabled == true && $VMSTATE == "running" ]]; then
						MSG="Resume domain $DOMAIN"
						echo "$MSG" >> $LOG
						echo "$MSG"
						CMDst="virsh resume $DOMAIN"
						echo "Command: $CMDst" >> $LOG
						echo "Command: $CMDst"
						eval "$CMDst"
						if [ $? -ne 0 ]; then
							ERR="can't resume domain $DOMAIN, continue anyway"
							echo "$ERR" >> $LOG
							echo "$ERR"
						else
							pause_enabled=false
						fi
					fi
				fi
				i=$i+1
			done
			[ $BREAK == true ] && continue
			#delete tmp images after blockcommit
			for IMAGE in $IMAGES; do
				set -o noglob
				if [[ $IMAGE == *${SNAPPREFIX}* ]]; then
					set +o noglob
					CMD="rm -f $IMAGE >> $LOG 2>&1"
					MSG=" Deleting temporary image $IMAGE"
					echo "$MSG" >> $LOG
					echo "$MSG"
					echo "Command: $CMD" >> $LOG
					echo "Command: $CMD"
					eval "$CMD"
				fi
				set +o noglob
			done
			#Reload images after blockcommit
			IMAGES=$(virsh domblklist $DOMAIN --details | grep disk | awk '{print $4}')
		fi
		DISKSPEC=""
		for TARGET in $TARGETS; do
			DISKSPEC="$DISKSPEC --diskspec $TARGET,snapshot=external"
		done

		# transfer the VM to snapshot disk image(s)
			CMD="virsh snapshot-create-as --domain $DOMAIN --name ${SNAPPREFIX}$DOMAIN --no-metadata --atomic --disk-only --quiesce $DISKSPEC >> $LOG 2>&1"
			echo "Command: $CMD" >> $LOG 2>&1
			echo "Command: $CMD"
			eval "$CMD"
			if [ $? -ne 0 ]; then
				MSG="Error create snapshot with --quiesce mode (it's require qemu-guest-agent), try without"
				echo "$MSG" >> $LOG
				echo "$MSG"
				CMD="virsh snapshot-create-as --domain $DOMAIN --name ${SNAPPREFIX}$DOMAIN --no-metadata --atomic --disk-only $DISKSPEC >> $LOG 2>&1"
				echo "Command: $CMD" >> $LOG 2>&1
				echo "Command: $CMD"
				eval "$CMD"
			fi
			if [ $? -ne 0 ]; then
				ERR="Failed to create snapshot for $DOMAIN"
				echo "$ERR" >> $LOG
				echo "$ERR"
				echo "$ERR
Host:    $HOST
Domain:  $DOMAIN
Command: $CMD" | mail -s "$SHCMD snapshot Exception for $DOMAIN" $EMR
				ERRORED=$(($ERRORED+1))
				continue
			fi
		#i=$i+1

	fi

	
	####################################
	#Rsync							   #
	####################################
	MSG="Rsync"
	echo "$MSG" >> $LOG
	echo "$MSG"
	for IMAGE in $IMAGES; do
		FILEFULLNAME=`basename "$IMAGE"`
		if test -f "$BACKUPFOLDER/$FILEFULLNAME"; then
			MSG="Backup exists, merging only changes to image $BACKUPFOLDER/$FILEFULLNAME"
			echo "$MSG" >> $LOG
			echo "$MSG"
			#CMD="rsync -apvhz --inplace --progress $IMAGE $BACKUPFOLDER/$FILEFULLNAME "
			CMD="rsync -apvhz --inplace --progress $IMAGE $BACKUPFOLDER/$FILEFULLNAME "
		else
			MSG="Backup does not exist, creating a full sparse copy of image $IMAGE"
			echo "$MSG" >> $LOG
			echo "$MSG"
			#CMD="rsync -apvhz --sparse --progress $IMAGE $BACKUPFOLDER/$FILEFULLNAME "
			CMD="rsync -apvhz --sparse --progress $IMAGE $BACKUPFOLDER/$FILEFULLNAME "
		fi
		echo "Command: $CMD" >> $LOG
		echo "Command: $CMD"
		eval "$CMD"
	done
	########################################
	# After sync file merge snapshots back #
	########################################

	if [[ $VM_RUNNING -eq 1 ]]; then
		# Update the VM's disk image(s) with any changes recorded in the snapshot
		# while the copy process was running.  In qemu lingo this is called a "pivot"
			BACKUPIMAGES=$(virsh domblklist $DOMAIN --details | grep disk | awk '{print $4}')
			for TARGET in $TARGETS; do
				#if [[ ${COMMITSNAPTAB[$i]} == true ]]; then
					#rsync can tak long time and state could changed so check once again 
					#if state shut there will simple error becouse blockcommi and blockjob can't be done on shutdown domain.
					VMSTATE=$(LC_ALL=en_EN virsh list --all | grep [[:space:]]$DOMAIN[[:space:]] | awk '{print $3}')
					if [[ $VMSTATE == "paused" ]]; then
						pause_enabled=true
					else
						pause_enabled=false
					fi
					if [[ $pause_domain_bc == true && $pause_enabled == false ]]; then
						MSG="Try to pause domain $DOMAIN first"
						echo "$MSG" >> $LOG
						echo "$MSG"
						CMDst="virsh suspend $DOMAIN"
						echo "Command: $CMDst" >> $LOG
						echo "Command: $CMDst"
						eval "$CMDst"
						if [ $? -ne 0 ]; then
							ERR="can't pause domain $DOMAIN, continue anyway"
							echo "$ERR" >> $LOG
							echo "$ERR"
						else
							pause_enabled=true
						fi
					fi
					MSG="$TARGET BLOCKCOMMIT previous snapshot "
					echo "$MSG" >> $LOG
					echo "$MSG"
					CMD="virsh blockcommit $DOMAIN $TARGET --active --pivot >> $LOG 2>&1"
					echo "Command: $CMD" >> $LOG
					echo "Command: $CMD"
					eval "$CMD"
					if [ $? -ne 0 ]; then
						ERR="Could not merge changes for disk $TARGET of $DOMAIN with blockcommit. VM may be in an invalid state."
						echo "$ERR" >> $LOG
						echo "$ERR"
						echo "$ERR
Host:    $HOST
Domain:  $DOMAIN
Command: $CMD" | mail -s "$SHCMD blockcommit Exception for $DOMAIN" $EMR
						if [[ $pause_domain_after_bc_fail == true && $pause_enabled == false ]]; then
							MSG="Try to pause domain $DOMAIN and retry bc"
							echo "$MSG" >> $LOG
							echo "$MSG"
							CMDst="virsh suspend $DOMAIN"
							echo "Command: $CMDst" >> $LOG
							echo "Command: $CMDst"
							eval "$CMDst"
							if [ $? -ne 0 ]; then
								ERR="can't pause domain $DOMAIN, continue anyway"
								echo "$ERR" >> $LOG
								echo "$ERR"
							else
								pause_enabled=true
							fi
							MSG="$TARGET BLOCKCOMMIT previous snapshot "
							echo "$MSG" >> $LOG
							echo "$MSG"
							CMD="virsh blockcommit $DOMAIN $TARGET --active --pivot >> $LOG 2>&1"
							echo "Command: $CMD" >> $LOG
							echo "Command: $CMD"
							eval "$CMD"
							if [ $? -ne 0 ]; then
								ERR="Could not merge changes for disk $TARGET of $DOMAIN (paused) with blockcommit. VM may be in an invalid state."
								echo "$ERR" >> $LOG
								echo "$ERR"
								echo "$ERR
Host:    $HOST
Domain:  $DOMAIN
Command: $CMD" | mail -s "$SHCMD blockcommit Exception for $DOMAIN" $EMR
							else
								#if bc succes need to disable bj, so set to false
								enable_bj_attemts=false
							fi
						fi	
						if [ $enable_bj_attemts == true ]; then
							if [[ $pause_while_bj_attemtps == true ]]; then
								MSG="Try to pause domain $DOMAIN first"
								echo "$MSG" >> $LOG
								echo "$MSG"
								CMDst="virsh suspend $DOMAIN"
								echo "Command: $CMDst" >> $LOG
								echo "Command: $CMDst"
								eval "$CMDst"
								if [ $? -ne 0 ]; then
									ERR="can't pause domain $DOMAIN, continue anyway"
									echo "$ERR" >> $LOG
									echo "$ERR"
								else
									pause_enabled=true
								fi
							else
								if [[ $pause_enabled == true && $VMSTATE == "running" ]]; then
									MSG="Resume domain $DOMAIN"
									echo "$MSG" >> $LOG
									echo "$MSG"
									CMDst="virsh resume $DOMAIN"
									echo "Command: $CMDst" >> $LOG
									echo "Command: $CMDst"
									eval "$CMDst"
									if [ $? -ne 0 ]; then
										ERR="can't resume domain $DOMAIN, continue anyway"
										echo "$ERR" >> $LOG
										echo "$ERR"
									else
										pause_enabled=false
									fi
								fi
							fi
						
							for (( j=1; j<=$blockjob_retrycount; j++ )) do
								CMD="virsh blockjob --domain $DOMAIN --pivot $TARGET"
								echo "Command: $CMD" >> $LOG
								echo "Command: $CMD"
								eval "$CMD"
								if [ $? -ne 0 ]; then
									ERR="virsh blockjob attempt no $j --domain $DOMAIN --pivot $TARGET failed"
									echo "$ERR" >> $LOG
									echo "$ERR"
									if [[ $j -eq $blockjob_retrycount ]]; then
										ERR="all blockjob attempt --domain $DOMAIN --pivot $TARGET failed."
										echo "$ERR" >> $LOG
										echo "$ERR"
										echo "$ERR
Host:    $HOST
Domain:  $DOMAIN
Command: $CMD" | mail -s "$SHCMD blockjob  Exception for $DOMAIN" $EMR
										if [[ $pause_domain_after_bj_fail == true && $pause_enabled == false && $VMSTATE == "running" ]]; then
											MSG="Try to pause domain $DOMAIN first"
											echo "$MSG" >> $LOG
											echo "$MSG"
											CMDst="virsh suspend $DOMAIN"
											echo "Command: $CMDst" >> $LOG
											echo "Command: $CMDst"
											eval "$CMDst"
											if [ $? -ne 0 ]; then
												ERR="can't pause domain $DOMAIN"
												echo "$ERR" >> $LOG
												echo "$ERR"
											else
												echo "Command: $CMD" >> $LOG
												echo "Command: $CMD"
												eval "$CMD"
													if [ $? -ne 0 ]; then
														ERR="can't blockjob --piovot $TARGET anyway, I'm giving up"
														echo "$ERR" >> $LOG
														echo "$ERR"
														echo "$ERR
Host:    $HOST
Domain:  $DOMAIN
Command: $CMD" | mail -s "$SHCMD blockjob Exception for $DOMAIN" $EMR
														echo "$ERR" >> $LOG
														echo "$ERR"
														BREAK=true
														ERRORORED=$(($ERRORED+1))
													fi
												MSG="Resume domain $DOMAIN"
												echo "$MSG" >> $LOG
												echo "$MSG"
												CMDst="virsh resume $DOMAIN"
												echo "Command: $CMDst" >> $LOG
												echo "Command: $CMDst"
												eval "$CMDst"
												if [ $? -ne 0 ]; then
													ERR="can't resume domain $DOMAIN, continue anyway"
													echo "$ERR" >> $LOG
													echo "$ERR"
												else
													pause_enabled=false
												fi
											fi
										else
											BREAK=true
											ERRORORED=$(($ERRORED+1))
											break
										fi
									fi
									sleep $blockjob_delay
								else
									MSG="virsh blockjob attempt no $j --domain $DOMAIN --pivot $TARGET success"
									echo "$MSG" >> $LOG
									echo "$MSG"
									break
								fi
							done
							[ $BREAK == true ] && break
						fi
					fi
					if [[ $pause_enabled == true && $VMSTATE == "running" ]]; then
						MSG="Resume domain $DOMAIN"
						echo "$MSG" >> $LOG
						echo "$MSG"
						CMDst="virsh resume $DOMAIN"
						echo "Command: $CMDst" >> $LOG
						echo "Command: $CMDst"
						eval "$CMDst"
						if [ $? -ne 0 ]; then
							ERR="can't resume domain $DOMAIN, continue anyway"
							echo "$ERR" >> $LOG
							echo "$ERR"
						else
							pause_enabled=false
						fi
					fi
				#fi
				i=$i+1
			done
		[ $BREAK == true ] && continue

		# Now that the VM's disk image(s) have been successfully committed/pivoted to
		# back to the main disk image, remove the temporary snapshot image file(s)
			for BACKUP in $BACKUPIMAGES; do
				set -o noglob
				if [[ $BACKUP == *${SNAPPREFIX}* ]]; then
					set +o noglob
					CMD="rm -f $BACKUP >> $LOG 2>&1"
					MSG=" Deleting temporary image $BACKUP"
					echo "$MSG" >> $LOG
					echo "$MSG"
					MSG="Command: $CMD"
					echo "$MSG" >> $LOG
					echo "$MSG"
					eval "$CMD"
				fi
				set +o noglob
			done

		# capture the VM's definition in use at the time the backup was done
			CMD="virsh dumpxml $DOMAIN > $BACKUPFOLDER/$DOMAIN.xml 2>> $LOG"
			MSG="Command: $CMD"
			echo "$MSG" >> $LOG
			echo "$MSG"
			eval "$CMD"
	else
		# copy the VM's definition
		CMD="cp $PATH_LIBVIRT_QEMU/$DOMAIN.xml $BACKUPFOLDER/$DOMAIN.xml 2>> $LOG"
		MSG="Command: $CMD"
		echo "$MSG" >> $LOG
		echo "$MSG"
		eval "$CMD"
	fi


	MSG="---- Backup done $DOMAIN ---- $(date +'%d-%m-%Y %H:%M:%S') ----"
	echo "$MSG" >> $LOG
	echo "$MSG"
done
MSG="$SHCMD: Finished backups at $(date +'%d-%m-%Y %H:%M:%S')
====================" >> $LOG
echo "$MSG" >> $LOG
echo "$MSG"

exit $ERRORED

@juliyvchirkov
Copy link

my 5 cents, live full and incremental backups of kvm guests

https://gist.github.com/juliyvchirkov/663eb6f5c18600a7414528beee6a7f3a

@tuomotalvitie
Copy link

I guess I need to do some converting to qcow3 to use the tool (at least to full potential). Thank you @juliyvchirkov

Meanwhile, "virsh blockcommit $DOMAIN $TARGET --active --pivot" in the script will most likely cause havoc if one uses base images.

Shouldn't be too difficult to fix to shorten the chain to the original length:
https://access.redhat.com/documentation/en-us/red_hat_enterprise_linux/6/html/virtualization_administration_guide/sub-sect-domain_commands-using_blockcommit_to_shorten_a_backing_chain

Seed image operations also take 0s to complete, so I dropped the division by time line.

@abbbi
Copy link

abbbi commented Nov 2, 2021

I guess I need to do some converting to qcow3 to use the tool (at least to full potential). Thank you @juliyvchirkov

yes, if you want to use the full feature set with incremental backups, you need to convert. Otherwise you can still
use backup mode copy, to do at least a full backup, heres the full documentation:

https://github.com/abbbi/virtnbdbackup#readme

@gokhan27
Copy link

Hi folks, how can we use vm-backup scripts with Bacula ? I need dir, fd and job files. Thanks.

@Ryushin
Copy link

Ryushin commented Nov 16, 2022

Hi folks, how can we use vm-backup scripts with Bacula ? I need dir, fd and job files. Thanks.

I use it with Bareos, which is a fork of Bacula. I just call it as a pre script:

RunScript {
    RunsWhen = Before
    RunsOnClient = yes
    Fail Job On Error = yes
    Command = "sudo /etc/bareos/scripts/virtual_machine_backups.bash"
  }

I use the borg backup script that I created that is posted above.

@gokhan27
Copy link

gokhan27 commented Nov 17, 2022

Thanks @Ryushin I looked at your script and it seems you backup your files with borg. But I want to learn how you arrange Fileset definition. You seem you don't need to backup any files in Fileset. Can we write any folders/directories in Fileset like below:
FileSet {
Name = "LocalhostFiles"
Include {
Options {
signature = MD5
compression=GZIP
}
File = /home
}
}
sorry I am new to Bacula/Bareos :(

@Ryushin
Copy link

Ryushin commented Nov 17, 2022

Thanks @Ryushin I looked at your script and it seems you backup your files with borg. But I want to learn how you arrange Fileset definition. I am new to Bacula/Bareos :(

Oh boy. Bareos/Bacula has a steep learning curve. It is Enterprise software with a learning curve to go with it. Threre is a lot of configuration needed. Explaining how to set up Bareos goes beyond the scope of this thread. I'll give you my tarball of /etc/bareos:
https://cloud.chrisdos.com/s/2egysaY5S9w2NxZ
Note, my bareos file layout is slightly customized from the new installs from Bareos. Bareos has changed their file structure slightly since when they first released and mine is customized a bit based on the original layout. Also, I back up to hard drives and use vchanger to simulate a virtual tape library.

Notes about this backup script. I exclude backing up the .qcow2 files in /var/lib/libvirt/images. I instead backup my borg repo. The vm-backup script is called by the Bareos Client config as a pre-job script so it is executed before Bareos starts backing up files.

@gokhan27
Copy link

Thanks @Ryushin this is really helpful for me especially for newbiees to bacula/bareos:) I will try bacup with bareos asap :)

@Ryushin
Copy link

Ryushin commented Nov 17, 2022

Thanks @Ryushin this is really helpful for me especially for newbiees to bacula/bareos:) I will try bacup with bareos asap :)

Don't thank me yet. You might find my setup a bit complicated. I have scripts in bareos/scripts that I call for backups. Along with sending backups and important files to other hosts. Take your time and go through it all. Use my stuff to create your setup. I also call sudo for a few scripts so you will need to create a bareos sudo user to call certain programs.

@MournfulRaven
Copy link

Hi everyone, did anyone adopt this to backup disks, which are also connected from pools?

@juliyvchirkov
Copy link

juliyvchirkov commented Nov 14, 2023

Hi guys

Below is my actual time-tested custom console backup solution to implement regular nighlty increment backups of live kvm machines w/o shutting the pool of virtual servers off in a completely automated non-interactive mode

The flow is mostly built upon great cli backup tool virtnbdbackup developed by @abbbi in Python with a number of extra related subroutines

Long story short

  1. Every night the routine produces 3 fresh backup copies, 2 local and 1 remote. Local backups are handled by dedicated 2Tb nvme
    drive, mounted under /backup point. Remote copies occupy remote rented storage box.
  2. Each vm is supplied with 2 virtual drives - vda with OS / data in qcow2 format and raw vdb utilized as swap partition. The last one is excluded from backups for obvious reasons.
  3. On 1st night of each month the fresh full backup is created. Then night by night it gets incremented from the live running machines during a month till the next one, when the cycle repeats.
  4. Primary local vm backups are stored under /backup/storage/kvm folder, the full path for each month is prepared automatically with this script by the scheme: /backup/storage/kvm/[guest-name]/MM.YYYY
  5. On 1st night of a month the step before full backups are created is the only step when machines are shutting down for a short term to maintain their virtual vda drives. Each drive gets defragmented and the wasted space are reclaimed to speed up the storage for the next month. For safety reasons drives are not defragmented in place, a new defragmented copy for each drive is created under the /tmp/ folder. And only If the image defragmentation tool and another image tool, utilized for post-check, both report zero status, castling is taking place.
  6. As soon as the process is fully automated, to insure vms against epic fail with a plot when despite zero statuses of tools defragmented drive appears broken and rejected by Qemu, the working origin of a drive in a state before the last defragmentaton is not wiped immediately, but instead is stored under /backup/storage/kvm/dev/block folder. A few days later, when admin feels completely assure a new copies of drives after defragmentation are working fine, she can drop these obsoleted extra backups by hands.
  7. The fresh defragmented copy of drive from /tmp/ takes places of the old one which has been moved to /backup/storage/kvm/dev/block, vm is started and the new full backup is created.
  8. Every night on the next step after backups are created or incremented, the whole tree under /backup/storage/kvm gets packed into squashfs entity and stored under the name /backup/storage/squashfs/kvm/kvm.sqsh to be the 2nd local copy in case the 1st one gets in a trouble.
  9. Some time later close to the dawn lftp tool (which are not covered by this script since it runs it's own separate plot aimed for remote) will locate this fresh night build of kvm.sqsh and transfer it to the remote storage box with other fresh backups in squashfs containers, thus providing the 3rd remote backup copy in case some evil thing exterminates both local along with live origin

The described flow is implemented night by night by the script below in a completely automated non-interactive mode with neither excess nor complain for over 2 last years, so I suppose for this moment it has been tested live long enough to share for the community with no worry

Please feel free to utilize this stuff for own purposes, to ask questions, to request related to this routine extra info (kinda mentioned above automated non-interactive lftp plot to populate backups to the remote), to discuss alternatives, or just to say thanks or hi

Glory to Ukraine! 🇺🇦
Juliy V. Chirkov, https://t.me/juliyvchirkov

#!/usr/bin/env bash
#
# Implements completely automated non-interactive backup routine for live kvm pool
#
# Designed, developed and tested in a long run under Ubuntu 20.04.4 and 22.04.3 at
# 2021-2023 by Juliy V. Chirkov <juliyvchirkov@gmail.com> https://juliyvchirkov.github.io/
# under the MIT licence https://juliyvchirkov.mit-license.org/2021-2023
#

[[ -z "${SHDEBUG}" ]] || set -vx

virtnbdbackupSocket="/var/tmp/virtnbdbackup.sock"
virtnbdbackupBackupType="stream"
virtnbdbackupBackupLevel="inc"
virtnbdbackupExcludeDrives="vdb"
virtnbdbackupWorkers="$(/usr/bin/nproc)"

vmBackupRoot="/backup/storage/kvm"
vmBackupImagesB4Defragment="${vmBackupRoot}/dev/block"
vmBackupSubstorageThisMonth="$(/usr/bin/date +%m.%Y)"

SQUASHFS="/backup/storage/squashfs/kvm"

defragment() {
    local kvmImg="$(/usr/bin/virsh domblklist "${1}" | /usr/bin/grep -Po "vda\s+\K\N+")"
    local tmpImg="/tmp/${kvmImg##*/}"
    local elapsed=0
    local restartLibvirtGuests
    local restartLibvirtd

    /usr/bin/virsh shutdown "${1}" --mode=agent &>/dev/null

    while /usr/bin/virsh list | /usr/bin/grep "${1}" -q; do
        /usr/bin/sleep 1

        elapsed=$(( elapsed + 1 ))

        [[ "${elapsed}" -eq 180 ]] && /usr/bin/virsh shutdown "${1}" &>/dev/null
    done

    if /usr/bin/virt-sparsify --compress "${kvmImg}" "${tmpImg}" && 
       /usr/bin/qemu-img check "${tmpImg}" &>/dev/null; then
        /usr/bin/virsh checkpoint-delete "${1}" --checkpointname virtnbdbackup.0 --children

        /usr/bin/systemctl -q is-active libvirt-guests && {
            restartLibvirtGuests=1
            /usr/bin/systemctl stop libvirt-guests
        }

        /usr/bin/systemctl -q is-active libvirtd && {
            restartLibvirtd=1
            /usr/bin/systemctl stop libvirtd
        }

        [[ -d "/var/lib/libvirt/qemu/checkpoint/${1}" ]] &&
            /usr/bin/rm -rf "/var/lib/libvirt/qemu/checkpoint/${1}"

        /usr/bin/chown root:kvm "${tmpImg}"

        /usr/bin/mv "${kvmImg}" "${vmBackupImagesB4Defragment}/${1}.${vmBackupSubstorageThisMonth}.vda"

        /usr/bin/mv "${tmpImg}" "${kvmImg}"

        [[ -z "${restartLibvirtd}" ]] || /usr/bin/systemctl start libvirtd
        [[ -z "${restartLibvirtGuests}" ]] || /usr/bin/systemctl start libvirt-guests
    fi

    /usr/bin/virsh start "${1}"
    /usr/bin/sleep 30s
}

[[ -d "${vmBackupRoot}" ]] || /usr/bin/mkdir -pm755 "${vmBackupRoot}"
[[ -d "${SQUASHFS}" ]] || /usr/bin/mkdir -m755 "${SQUASHFS}"

mapfile -t vmlist < <(/usr/bin/virsh list | /usr/bin/grep -oP "[-\d]+\s+\K[^\s]+" || :)

for vm in "${vmlist[@]}"; do
    [[ -d "${vmBackupRoot}/${vm}" ]] || /usr/bin/mkdir -m755 "${vmBackupRoot}/${vm}"

    [[ -d "${vmBackupRoot}/${vm}/${vmBackupSubstorageThisMonth}" ]] || {
        /usr/bin/mkdir -m755 "${vmBackupRoot}/${vm}/${vmBackupSubstorageThisMonth}"
        virtnbdbackupBackupLevel="full"
        defragment "${vm}"
    }

    /usr/bin/virtnbdbackup --domain "${vm}" \
                           --socketfile "${virtnbdbackupSocket}" \
                           --level "${virtnbdbackupBackupLevel}" \
                           --type "${virtnbdbackupBackupType}" \
                           --exclude "${virtnbdbackupExcludeDrives}" \
                           --worker "${virtnbdbackupWorkers}" \
                           --output "${vmBackupRoot}/${vm}/${vmBackupSubstorageThisMonth}" \
                           --compress
done

[[ -f "${SQUASHFS}/kvm.sqsh" ]] && /usr/bin/rm "${SQUASHFS}/kvm.sqsh"

/usr/bin/mksquashfs "${vmBackupRoot}/" "${SQUASHFS}/kvm.sqsh" \
            -noI -noD -noF -noX -comp xz -Xbcj x86 -keep-as-directory

@abbbi
Copy link

abbbi commented Nov 15, 2023

@juliyvchirkov thanks for sharing! Just a few notes from my side:

virtnbdbackupWorkers="$(/usr/bin/nproc)"

this doesnt make quite that much sense, virtnbdbackup uses one worker for each disk the virtual machine has attached, to speed up the backup process (process multiple disks in different threads).

So it is always limited to the amount of disks, not to the amounts of cpu's that your host system has. If you use the amount of cpus (which is most probably bigger than the amount of disks the virtual machine has) it will always default to the amount of disks, as the value is bigger ;)). So if you dont want to limit the amount of concurrent disks to process during backup, leave this value to default or limit it to a lower amount of workers if you want to throttle backup process.

virtnbdbackupSocket="/var/tmp/virtnbdbackup.sock"

by default virtnbdbackup uses a process related socket file, that allows to start multiple backups from different virtual
machines at the same time. If you ever want to enhance your script to backup multiple virtual machines at the same time (not sequentially as now) you dont want to set this option to a hardcoded socked.

@LordPinhead
Copy link

Is there a reason you all do not use blockcopy? I can backup/clone complete VM's without shutting them down. A snapshot relies on the last backup.
So a mix of a weekly Full Backup and snapshots would be a smart thing right? I have to read up on it, but when I make 6 snapshots and one full backup, I need basically just the fullbackup plus the latest snapshot to restore my machine.

@lightninjay
Copy link

Is there a reason you all do not use blockcopy? I can backup/clone complete VM's without shutting them down. A snapshot relies on the last backup. So a mix of a weekly Full Backup and snapshots would be a smart thing right? I have to read up on it, but when I make 6 snapshots and one full backup, I need basically just the fullbackup plus the latest snapshot to restore my machine.

The point of this script (at least the last time I checked it), is to pivot your live OS to a temporary snapshot file, while it is still booted and running. This frees up the ability to copy the OS virtual disk without risk of data corruption while it is running live. Once the backup of the VM disk is finished, the live snapshot is pivoted back onto the VM disk file, and resumes (or rather, maintains) live running of the OS.

@techlabhq
Copy link

techlabhq commented Mar 8, 2024

Example of how I automate this script in the following loop and it backs up all the VMs on the system. Note: I did have to make sure permissions were correct on the disk image files so the script would actually back them up, but besides that, it works great! Love it! Thanks for this.

I also rsync the /etc directory to the backup drive so I have a copy of the existing configs. I suppose I should copy the /etc directory into a date-stamped directory with the images, I just haven't written that yet.

#!/bin/bash
date
date > /etc/date.log

BACKUPDIR=/media/usbbackup
IMGDIR=$BACKUPDIR/image-backups
# However many backups you want to keep...
NUMBACKUPS=5

rsync -arp --delete /etc $BACKUPDIR

# Selects all domains (on or off) to begin the loop (i.e. VMs with a number: running, or VMs with a dash [-]: stopped.)
for VM in $( virsh list --all | awk '( $1 ~ /^[0-9]|-+$/ ) { print $2 }' )
do
        /usr/local/bin/vm-backup.sh $IMGDIR $VM $NUMBACKUPS
        #echo "Next"
done
date
exit

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment