qmpbackup is designed to create live full and incremental backups of running qemu virtual machines via QMP protocol. It makes use of the dirty-bitmap feature introduced in later QEMU versions. It works with standalone QEMU processes.
If you want to backup QEMU virtual machines managed by libvirt, see this
project:
https://github.com/abbbi/virtnbdbackup
Table of Contents
qmpbackup makes use of qemu.qmp
python3 -m venv venv
source venv/bin/activate
pip3 install -r requirements.txt
python3 setup.py install
The virtual machine must be reachable via QMP protocol on a unix socket, usually this happens by starting the virtual machine via:
qemu-system-<arch> <options> -qmp unix:/path/to/socket,server,nowait
qmpbackup uses this socket to pass required commands to the virtual machine.
In order to create a full backup use the following command:
# remove already existent bitmaps from prior full backups:
qmpbackup --socket /path/to/socket cleanup --remove-bitmaps
# create a new full backup to an empty directory:
qmpbackup --socket /path/to/socket backup --level full --target /tmp/backup/
the command will create a new unique dirty bitmap and backup the virtual
machines disks to /tmp/backup/<disk-bus-id>/FULL-<timestamp>. It ensures
consistency by creating the bitmap and backup within one QMP transaction.
Multiple disks attached to the virtual machine are backed up concurrently.
During full and incremental backup, bitmaps will be created with persistent option flag. This means QEMU attempts to store them in the QCOW images, so
they are available between virtual machine shutdowns. The attached QCOW images
must be in qcow(v3) format, for this to work.
If you can't convert your QCOW images to newer formats, you still can use the
backup mode copy: it allows to execute a complete full backup but no further
incremental backups.
Second step is to change some data within your virtual machine and let qmpbackup create an incremental backup for you, this works by:
qmpbackup --socket /path/to/socket backup --level inc --target /tmp/backup/
The changed delta since your last full (or inc) backup will be dumped to
/tmp/backup/<disk-bus-id>/INC-<timestamp>, the dirty-bitmap is automatically
cleared after this and you can continue creating further incremental backups by
re-issuing the command likewise.
There is also the auto backup level which combines the full and inc
backup levels. If there's no existing bitmap for the VM, full will run. If a
bitmap exists, inc will be used.
By default a new full backup to an empty directory will create a new unique id for the bitmap that is used to start a new backup chain.
This way you can create multiple backup chains, each of them using an unique bitmap to track the changes.
The qmpbackup utility will not cleanup those bitmaps by default if you can
cleanup bitmaps that are not required via:
qmpbackup --socket /path/to/socket cleanup --remove-bitmaps
qmpbackup --socket /path/to/socket cleanup --remove-bitmaps --uuid <uuid>
Alternatively you can specify the uuid to be used for the bitmap names during the first full backup you create. This way the bitmaps will be re-used and must not be cleaned:
qmpbackup --socket /path/to/socket backup -l full -t /tmp/backup --uuid testme
qmpbackup --socket /path/to/socket backup -l inc -t /tmp/backup
Using the --monthly flag with the backup command, backups will be placed in
monthly folders in a YYYY-MM format. The above combined with the auto backup
level, backups will be created in monthly backup chains.
Executing the backup and the date being 2021-11, the following command:
qmpbackup --socket /path/to/socket backup --level auto --monthly --target /tmp/backup
will place backups in the following backup path: /tmp/backup/2021-11/
When the date changes to 2021-12 and qmpbackup is executed, backups will be
placed in /tmp/backup/2021-12/ and a new full backup will be created.
Disks can be excluded from the backup by using the --exclude option, the name must match the devices "node" name (use the info --show blockdev option to get a list of attached block devices considered for backup)
If only specific disks should be saved, use the --include option.
In case the virtual machine has an guest agent installed you can set the QEMU Guest Agent socket (--agent-socket) and request filesystem quiesce via --quiesce option:
qmpbackup --socket /path/to/socket --agent-socket /tmp/qga.sock backup --level full --target /tmp/ --quisce
Use the following options to QEMU to enable an guest agent socket:
-chardev socket,path=/tmp/qga.sock,server,nowait,id=qga0 \
-device virtio-serial \
-device "virtserialport,chardev=qga0,name=org.qemu.guest_agent.0" \
If you want to backup virtual machines without the virtual machine being in
fully operational state, it is sufficient to bring up the QEMU process in
prelaunch mode (The QEMU blocklayer is operational but no code is executed):
qemu-system-<arch> -S <options>
If the virtual machine uses UEFI, it usually has attached pflash devices
pointing to the UEFI firmware and variables files. These will be included in
the backup by default.
Restoring your data is a matter of rebasing the created qcow images by using standard tools such as qemu-img or qmprestore. There are three major features implemented within the restore command: rebase, merge and snapshotrebase.
The rebase and snapshotrebase commands will alter the directory
in-place: this means your backup files will be changed.
The merge functionality will merge the data into a separate, new qcow file
outside of your backup folder.
A image backup based on a backup folder containing the following backups:
/tmp/backup/ide0-hd0/
├── FULL-1706260639-disk1.qcow2
├── INC-1706260646-disk1.qcow2
└── INC-1706260647-disk1.qcow2
can be recovered the following ways:
A regular rebase will update the backing image for each backup file in-place:
qmprestore rebase --dir /tmp/backup/ide0-hd0
After rebase you will find an symlink /tmp/backup/image, which points to the
latest image to use with qemu or other tools.
Note: It makes sense to copy the existing backup directory to a temporary
folder before rebasing, if you do not want to alter your existing backups.
Using the --until option rollback to a specific incremental point in
time is possible:
qmprestore rebase --dir /tmp/backup/ide0-hd0 --until INC-1480542701
If you want to rebase and actually commit back the changes to the images use:
qmprestore commit --dir /tmp/backup/ide0-hd0
After rebase you will find the merged image file with all changes committed in the target folder.
Note: It makes sense to copy the existing backup directory to a temporary
folder before rebasing, if you do not want to alter your existing backups.
It is also possible to restore and rebase the backup files into a new target file image, without altering the original backup files:
qmprestore merge --dir /tmp/backup/ide0-hd0/ --targetfile /tmp/restore/disk1.qcow2
Using the snapshotrebase functionality it is possible to rebase/commit the
images back into an full backup, but additionally the rebase process will
create an internal snapshot for the qemu image, for each incremental backup
applied.
This way it is easily possible to switch between the backup states after rebasing.
qmprestore snapshotrebase --dir /tmp/backup/ide0-hd0/
[..]
qemu-img snapshot -l /tmp/backup/ide0-hd0/FULL-1706260639-disk1.qcow2
Snapshot list:
ID TAG VM SIZE DATE VM CLOCK ICOUNT
1 FULL-BACKUP 0 B 2024-10-21 12:50:45 00:00:00.000 0
2 2024-10-21-12:42:48 0 B 2024-10-22 09:23:39 00:00:00.000 0
3 2024-10-21-12:42:49 0 B 2024-10-22 09:23:39 00:00:00.000 0
The --compress option can be used to enable compression for target files
during the blockdev-backup operation. This can save quite some storage space on
the created target images, but may slow down the backup operation.
qmpbackup --socket /path/to/socket backup [..] --compress
qmpbackup --socket /path/to/socket info --show blockdev
Attached raw devices (format: raw) do not support incremental backup. The only way to create backups for these devices is to create a complete full or copy backup.
By default qmpbackup will ignore such devices, but you can use the
--include-raw option to create a backup for those devices too.
Of course, if you create an incremental backup for these devices, the complete image will be backed up.
To query existing bitmaps information use:
qmpbackup --socket /path/to/socket info --show bitmaps
In order to remove existing dirty-bitmaps use:
qmpbackup --socket /path/to/socket cleanup --remove-bitmaps
If you create a new backup chain (new full backup to an empty directory) you should cleanup old bitmaps before.
You can set an speed limit (bytes per second) for all backup operations to limit throughput:
qmpbackup --socket /path/to/socket backup [..] --speed-limit 2000000
-
Using the QMP protocol it cannot be used together with libvirt as libvirt exclusively uses the virtual machines monitor socket. See virtnbdbackup.
-
QEMUs
drive-backupfunction does currently not support dumping data as a stream, it also cannot work with fifo pipes as the blockdriver expects functions like ftruncate and fseek to work on the target file, so the backup target must be a directory.