On 4/28/25 10:14, vrms wrote:
hei jed,
that sounds slick. thx for the pointer.
out of interest ...
1. how do you handle snapshots for VMs?
I guess you have the choice between btrfs/zfs and qemu/virsh
snapshots on the images to your avail with a setup as yours
Since ZFS snapshotting is so fast, I rely on it. Occasionally I will do
overlays, but I've never relied on qcow2 snapshots. I don't run btrfs so
I shouldn't give advice on that.
2. is there anything else then the hd images inside those subvolumes?
Libvirtd stores the domain xml files in /var/lib/libvirtd somewhere, and
if you need to back up those settings you can copy them. I have not run
that on ZFS because that's bending the OS into a hard to recover
combination of file systems. Likewise, I avoid installing VMs into the
default libvirtd storage area /var/lib/libvirtd/images. I suggest
creating some scripts that can ease the domain pool storage creation to
help out with that. Below are some example bash functions I use:
function addpool() {
[ -z "$1" ] && echo "Whaz da poo name?" && return
local longname="$PWD/$1"
sudo virsh pool-define-as --name "$1" --type dir --target "$longname" --source-path
"$longname"
sudo virsh pool-start "$1"
sudo virsh pool-autostart "$1"
}
function NewVmOneDisk() {
set -o pipefail
# set -x
local L="/VMs"
local V="cholla6-vms${L}"
cd $L
local recent=`ls -1d l_45* | awk -F'[_-]' '{print $2}' | sort | tail -1`
local new_ser=$(( recent + 1 ))
#echo "New ser: $new_ser"
read -p "[$new_ser] One word name: " name
if [[ x$name == x ]]; then
echo "No name, bye."
return 1
fi
local vmname="${new_ser}-${name}"
local fqname="${V}/l_${vmname}"
local clustersz="32K"
local qi_opts=("create" "-f" "qcow2"
"-o" "cluster_size=$clustersz"
"-o" "preallocation=metadata"
"-o" "lazy_refcounts=on"
"-o" "extended_l2=on"
)
sudo zfs create $fqname
sudo zfs set recordsize=$clustersz $fqname
sudo zfs set compression=zstd-5 $fqname
sudo chown jreynolds: l_$vmname
cd l_$vmname || {
echo "Unable to enter directory l_$vmname"
return 1
}
sudo modprobe nbd
local nb="/dev/nbd2"
sudo modprobe nbd
qemu-img ${qi_opts[@]} ${vmname}-00-root.qcow2 60G
sudo qemu-nbd -c $nb ${vmname}-00-root.qcow2 || {
echo "Unable to mount ${vmname}-00-root.qcow2 -> $nb, bye"
set +x
return 1
}
sudo partprobe $nb
sudo parted -a optimal $nb -s \
mklabel msdos -- \
mkpart primary fat32 1M 2M \
mkpart primary fat32 2M 256M \
mkpart primary ext4 256M 1280M \
mkpart primary ext4 1280M -1s
sync
sudo partprobe $nb
# this changes type 83 (ext) to 8e (lvm)
sudo sfdisk --change-id $nb 2 ef
sudo sfdisk --change-id $nb 4 8e
sudo fdisk -l $nb
sleep 5
local vgname=`date +%H%M`
vgname="ctvg_$vgname"
sudo mkfs.vfat ${nb}p1
sudo mkfs.vfat -n EFI ${nb}p2
sudo mkfs.ext4 -L BOOT ${nb}p3
sudo pvcreate ${nb}p4
sudo vgcreate --pvmetadatacopies 2 -y $vgname ${nb}p4
sudo lvcreate -n lv_swap -L 2G $vgname
sudo lvcreate -n lv_root -L 20G $vgname
sudo lvcreate -n lv_home -L 2G $vgname
sudo mkswap -L SWAP /dev/mapper/${vgname}-lv_swap
sudo mkfs.ext4 -L ROOT /dev/mapper/${vgname}-lv_root
sudo mkfs.ext4 -L HOME /dev/mapper/${vgname}-lv_home
sync
sudo umount /dev/mapper/${vgname}-lv_swap
sudo umount /dev/mapper/${vgname}-lv_root
sudo umount /dev/mapper/${vgname}-lv_home
sudo vgchange -a n $vgname
sudo qemu-nbd -d $nb
cd ..
addpool l_$vmname
echo "...done"
set +x
} # ~NewVmOneDisk
--
Jed Reynolds -- Sr Software Developer and Sysadmin
Candela Technologies, Washington USA PST GMT-8
Please CC:supp...@candelatech.com on support topics.
