Skip to content
Hands-on IT IT Drills

Day 3: Images and Datastores

Goal: Know where disks live, create your own data disk, and experience what “persistent” means in OpenNebula. · Duration: approx. 60 min · Prerequisite: Day 2

Briefing

  • An image is a disk that OpenNebula manages: an operating system, an empty data disk, a CD. It lives in an image datastore.
  • When a VM starts, the Transfer Manager copies its disks to the system datastore. That is where the VM works, not on the image itself.
  • Whether changes flow back into the image is decided by a switch:
Image At start When the VM is deleted
non-persistent (default) copy or qcow2 overlay in the system datastore copy is discarded
persistent the VM works directly with the image changes remain in the image
  • Image types: OS (bootable), DATABLOCK (empty or filled data disk), CDROM (read-only, such as ISO files).
 Image-Datastore 1 (default)          System-Datastore 0
 ├─ Alpine Linux 3.20 (OS)   ──Kopie──►  /var/lib/one/datastores/0/<VM-ID>/disk.0
 └─ daten (DATABLOCK, persistent) ────►  Verweis, die VM schreibt ins Image

Drill 1: The datastores

onedatastore list

Expected:

  ID NAME         SIZE AVA CLUSTERS IMAGES TYPE DS      TM      STAT
   2 files         28G 76% 0             0 fil  fs      local   on
   1 default       28G 76% 0             3 img  fs      qcow2   on
   0 system        28G 76% 0             0 sys  -       qcow2   on
Column Meaning
TYPE sys for running VMs, img for images, fil for files such as kernels or context scripts
DS how images get in (fs = file system)
TM how disks get to the hosts (qcow2 = overlays in the shared file system)

All three live on the same disk here, under /var/lib/one/datastores. Images are plain files with hash names there:

sudo ls -la /var/lib/one/datastores/1/

Expected (shortened):

-rw-r--r-- 1 oneadmin oneadmin 1398682112 Nov 12 18:47 3004b2effa884b510c08267193825668
-rw-r--r-- 1 oneadmin oneadmin   89182208 Nov 12 18:45 59fed322469dcb8c2b10e5a346059c7f

The 1.4 GB file is the image of the OneKS appliance, the 89 MB one is Alpine. Which file belongs to which image is revealed by oneimage show <name> | grep SOURCE.

Save your progress? Sign up for free

Drill 2: A data disk

oneimage create --name daten --type DATABLOCK --size 512 --persistent --datastore default
oneimage list

Expected:

ID: 2
  ID USER     GROUP    NAME            DATASTORE     SIZE TYPE PER STAT RVMS
   2 oneadmin oneadmin daten           default       512M DB   Yes rdy     0
   1 oneadmin oneadmin Alpine Linux 3. default       512M OS    No rdy     0
   0 oneadmin oneadmin OneKS Appliance default         4G OS    No rdy     0

PER Yes means persistent, STAT rdy means ready. RVMS counts the VMs currently using the image.

A VM with an operating system and a data disk. --disk replaces the template’s disks, which is why Alpine is in the list as well:

onetemplate instantiate 1 --name speicher --disk "Alpine Linux 3.20,daten"

After a minute:

onevm show speicher | grep -A3 '^VM DISKS'

Expected:

VM DISKS
 ID DATASTORE  TARGET IMAGE                               SIZE      TYPE SAVE
  0 default    vda    Alpine Linux 3.20                   -/512M    file   NO
  1 default    sdb    daten                               -/512M    file  YES

SAVE YES is the persistent disk. TARGET sdb is the name OpenNebula gives the disk in libvirt. Inside the VM, it has a different name:

IP=$(onevm show speicher | grep -o 'ETH0_IP="[0-9.]*' | cut -d'"' -f2)
ssh root@$IP cat /proc/partitions

Expected (shortened):

major minor  #blocks  name

  11        0        370 sr0
   8        0     524288 sda
 253        0     524288 vda
  • vda is the operating system (virtio disk).
  • sda is the data disk. Linux numbers SCSI disks inside the VM on its own, and this is the first one.
  • sr0 is a small CD drive with the context, the VM’s startup data (Day 4).

Remember: Inside the VM, never rely on the TARGET name from OpenNebula. If you need fixed names, set DEV_PREFIX=vd on the image. Then the data disk is also called vdb inside the VM.

Save your progress? Sign up for free

Drill 3: Persistent or not

Format, mount, write two files, one on the data disk and one on the operating system:

ssh root@$IP "apk add -q e2fsprogs; mkfs.ext4 -q /dev/sda; mkdir -p /daten; \
  mount /dev/sda /daten; echo Hallo OpenNebula > /daten/notiz.txt; \
  echo Weg > /root/fluechtig.txt; sync; df -h /daten | tail -1"

Expected:

/dev/sda        488M  156K  452M   1% /daten

Now the VM is deleted and a new one is created with the same images:

onevm terminate speicher
onetemplate instantiate 1 --name speicher2 --disk "Alpine Linux 3.20,daten"

After a minute:

IP=$(onevm show speicher2 | grep -o 'ETH0_IP="[0-9.]*' | cut -d'"' -f2)
ssh root@$IP \
  "mkdir -p /daten; mount /dev/sda /daten; cat /daten/notiz.txt; ls /root/fluechtig.txt"

Expected (shortened):

Hallo OpenNebula
ls: cannot access '/root/fluechtig.txt': No such file or directory

The note survived, the file on the operating system did not. e2fsprogs is gone again too: the Alpine image is not persistent, and every VM starts from the original state.

Save your progress? Sign up for free

Checkpoint

oneimage list | grep daten

Expected: daten with PER Yes, STAT used, RVMS 1.

onevm show speicher2 | grep -c 'daten'

Expected: 1 or more; the disk is attached to the VM.

Challenge

A second VM should get the same data disk while speicher2 is running. What happens?

Solution

onetemplate instantiate 1 --name speicher3 --disk "Alpine Linux 3.20,daten"

Expected:

[one.template.instantiate] Error allocating a new virtual machine template. DISK 1: Cannot acquire image 2, it is persistent and already in use

OpenNebula refuses even to create it. A persistent image always belongs to exactly one VM. Two VMs writing to the same ext4 at the same time would destroy it. Shared storage for multiple VMs needs a cluster file system or a network service such as NFS.

Clean up:

onevm terminate speicher2

The image daten remains. You no longer need it, but you never delete a persistent image with data by accident. Once you are sure:

oneimage delete daten

Save your progress? Sign up for free

Enterprise Note

  • Shared storage: In production, system and image datastores live on Ceph, NFS, or a SAN. Then disks do not have to be copied, and a VM can move to another host without interruption (live migration).
  • qcow2 chains: Non-persistent VMs start as a thin overlay on top of the image. Many VMs from one image save space this way, but the image must not change as long as VMs build on it.
  • Backups: Persistent images are where data lives. OpenNebula 7 comes with its own backup datastores (Day 7). Images without backups are off-limits in production.