Sauvegarder une instance

From Alliance Doc
Revision as of 17:22, 30 July 2018 by Diane27 (talk | contribs) (Created page with "==Méthodes de sauvegarde OpenStack== OpenStack offre deux options pour le stockage : * stockage dans un volume, avec triple réplication; cette option protège les données e...")
Jump to navigation Jump to search
Other languages:

Page enfant de Service infonuagique de Calcul Canada

Il existe plusieurs stratégies pour faire une copie de sauvegarde d'une instance et pour rétablir la situation en cas de problème; le choix de la stratégie à adopter dépend de vos exigences et de votre cas particulier. Il est fortement recommandé de créer des copies de sauvegarde à l’extérieur du nuage. Une des règles fréquemment appliquées en matière de sauvegarde est celle du 3-2-1 qui veut que trois copies de vos données soient enregistrées sur au moins deux types de médias et qu’une de ces copies se trouve hors du site. Nous discutons ici de quelques méthodes usuelles pour sauvegarder votre instance et en préserver l’état et nous présentons l’exemple d’une combinaison de certaines ces méthodes qui représente une stratégie de sauvegarde complète.

Sauvegarde de fichiers

Plusieurs stratégies employées avec les ordinateurs physiques s’appliquent aussi aux instances virtuelles; par exemple, rsync et duplicity sont des outils qui peuvent sauvegarder les données de votre instance dans un endroit à distance.

Configuration automatique

Des outils d’approvisionnement comme ansible, puppet, chef et saltstack peuvent être utilisés pour automatiser la configuration des logiciels et du système d’exploitation. Avec les fichiers de spécification appropriés pour chacun de ces outils, il est très facile de recréer une instance. Les fichiers de spécification peuvent être gérés par une application de gestion de versions comme git. Les outils d’approvisionnement et d’orchestration (par exemple heat et terraform) peuvent être utilisés ensemble pour automatiser le processus complet de la création d’une instance et de la configuration des logiciels; voyez Automating VM creation; les données qui ne seraient pas alors générées ou créées devront être sauvegardées en utilisant une méthode mentionnée à la section Sauvegarde de fichiers.

Méthodes de sauvegarde OpenStack

OpenStack offre deux options pour le stockage :

  • stockage dans un volume, avec triple réplication; cette option protège les données en cas de problème de matériel, mais non en cas de suppression involontaire ou malicieuse;
  • stockage éphémère sur un nœud local; cette option protège aussi en cas de problème de matériel, mais ne devrait pas être utilisée avec des données critiques; on l’utilise surtout de façon temporaire.

OpenStack provides tools to create disk images and snapshots of your virtual machines. The two main VM flavors - persistent (p) and compute (c) - have different behaviors and creating disk images or snapshots should be considered differently for each.

Persistent VMs

Persistent VMs are designed to boot from volumes (see booting from a volume) thus creating a copy of the volume(s) the VM has attached to it will produce a backup. However this would not preserve things like VM flavor, public IP, and security rules. The best way to create a copy of a volume for backup purposes is to create an image from that volume. An image can be downloaded and reused to create multiple new VMs, can be accessed by VirtualBox on your desktop or laptop, and uploaded to other clouds.

To be able to create an image from a volume, that volume must be detached from the VM. In addition, if the volume is the root volume of the VM it cannot be detached unless the VM is deleted. If you are sure that when you created the VM you did not check the box "Delete Volume on Instance Delete", then you can delete your VM knowing you will not lose any data. However, if you are unsure whether or not you checked this box, OpenStack unfortunately doesn't tell you if this box was checked when you created a VM. One trick which may be useful for getting around this deficiency is to create a snapshot of the volume provided you have a storage quota which allows it, since snapshots count towards your storage quota. As volumes cannot be deleted if there is a volume snapshot created from them, when you delete the VM the volume will not be deleted even if you checked the box.

At this point all the volumes you wish to create images of should be in the "Available" state. To create an image from a volume, select 'Upload to Image' from the drop down menu for the volume. Select the 'QCOW2' disk format and give your image a name. There are several formats for disk images but QCOW2 works well with OpenStack and typically does not take up as much space as "Raw" images. Other formats "vmdk" and "vdi" can be useful when working with other virtualization tools such as VirtualBox.

Once you have created images of all the volumes you wish to backup you can then re-create your VM booting from the original VM's root volume and attaching any additional volumes you may have had attached to the original VM.

What about volume snapshots?

Another alternative might be to create a snapshot of a volume, which will save the state of a volume at the time the snapshot was created, however volume snapshots depend on the original volume remaining intact and as such are not ideal for backups. It is also not possible to download volume snapshots as they depend on the original volume. However, they do allow you to create a new volume from a snapshot of a volume at a given point in time. For example, if there were file changes since the last volume snapshot you would like to revert. Or if there were file changes specific to that VM which should not be included in other VMs.

What about VM snapshots?

Unfortunately, OpenStack uses the word "snapshot" to mean two different things. There are volume snapshots, as discussed above, and snapshots of VMs. Snapshots of VMs behave in different ways depending on the flavor of your VM. If you have a persistent VM and create a snapshot of the VM, OpenStack creates a nearly empty image, which contains pointers to volume snapshots. These pointers point to volume snapshot(s) of the persistent VM's boot volume and any attached volumes which were created as part of creating a snapshot of the VM. You can then create a new VM (boot from image (creates a new volume)) which will create new volumes from the snapshots of the volumes taken previously, boot a new VM from the root volume and attach any other duplicated volumes.

Compute VMs

As with creating backups of persistent VMs, the main goal is to create an image of at least the root drive, and perhaps also other attached volumes if needed. However, differences with compute flavor VMs change the process of creating that image. Compute VMs are not designed to boot from volumes accessed over the network as persistent VMs are, instead they are meant to be booted from disk images which reside locally on the computer which is actually running your VM. This means there is no volume which you can click on in the OpenStack dashboard to create an image of your root disk. Instead you can do this by clicking "Create Snapshot" on your VM's drop-down menu on the "Instances" tab. As with creating a snapshot with a persistent VM this creates an image; however, in this case the image isn't nearly as empty (i.e. containing only pointers to volume snapshots) since the image instead contains a copy of the VM's root drive.

Compute VMs also come with an extra data drive mounted at /mnt and the data on this drive is not captured in the image created of a compute VM. Other arrangements must be made to save this data, such as copying it off the disk before the VM is terminated.

An example backup strategy

Very large disk images (larger than 10-20 GB) can become difficult to manage with relatively long upload times and long VM creation times. A good basic strategy might be to separate large data sets from your operating system and software stack. The operating system and software stack can be backed up either using a disk image or recreated using some provisioning software (see setup automation). The data sets can then be backed up with using normal file backup methods to a remote location. If you are using any database software such as MySQL or PostgreSQL you will want to create dumps of the databases to include in your backups. Finally, and most importantly, test restoring from your backup. If you can't restore from your backup it isn't very useful.

Voir aussi