What is your backup strategy?

[breakaway9000]

Hello,

Historically I have only ever used Hyper-V or VMWare in "Production" environments, both mature products with good support for third party backup tools that can connect to the host directly and make a backup of the VM. You can then restore the entire VM or do a more granular file-level recovery in case you need to only pull one file out for instance. One piece of software that can do this for example is Veeam Backup & Replication (not sure if any of you are familiar with it).

I have been using Proxmox on and off and now I'm approaching that point in time where I feel confident recommending it for new projects and new deployments.

My main question surrounds backups. How do you all back up your VMs? I need to protect approximately 200 VMs, with:

1. Daily backups going back 2 months
2. Weekly backup going back 6 months
3. Monthly backups going back 12 months

I know Proxmox has a builtin backup system but it simply compresses up the VM and dumps it to disk. This may work for keeping a few points in time for small VMs. But when you need a large amount of data backed up with multiple points in time and complex retention policies it becomes very inefficient with lots of duplicated data. So this means you'd need a dedupe capable system or other such technology at the storage layer? My VMs are a mix of Windows, Linux and BSD so a VM level backup running inside each VM is not going to do it.

Just wondering how the rest of you are getting around this? Anything obvious I've missed?

 

[Richard]

1. Daily backups going back 2 months
2. Weekly backup going back 6 months
3. Monthly backups going back 12 months

To go back "Daily" etc. such a long period is rather a case for snapshots than backups, i.e. you want to back in the history. In contrast the main idea of backup is to get back the data in case of a (hardware) defect.

Combining the latter witch snapshots would be an "Incremental backup", but this is not supported by Proxmox (even some users have their own solution for it, but it is not reliable and not recommended for use).

What you can do:

make regularly snapshots as well as regularly backups (without keeping to much older data).

So you have the recent data in case of a hardware defect, and can go back "Daily" etc. in all other cases.

 

[kev974]

To go back "Daily" etc. such a long period is rather a case for snapshots than backups, i.e. you want to back in the history. In contrast the main idea of backup is to get back the data in case of a (hardware) defect.

Combining the latter witch snapshots would be an "Incremental backup", but this is not supported by Proxmox (even some users have their own solution for it, but it is not reliable and not recommended for use).

What you can do:

make regularly snapshots as well as regularly backups (without keeping to much older data).

So you have the recent data in case of a hardware defect, and can go back "Daily" etc. in all other cases.

Hi Richard,

When you say make regularly snapshots you talk about zfs snaphsots or manual snaphsots you can do in GUI ?

Breakaway9000 talks about 200 vms with lots of data to backup,so i think one daylong is not suficient to backup all of them with vzdump.Do you have any suggestion for peaples who have lots of vms and lots of data to backup like him ?

Sincerely,

 

[FastLaneJB]

To go back "Daily" etc. such a long period is rather a case for snapshots than backups, i.e. you want to back in the history. In contrast the main idea of backup is to get back the data in case of a (hardware) defect.

Combining the latter witch snapshots would be an "Incremental backup", but this is not supported by Proxmox (even some users have their own solution for it, but it is not reliable and not recommended for use).

What you can do:

make regularly snapshots as well as regularly backups (without keeping to much older data).

So you have the recent data in case of a hardware defect, and can go back "Daily" etc. in all other cases.

Hi Richard,

I'd have to disagree that this is a sensible approach. If you're taking a storage snapshot on ZFS then you've not got the RAM, if you snapshot in Proxmox then you've got a performance hit the more snapshots you get.

Proxmox's backup with the qemu agent does support freezing the file system on Windows with VSS but it's not application aware so the log files for AD / Exchange / SQL, etc won't get truncated. So this also isn't ideal for Windows Server backups in most cases.

The only option really here is a backup solution that puts an agent onto each of the VM's your backing up. However this is going to have quite a RAM impact vs backup systems which are hypervisor aware however I don't believe you've much choice here in my opinion if you want to do this right.

 

[Richard]

When you say make regularly snapshots you talk about zfs snaphsots or manual snaphsots you can do in GUI ?

Snap shots like you usually make by GUI. But in that particular case the should be made by CLI running (e.g.for VM 101 making a snapshot called "snap123"

qm snapshot 101 snap123

Then this sequence can be put into a daily cron job (with adaption of VM IDs, snap shot IDs etc.)

Btw.: numerous snapshots don't have negative impact to performance.

Breakaway9000 talks about 200 vms with lots of data to backup,so i think one daylong is not suficient to backup all of them with vzdump.Do you have any suggestion for peaples who have lots of vms and lots of data to backup like him ?

It's a question of system design to have sufficient resources.