PBS advices on backup

[CedricKid]

Hello, I'm new to Proxmox Backup Server and I'd like to know if there are any good practices I should be aware of.

For installation, I have 3 nodes in a cluster. I installed Proxmox Backup Server in a virtual machine on Ceph storage, and I'm storing my backups on another Ceph pool. Does this seem secure, or are there potential issues? (This is a home lab, so I don't have unlimited resources.)

Regarding backups, I don't have a clear idea of how many backups I should do per day, week, or month for a typical web server, for instance. Same problem on retention policies. Do you have advices ?

Thank you in advance for your feedback.

 

[Dunuin]

Regarding backups, I don't have a clear idea of how many backups I should do per day, week, or month for a typical web server, for instance. Same problem on retention policies. Do you have advices ?

Thanks to the deduplication it doesn't matter that much as long as your data isn'T changing much. Something like 14 daily + 8 weeky + 12 monthly + 10 yearly might take less space than a few daily VZDump backups.

For installation, I have 3 nodes in a cluster. I installed Proxmox Backup Server in a virtual machine on Ceph storage, and I'm storing my backups on another Ceph pool. Does this seem secure, or are there potential issues? (This is a home lab, so I don't have unlimited resources.)

Not great for ransomware protection. See: https://pbs.proxmox.com/docs/storage.html#ransomware-protection-recovery
Best would be one or two dedicated PBS serverswith local SSD storage for fast IOPS and that those won't be affected if you lose your PVE hosts.

 

[CedricKid]

Thanks to the deduplication it doesn't matter that much as long as your data isn'T changing much. Something like 14 daily + 8 weeky + 12 monthly + 10 yearly might take less space than a few daily VZDump backups.

Ok, I'll try this. If I understand the principle of prune rules with these parameters, and I'm doing backups every 2 hours, I'll keep the latest backup from the last 14 days (keep daily). Once the 14 days are over, does it end, or does it increment, always keeping the last 14 days by deleting the oldest (j1 deleted for j15 for example) ?

Not great for ransomware protection. See: https://pbs.proxmox.com/docs/storage.html#ransomware-protection-recovery
Best would be one or two dedicated PBS serverswith local SSD storage for fast IOPS and that those won't be affected if you lose your PVE hosts.

I currently have 3 Proxmox VE servers: one with VMs, and the other two are still empty for now. Ideally, should I use both of them to set up Proxmox Backup Servers? If I only use one of them for PBS, does that mean if I lose that PVE, I lose everything?

 

[Dunuin]

If I understand the principle of prune rules with these parameters, and I'm doing backups every 2 hours, I'll keep the latest backup from the last 14 days (keep daily).

Correct. With backupy every 2 hours you probably alao want something like keep-hourly=24 in addition to the daily, weekly, ... Backups.

Once the 14 days are over, does it end, or does it increment, always keeping the last 14 days by deleting the oldest (j1 deleted for j15 for example) ?

It will keep the last backup of the last 14 days. The 15th day it will remove the oldest daily backup so there are 14 again. Same for weekly, monthly backups and so on where it will keep the last backup of a week or month.

Have a look at the prune simulator for a better understanding: https://pbs.proxmox.com/docs/prune-simulator/

Ideally, should I use both of them to set up Proxmox Backup Servers? If I only use one of them for PBS, does that mean if I lose that PVE, I lose everything?

Ideally you want PBS to be running bare metal and not virtualized on top of PVE or being part of the cluster.
Also good idea to have a second PBS offsite.

 

[CedricKid]

Okay, perfect, thanks for your answers. One last question: earlier, you mentioned deduplication. I'm having a bit of trouble understanding the concept. From what I've read, it seems like it deletes backups that are identical to save storage space, but that's not really it, is it?

 

[Dunuin]

Okay, perfect, thanks for your answers. One last question: earlier, you mentioned deduplication. I'm having a bit of trouble understanding the concept. From what I've read, it seems like it deletes backups that are identical to save storage space, but that's not really it, is it?

No, it chops everything you store in millions of chunk files of max 4MB. None if these chunk files will ever be stored twice. So if you have multiple backups and some data is identical (doesn't even have to be backups of the same VM), only the first backup will need to store this identical data and all other backups will just reference the existing chunk file.
So this saves lots if space but also means if a single 4MB chunk file corrupts, this mjght affect dozens of backups that then won't be restorable anymore. So you better have a reliable storage like a ZFS striped mirror that could heal itself or you should use multiple datastores or even multiple PBSs.

 

[CedricKid]

No, it chops everything you store in millions of chunk files of max 4MB. None if these chunk files will ever be stored twice. So if you have multiple backups and some data is identical (doesn't even have to be backups of the same VM), only the first backup will need to store this identical data and all other backups will just reference the existing chunk file.
So this saves lots if space but also means if a single 4MB chunk file corrupts, this mjght affect dozens of backups that then won't be restorable anymore. So you better have a reliable storage like a ZFS striped mirror that could heal itself or you should use multiple darastores or even multiple PBSs.

Okay, that's very interesting. So, if I understand correctly, if I back up my Zabbix VM without making any changes to it, the storage used will be equivalent to just a single backup?

 

[Dunuin]

Okay, that's very interesting. So, if I understand correctly, if I back up my Zabbix VM without making any changes to it, the storage used will be equivalent to just a single backup?

Correct.

 

[CedricKid]

Okay, interesting, but if I lose a single block, I'll lose all my backups. So what's the point of making multiple backups of the same VM if a single defective block can ruin all the backups for that VM?

 

[Dunuin]

Okay, interesting, but if I lose a single block, I'll lose all my backups. So what's the point of making multiple backups of the same VM if a single defective block can ruin all the backups for that VM?

You need to ensure that you won't lose a single block. Or have some redundancy by having multiple datastores or PBS servers so you got multiple copies of each block.
Deduplication is only datastore wide. Do 100 backups to a single datastore and you will only need to store 1 copy of each block instead of 100 copies. You could then sync those 100 backups to another local or remote datastore. Then you need to store everything twice but still better to have only 2 copies of each block instead of 100 copies. If then a block on one datastore corrupts, you coukd sync a healthy version back from the other datastore.