[SOLVED] Question about Backups

[Thomas Plant]

Hello,

do not know if the answer to my question is obvious.....

If I make on our 4 hosts cluster a backup job of all VMs, are the VMs backed up sequentially? Or for example is a backup started on every host, so there wil be 4 backups running in parallel? Latter would not be a good thing in our case.

Thanks,
Thomas

 

[t.lamprecht]

Hi,

If I make on our 4 hosts cluster a backup job of all VMs, are the VMs backed up sequentially? Or for example is a backup started on every host, so there wil be 4 backups running in parallel? Latter would not be a good thing in our case.

The backup lock is per host (node), so they will run in parallel on all nodes in a cluster with a virtual guest from a backup job.
If there are multiple guests to back up on a node, they are then done so sequentially.

 

[Thomas Plant]

Ok, thanks. So we will make a backup job for every node on different day.

Regards,
Thomas

 

[Thomas Plant]

Where can I write about a feature request?

'Backup only one VM at once over the entire cluster'

 

[t.lamprecht]

'Backup only one VM at once over the entire cluster'

What's your reason for that?

Would bandwidth limits help if backup target storage is an issue?

 

[Thomas Plant]

Problem is bandwidth to our backup server, it has only 2x1 Gbit bonded in alb mode. So 6 (wrongly wrote 4 hosts before) hosts at once pushing each on 1 gbit to PBS I think could be problematic.

 

[t.lamprecht]

Ok, thanks for clarifying your reasons.

You can file an enhancement request over at
https://bugzilla.proxmox.com/

 

[RolandK]

another reason why you want to limit that is, that nas storage like synology may be totally overwhelmed.

have a look - this is the load on our diskstation when backup runs on our 5-node cluster.

totally unusable and the backup throughput crawls....

diskstations have very low cpu power...

 

[t.lamprecht]

NAS storage for backup is often not ideally in general, but especially if it has to send the storage traffic over the same network as the backup traffic got sent, we recommend local storage for best experience.

If the NAS is already pre-existing I can understand the desire to use it as backup target, but for new setups a dedicated server with a local storage (e.g. using ZFS) can bring a lot of performance and also reliability (avoids having another failure point in the system) without high costs, the preconfigured proprietary nas boxes often have high premiums while still being under powered after all.

That said, did you already try setting up traffic control from the PBS side? It's integrated in the web interface since PBS 2.1 and can be used to slow down incoming api/backup traffic, which then also means less storage throughput produced.
https://pbs.proxmox.com/docs/network-management.html#traffic-control

A general job limit from the PVE cluster side is planned long term but not trivial, a nobody showed up to spearhead that feature until now. Note also that it's only helpful sometimes (lots of setups use a single PBS for multiple PVE cluster/single-host setups, that's why traffic control got added on PBS side).

 

[RolandK]

yeah, i'm aware that nas is not optimal, but one of my bosses loves synology, so i'm having a hard time to get it replaced

regarding traffic throttling, i'm aware of this options, thanks - but whereas i have no overload problem with pbs i don't need to use it.

my problem is vzdump full backups to nas share, as there is no limit the number of concurrent backup tasks in a cluster...

>A general job limit from the PVE cluster side is planned long term but not trivial,
thanks for letting us know

 

[hepo]

Does anyone knows if feature/enhancement request was actually made?

We are running dedicated PBS with HDDs in mirrored vdevs (ZFS), 7 nodes running backups in the same time is choking the server/disks.
Slow PBS server causes issues on the VM's while the backup is running.
To mitigate this at present we have 7 backup jobs, one for each node.

 

[Neobin]

Does anyone knows if feature/enhancement request was actually made?

Looks so:
https://bugzilla.proxmox.com/show_bug.cgi?id=3086