solved: understanding PBS performance after migration from no-Enterprise to Enterprise

[Tom8]

Hi all,

I have migrated a VMware ESXi running on Dell Poweredge with Qnap ZFS with Hyperdataprotector to
ProxmoxVE 8.4 and PBS 4.1 to reuse current hardware.

The PBS is running as VM on Qnap Virtualisationstation and has storage connected as NFSv4 (I know not optimal, but that's the existing hardware).
On VE I have about 2,4TB storage used and backup to PBS. The daily backup require about 10 minutes.

Now I have upgraded the PBS to CommunitySubscription and before upgrading the VE to Standardsubcripton I take a backup with the same existing backupjobs.
This backupjob require is now running more than 2 hours.

What is different to yesterday:
1. 5x Debian Securityupdates got installed (between 5 and 50 packages updated
2. PBS got rebootet with subscription key
3. VMs are now in powered off

Just for understanding
* imho this is still running an incremental backup to yesterday?
* what are the reasons the backup require now much more time?

many greetings
Tom

 

[SteveITS]

If a VM is off the “dirty bitmap” is lost…there should be a line at the top of the log section to that effect.

All backups in PVE are full backups. The dirty bitmap just knows to skip unused changes. PBS deduplicates either way.

Is the PBS storage SSD?

 

[Tom8]

Hi @SteveITS,
ok, understood and you are right. On each start of backup is a line
scsi0: dirty-bitmap status: created new
And explains the long backup-runtime, however I don't understand, why the bitmap is not persistent.

Nevertheless with your hint I found the relation for it.
* https://forum.proxmox.com/threads/about-the-dirty-bitmaps.92754/
* https://qemu-project.gitlab.io/qemu/interop/bitmaps.html

no, the PBS storage are a ZFS mirror of two Seagate spindles.

thank you very much
Thomas

 

[Impact]

I don't understand, why the bitmap is not persistent.

Can you share your VM config and the backup log?

 

[Tom8]

Hi @Impact, the link described it.
The bitmap is only inmemory and get invalidated, when a VM get stopped to avoid invalid backups
Still in learning mode from VMware

 

[SteveITS]

If a VM will remain off, options to improve speed include unchecking the backup for the disk(s), or excluding the VM from the backup job.

 

[fabian]

https://bugzilla.proxmox.com/show_bug.cgi?id=3233 persistent bitmaps are tracked here, in case you want to get notified about updates