I do not think that you are running into the same issue as you are running older versions. Please upgrade to the latest packages available and see if the issue persists.
[SOLVED] Super slow, timeout, and VM stuck while backing up, after updated to PVE 9.1.1 and PBS 4.0.20
- Thread starter Heracleos
- Start date
This does not sound like the same issue as reported in this thread. Please check the systemd journal on the PBS side. Further, make sure you have enough memory available on the PBS and check the io pressure.
I had the same issues with backups after upgrading to PBS 4.1. I reverted to the 6.11.11-2-pve kernel that was previously used and backups work fine again. I didn't want to upgrade to 6.17.4-2-pve as on my home lab my Proxmox box crashed after a few days running that kernel.
Now I'm a little concerned running the 6.17 on my production cluster. My home lab box is AMD 7945HX cpu, 64GB RAM, and a couple NVME drives. I also run PBS in a container that backs up to a usb4 flash drive. Running AMD EPYC 9 CPUs on our prod cluster with Intel NICs.
thanks all
Now I'm a little concerned running the 6.17 on my production cluster. My home lab box is AMD 7945HX cpu, 64GB RAM, and a couple NVME drives. I also run PBS in a container that backs up to a usb4 flash drive. Running AMD EPYC 9 CPUs on our prod cluster with Intel NICs.
thanks all
Last edited:
After a upgrade to the latest kernel we noticed a severe drop in performance when it comes to operations like cloning a VM or CT.
We use a seperate SSD based CEPH cluster on a 40GB network. Before the update speeds were normal and after the update a cloning procedure of a 600GB vm disk takes more than 5 hours. Nothing has been changed when it comes to the config of the storage cluster.
We use a seperate SSD based CEPH cluster on a 40GB network. Before the update speeds were normal and after the update a cloning procedure of a 600GB vm disk takes more than 5 hours. Nothing has been changed when it comes to the config of the storage cluster.
from which kernel version to which kernel version? were other components/packages upgraded at the same time? do you see other symptoms described in this thread (connections stalling with tiny receive windows)?
After updating PBS from 4.0.4 to the current version 4.1.0 (Enterprise repository) with kernel 6.17.2 on Monday, we suddenly experienced backup failures, even resulting in the destruction of VMs during the backup process. This cost us a great deal of work. After finding this thread, we temporarily switched PBS to the Community repository and performed the update to PBS 4.1.1 with kernel 6.17.4. The backup process now seems to be running stably again.
Therefore, two questions:
1. Why is a kernel released in the Enterprise repository and redistributed across multiple versions (PBS 4.0.20 - 4.1.0) when a bug has been known since the end of November 2025? We use the Enterprise repository precisely to prevent such issues!
2. When will the 6.17.4 kernel be available in the Enterprise repository?
Therefore, two questions:
1. Why is a kernel released in the Enterprise repository and redistributed across multiple versions (PBS 4.0.20 - 4.1.0) when a bug has been known since the end of November 2025? We use the Enterprise repository precisely to prevent such issues!
2. When will the 6.17.4 kernel be available in the Enterprise repository?
Last edited:
1. we did fix the issue we could reproduce quite fast
2. despite this long thread, only a (small!) subset of systems is affected, which is the reason why we haven't been able to reproduce the remaining issue at all so far, despite many hours spent by multiple team members on our end
3. rebooting into a working kernel as a workaround is possible, if your system/setup is affected
4. the 6.17.4-2 kernel will probably be moved to enterprise today or on Monday (we've waited for feedback by affected users, which only came in this week after everybody returned from their respective holidays). the package will be identical to the one on no-subscription (this is always true), so you can already install it from there if you want to have it now
2. despite this long thread, only a (small!) subset of systems is affected, which is the reason why we haven't been able to reproduce the remaining issue at all so far, despite many hours spent by multiple team members on our end
3. rebooting into a working kernel as a workaround is possible, if your system/setup is affected
4. the 6.17.4-2 kernel will probably be moved to enterprise today or on Monday (we've waited for feedback by affected users, which only came in this week after everybody returned from their respective holidays). the package will be identical to the one on no-subscription (this is always true), so you can already install it from there if you want to have it now
Many thanks for your reply. The good news: "the 6.17.4-2 kernel will probably be moved to enterprise today or on Monday", so we can change back to enterprise-repo. Especially since this kernel also runs on the PVEs, and I have little confidence in it anymore.
Let me elaborate a bit more regarding this issue. After the kernel upgrade of PBS the situation has improved. But I need to say it is still not as it used to be. I have noticed that some backups of a random VM takes way longer as it should be. For instance, a backup of a same VM takes one day 28 seconds while next day can be 2 hours or it even fails. If it fails there is a great chance that dirty-map is invalid so the next backup will take way longer.
To quantify a bit, before the kernel upgrade of PBS to the latest version some backups were running at kilo bits per second, now this slowness is "up" to few tens of M bits per second (10-40).
As I mention before, we have two clusters, one is still being populated - 9.1.1 (VMs imported from VMware), while the other one was just recently upgrade from PVE 8.4.14 to 9.1.1. So basically both clusters are now exactly same version with kernel 6.17.2-1.
So my assumption is that the current slowness is coming from PVE and also PVE kernel needs to be upgraded to 6.17.4-2. As we are using Enterprise repo I assume this kernel will be there soon(hopefully).
I am surprised that this regression has not been noticed before, as to me it looks quite serious one. I just need to say luckily this slowness is not being "noticed" by Ceph, otherwise it would have been way bigger problem.
To quantify a bit, before the kernel upgrade of PBS to the latest version some backups were running at kilo bits per second, now this slowness is "up" to few tens of M bits per second (10-40).
As I mention before, we have two clusters, one is still being populated - 9.1.1 (VMs imported from VMware), while the other one was just recently upgrade from PVE 8.4.14 to 9.1.1. So basically both clusters are now exactly same version with kernel 6.17.2-1.
So my assumption is that the current slowness is coming from PVE and also PVE kernel needs to be upgraded to 6.17.4-2. As we are using Enterprise repo I assume this kernel will be there soon(hopefully).
I am surprised that this regression has not been noticed before, as to me it looks quite serious one. I just need to say luckily this slowness is not being "noticed" by Ceph, otherwise it would have been way bigger problem.
fwiw, I had a stall on backups after upgrading and unpinning my PBS server but not upgrading PVE to the newest kernel a couple days ago. I then upgraded the PVE cluster to the new kernel and so far its working as it should.