But I used the writeback cache in the Proxmox 8.4 as well I got bugged!
So, still, I can't explain why works on newest version of kernel and PVE.
Anyway... It's works.... That what's matter.
Thanks for your reply @Gilberto Ferreira .
The difference to my setup is your cache setting (writeback) which does not use O_DIRECT semantics and therefore does not trigger the bug.
just to validate this process, I create a 2 node cluster with Proxmox VE 8.4.19 and with the very same storage, created with Debian 13 and TargetCLI.
Now I got that error:
With Proxmox VE 9 and newer kernels, no error so far, at least here for...
So... I went ahead and create a quick and simple target iscsi with Debian and targetcli.
I've created the backstore, target and luns with sinble IP.
After that, I create the device in proxmox with:
iscsiadm -m discovery -t st -p 172.18.100.120...
Try to append this options in the grub:
pcie_aspm=off intel_iommu=off intel_idle.max_cstate=0 acpi_power_meter=off nomodeset
This options has save me for a lot of issue.
pcie_aspm=off
What it does: Disables Active State Power Management for...
I am using the very same kernel version, updated today, 07-06-2026.
Proxmox VE 9.2.4, 2-nodes.
Perhaps I don't have had the issue because the storage underlying is DRBD 9 dual-primary + pacemaker + dlm + ocfs2?
I really don't know!
@Gilberto Ferreira I just tested it with the current Proxmox kernel 7.0.14-3 and the problem still persists on my system.
Which kernel version do you have and what are your VM disk options if I may ask?
Turns out I need to fall back to the old postfix configuration with sasl_passwd support.
After set this in the main.cf postfix file:
alias_maps = hash:/etc/aliases
alias_database = hash:/etc/aliases
mydestination = $myhostname...
Hi there.
I have a strange situation here, in that PBS are not sent email after prune job has done.
Bellow some imagens showing up my configuration
Other notifications like backup done are working well.
Thanks
This problem with direct_io was solved in recent versions of kernel 7.
In fact, I have scenarios with ocfs2 and I didn't change Async IO from default to threads and VM is running perfect.
For now, I am using raw interface like vmbr1 with physical interface.
The caveat is that needs to use cable instead of using the IPs from an exist interface.
With RAW interface I was able to get full data rate.
Same problem here, with similar setup to OP. Full 10Gbps data rate between hosts when tested with iperf, but on VXLAN connected VMs only 3-4Gbps. Anybody know a fix? Thanks
Ok.... Here we go again...
First, the physical interfaces on each server:
pve01
pve02
Yes! Two nodes. This is just for test purpose.
Now the fabric itself, in the Datacenter -> SDN -> Fabric
In the fabric/sdn panel I got this:
and
and...
Sorry.... Thank you for your reply.
I couldn't stay much longer testing the SDN in this configuration.
As my deadline with the customer was very short, turns out that the traditional way to create vlan and stuff, works better.
So, in order to...
I spoke too soon.
Fabric doesn't work at all!
For time to time, the status change it to 'not ok'!
What's that suppose to mean?
I have this in the log
fabricd[891055]: [NBV6R-CM3PT] OpenFabric: Needed to resync LSPDB using CSNP!