Thought I would pass this on since it took me a while to figure out.
My setup consist of 2 PVE nodes directly attached to a drobo 1200, each has a dedicated NIC for storage, all interfaces are setup for jumbo frames, multipath is configured. The LUN on the Drobo is setup with LVM. Storage is shared, HA and migration works fine. There is also an instance of PBS running on dedicated bare metal with dual NICs in a bond.
The problem I was having was that backups were failing with error 61 whenever I would try to backup to the new PBS server. I had noted other odd behaviors that I've attributed to possible corruption of the disk images.
This is an unusual build so there is a chance that I may be the only person to face this specific problem.
In building VMs from the Red Hat family I noticed that the only way to make them bootable was to change the disk cache to writethrough. Debian machines seemed unaffected, boy was I wrong!
In brief if you are using a drobo SAN for you PVE cluster set your drives as folllows:
cache = writethrough
ssd = no
discard = no
Async IO =default (io_uring)
That's all
My setup consist of 2 PVE nodes directly attached to a drobo 1200, each has a dedicated NIC for storage, all interfaces are setup for jumbo frames, multipath is configured. The LUN on the Drobo is setup with LVM. Storage is shared, HA and migration works fine. There is also an instance of PBS running on dedicated bare metal with dual NICs in a bond.
The problem I was having was that backups were failing with error 61 whenever I would try to backup to the new PBS server. I had noted other odd behaviors that I've attributed to possible corruption of the disk images.
This is an unusual build so there is a chance that I may be the only person to face this specific problem.
In building VMs from the Red Hat family I noticed that the only way to make them bootable was to change the disk cache to writethrough. Debian machines seemed unaffected, boy was I wrong!
In brief if you are using a drobo SAN for you PVE cluster set your drives as folllows:
cache = writethrough
ssd = no
discard = no
Async IO =default (io_uring)
That's all