I'd like to propose a new cluster-wide configuration option allowing one to set the default MTU for (VirtIO) vNICs.
This would allow one to overwrite the default behavior changes made in Proxmox 9, which changed from "default MTU is 1500" to "default MTU is same as bridge".
As it is now, this causes significant issues for anyone that has a trunked bridge with its MTU set to 9000 to accommodate those interfaces which need it, such as a VM with an interface used for some storage solution (ceph, for example).
Consequently, the nodes that were upgraded from Proxmox 8 to Proxmox 9 now leave us without a clean solution to prevent the MTUs in our VMs from changing from the previous default of 1500 to the 9000 of our bridge, aside from going over each VM and setting the MTU on each individual NIC.
Indeed, in the use case I'm presenting, only a handful of NICs need an MTU different from 1500.
Having to do this for every single NIC is both error prone and adds a step to the configuration of every single of vNICs, which could be avoided by having a global setting.
Additionally setting the vNIC MTU on the fly is not without risk, any interfaces in for example Debian that do not have allow-hotplug set will go down, regardless of if the MTU inside the VM already matches or not.
Thanks for considering this!
This would allow one to overwrite the default behavior changes made in Proxmox 9, which changed from "default MTU is 1500" to "default MTU is same as bridge".
As it is now, this causes significant issues for anyone that has a trunked bridge with its MTU set to 9000 to accommodate those interfaces which need it, such as a VM with an interface used for some storage solution (ceph, for example).
Consequently, the nodes that were upgraded from Proxmox 8 to Proxmox 9 now leave us without a clean solution to prevent the MTUs in our VMs from changing from the previous default of 1500 to the 9000 of our bridge, aside from going over each VM and setting the MTU on each individual NIC.
Indeed, in the use case I'm presenting, only a handful of NICs need an MTU different from 1500.
Having to do this for every single NIC is both error prone and adds a step to the configuration of every single of vNICs, which could be avoided by having a global setting.
Additionally setting the vNIC MTU on the fly is not without risk, any interfaces in for example Debian that do not have allow-hotplug set will go down, regardless of if the MTU inside the VM already matches or not.
Thanks for considering this!
Last edited:
