Interfaces have changed after upgrade from 6.4 to 7.1

[Mister X]

Yesterday we made the upgrade from PVE 6.4 to 7.1. After that the cluster did'nt come up because the interfaces has been renamed. I don't really know what the names were before.

I saw that the 25 MBit interfaces for the cluster communication in /etc/network/interfaces after the upgrade were titled enp68s0f0 and enp68s0f1 and also the bond02 for the cluster used them.

But when I used

ip addr

the interfaces called enp68s0f0np0 and enp68s0f1np1. Is it a bug? I did'nt see in the documentation that the interfaces will be renamed.

So I changed the /etc/network/interfaces to

auto enp68s0f0np0
iface enp68s0f0np0 inet manual

auto enp68s0f1np1
iface enp68s0f1np1 inet manual

and in the same file the bond2 for the cluster to

auto bond2
iface bond2 inet static
        address 172.18.10.101/24
        bond-slaves enp68s0f0np0 enp68s0f1np1
        bond-miimon 100
        bond-mode broadcast

After rebooting the nodes the cluster was present again and was working fine.

Did I not read something or is it really a undiscovered problem?

 

[fabian]

likely the new kernel (or NIC driver module) started to provide a phys_port_name, and that got picked up by udev generating the device name.. which NICs are you using?

 

[mvs]

We had the same issue with Mellanox ConnectX-5 EN NICs (PCIe and OCP3.0 versions).
Intel X710 NICs (PCIe and FlexibleLOM versions) were not affected.

 

[Mister X]

The same on our nodes:
With the interfaces Ethernet controller: Mellanox Technologies MT27710 Family [ConnectX-4 Lx] with 25 GB this issue came up.

The following interfaces are not affected:
Ethernet controller: Broadcom Inc. and subsidiaries BCM57412 NetXtreme-E 10Gb RDMA Ethernet Controller (rev 01)
Ethernet controller: Intel Corporation I350 Gigabit Network Connection (rev 01)