Guest VM network not working when host is using balance-rr bonding and 2 active interfaces

[Fatz]

I will try and find time to dabble with the above setting on .102, but i'm in quite a rush to get this machine in to use for some tasks.

I can say though, that the 6.2 kernel (6.2.6-1-pve) seems to be working for me without issue, details are in the bugzilla thread.

 

[john12]

Not exactly solved.
"bridge_ageing 0" solves the issue of DHCP not working and/or network suddenly freezing on the VMs, but it turns vmbr0 into a hub (packets in the same VLAN are visible between machines on the same physical host)
"bridge-disable-mac-learning 1" brings us back to square 1 (DHCP doesn't work).

Details: https://forum.proxmox.com/threads/using-2-nics-in-bonding-mode.124472/post-542990

 

[yonoss]

I am seeing the same problem in proxmox 8.0.3:
m:~# pveversion
pve-manager/8.0.3/bbf3993334bfa916 (running kernel: 6.2.16-3-pve)

 

[kennychennetman]

I got the exact same problem with v8.1.10(kernel 6.5.13-3pve)

note:
guest HDCP doesn't work with bonding in RR mode, but works fine in LACP

 

[SteveCliff]

Just trying this out now and getting exactly the same issue

Proxmox v8.2.4 (kernel 6.8.8-3-pve)

Purely by chance I am only running 1 x VM each on these nodes (it's a test K3s cluster) so I don't think turning vmbr0 in to a hub matters too much .... but would definitely like to be able to find a better way around the issue

Note: The reason I chose balance-rr is purely for speed - AFAIK it's the only bonding type that gives you the maximum speed, e.g. 2 x 2.5Gbit NICs is pushing through around 4-4.2Gbps host-to-host.