3 Network Address for Proxmox Cluster with CEPH?

apap

Member
Apr 18, 2021
24
2
8
50
Trying to understand the basic network setup for proxmox cluster with CEPH.

Say I have 6 physical nics: eno1 - 6

I would need to have 3 bonds (assume lacp bond):
bond0, eno1 eno2, lacp, layer 2+3
bond1, eno3 eno4, lacp, layer 2+3
bond2, eno5 eno6, lacp, layer 2+3

I would also need to have 3 bridges and 3 ips:
vmbr0, bond0, ip address 10.10.10.1/24 (for office network)
vmbr1, bond1, ip address 10.10.20.1/24, no gateway (for PVE cluster operations)
vmbr0, bond2, ip address 10.10.30.1/24, no gateway (for ceph public and cluster network)

Is this correct?

Thanks
 
Last edited:
The wiki (https://pve.proxmox.com/wiki/Network_Configuration) states:

"If you intend to run your cluster network on the bonding interfaces, then you have to use active-passive mode on the bonding interfaces, other modes are unsupported."

Does this mean if the ip address is set directly on bond1 the cluster will not work?

In my setup above, as the ip is on vmbr1 (which references bond1) and not on bond1, will the cluster work when bond1 is setup with lacp?
 
"If you intend to run your cluster network on the bonding interfaces, then you have to use active-passive mode on the bonding interfaces, other modes are unsupported."
Maybe the things are changed, but i don't see that sentence.
Instead i find:

"If your switch support the LACP (IEEE 802.3ad) protocol then we recommend usingthe corresponding bonding mode (802.3ad). Otherwise you should generally use theactive-backup mode."

So looks like the active-passive mode is just a suggestion when not possible to use the lacp.
 
  • Like
Reactions: Falk R.
Maybe the things are changed, but i don't see that sentence.
Instead i find:

"If your switch support the LACP (IEEE 802.3ad) protocol then we recommend usingthe corresponding bonding mode (802.3ad). Otherwise you should generally use theactive-backup mode."

So looks like the active-passive mode is just a suggestion when not possible to use the lacp.

For reference: