I thought I would post this to maybe assist anybody else that has a HP Procurve 2910al-24G network switch and using 'bridge-vlan-aware' parameter in their server interfaces.
I have a 3 server cluster consisting of 2x Dell R420 and 1 x R710. I converted them all recently from Proxmox 5.4 to 6.1 with no issues even converting Ceph to Nautilus went without a hitch. The problems started when I bought a HP 2910al-24G network switch. I have 4 vlans on my network and all have worked fine using my old Netgear switch. With a bond0 active-passive bond and vmbr0 has 'bridge-vlan-aware yes' this allows me to tag the vm's and keep everything simple especially when using a virtual firewall.
Problem:
The issue I had was when I changed the bond0 to LACP type layer2+3 on the servers and connected them to the HP switch.The bond would come up and be healthy at both ends. Any untagged traffic would go out to the switch but tagged traffic would not get to the switch. After alot of Goolge-ing and reading Proxmox network docs I had to give up and split out the tagged vlans on the bond so bond0.2, bond0.5 etc then create new bridges for each vlan. This allowed traffic vlans to start working again. The only problem with this set up is that the virtual firewall needs 5 interfaces (untagged and 4 vlan).
Solution:
Although this allowed traffic to start working across the servers its not ideal for me and I needed to know why the 'bridge-vlan-aware' parameter on the bridge interfaces did not work any more. I finally traced it to a debian module not being explicitly set in /etc/modules as doing a modprobe returned an error . To get the 'bridge-vlan-aware yes' to work I had to put '8021q' in the '/etc/modules' file then reboot the servers.
I could not find a mention of this in Proxmox manuals, but after a lot of digging I found it on some Debian networking sites. So maybe it's just my setup, or converting from 5.4 to 6.1 removes 8021q for some reason.
Hope this helps somebody else, and saves alot of time and head scratching
I have a 3 server cluster consisting of 2x Dell R420 and 1 x R710. I converted them all recently from Proxmox 5.4 to 6.1 with no issues even converting Ceph to Nautilus went without a hitch. The problems started when I bought a HP 2910al-24G network switch. I have 4 vlans on my network and all have worked fine using my old Netgear switch. With a bond0 active-passive bond and vmbr0 has 'bridge-vlan-aware yes' this allows me to tag the vm's and keep everything simple especially when using a virtual firewall.
Problem:
The issue I had was when I changed the bond0 to LACP type layer2+3 on the servers and connected them to the HP switch.The bond would come up and be healthy at both ends. Any untagged traffic would go out to the switch but tagged traffic would not get to the switch. After alot of Goolge-ing and reading Proxmox network docs I had to give up and split out the tagged vlans on the bond so bond0.2, bond0.5 etc then create new bridges for each vlan. This allowed traffic vlans to start working again. The only problem with this set up is that the virtual firewall needs 5 interfaces (untagged and 4 vlan).
Solution:
Although this allowed traffic to start working across the servers its not ideal for me and I needed to know why the 'bridge-vlan-aware' parameter on the bridge interfaces did not work any more. I finally traced it to a debian module not being explicitly set in /etc/modules as doing a modprobe returned an error . To get the 'bridge-vlan-aware yes' to work I had to put '8021q' in the '/etc/modules' file then reboot the servers.
I could not find a mention of this in Proxmox manuals, but after a lot of digging I found it on some Debian networking sites. So maybe it's just my setup, or converting from 5.4 to 6.1 removes 8021q for some reason.
Hope this helps somebody else, and saves alot of time and head scratching
