Not Tagging on OVS VLAN Ports

[dlaube]

Hi,

I created a cluster with OVS networking stack and vlan to sperate networks.
The configuration on each cluster node is the same as described in the wiki (https://pve.proxmox.com/wiki/Open_vSwitch#Example_2:_Bond_.2B_Bridge_.2B_Internal_Ports)
except that i used 4 interfaces in the bond0 which should not make a difference in the behaviour.
The switch is configured properly to handle LACP on the bonded ports and to tag the appropriate vlan tags.

Now there is an issue with one specific vlan id (102). When two vms are given a network port with that id a connection is only established when the reside on the same proxmox node.
When i move one vm to another proxmox node the vms can't reach each other.

I thought that the switch might be misconfigured but that does not seem to be the case because other vlan ids work fine. The same setting is working for vlan 16 and vlan 17. The vlan configuration for 16,17 and 102 is the same on the switch (i double checked that).

When i plug myself into an untagged port 102 on the switch i can reach all other devices on the network that have these vlan id but the proxmox vms can't.

I don't know if this is a proxmox issue or something in ovs going horribly wrong or even if the switch is the fault here but i wonder if someone can come up with an explanation for it.

(The Switch is an HP 1810-24G with LACP active on the proxmox side and vlan ids tagged on these trunk ports)

Thank you for your help
Daniel

 

[dlaube]

Regarding the issue: It seems like the management VLAN on the HP switch is causing the traffic to be dropped when coming from the proxmox servers.

 

[Stoiko Ivanov]

Hm - that sounds curious - what is the management VLAN (haven't heard it until now, but my experience with HP-switches is a bit dated).

Else - if you want to investigate further - I would suggest that you take a look at the traffic leaving/entering the PVE-hosts with tcpdump - maybe you can find a difference between the working VLANs 16,17 and the not working 102

 

[dlaube]

The switch is a managemed switch that you can configure through web. You can assign a specific vlan id to that internal management interface. We did that and ever traffic coming from the proxmox host with that id was silently dropped.
We changed the management vlan id to some other vlan w had and the problem shifted to that new vlan.
So it is indeed an issue with the management vlan (perhaps with management vlan on lacp bonds because the other connections (non proxmox hosts) didn't suffer from this problem).

 

[Stoiko Ivanov]

It could also be a bug in the switch's firmware - if you have the time you could ask HP-Support for an explanation.

Thanks for sharing!