Proxmox 4.1, Open vSwitch & MTU Settings

[WvdW]

Hi,

We have been using open vswitch since its introduction in 3.x without any problems. We have just deployed a new 4.1 server with exactly the same networking configuration as we had before on 3.4. In syslog we are now getting a huge number of incorrect mtu size errors for one of the vswitch tap devices, and it looks like all of these packets are being dropped by Proxmox because of this. Its mostly http traffic. The errors state that the packet mtu is exceeding 1500 / 1501 / 1502.

In turn this is causing a number of cascading issues in that the dropped packets aren't reaching the intended VM and then causing errors with the downstream clients connected to the VM.

No firewall has been configured or started on the Proxmox server to eliminate that from the problem.

Has anybody been experiencing the same problems? Anybody have an idea of what is causing it? For right now it looks like there is a problem with the vswitch implementation in 4.1 as nothing else in the network has changed and if we start the original 3.4 server up again everything is working 100%.

Werner

 

[WvdW]

After having performed various tests and closely monitoring what's happening on the Proxmox host, I decided to change the network config from open vswitch to standard linux bridges. Retested all the items in the testing matrix and everything was solved!
The only difference between the two configs is the move from vswitch to linux bridges. So it definitely seems like there is a problem with the vswitch package in Proxmox 4.1 with the latest updates (well up until today 2016/03/15). These tests were done using several different machines, with several fresh installs and the results remain consistent, so its not a machine or install specific issue.

Werner

 

[WvdW]

Okay, slight problem... everything seemed to be working okay after the change to Linux bridges but upon closer inspection of the network interfaces at both a physical and virtual layer I found the following:
- No packet transmission or receiving errors on any physical nics connected to the proxmox host.
- A lot of of packet receiving errors on all virtual nics (vnics in different vms and also connected to different bridges). More specifically frame errors with an indication of a too long frame lengths.

Packet receiving errors potentially indicate a faulty nic or switch port. In this case as these are vnics it indicates a faulty transmission from the bridge "switch" to the vnic. So in the end even though most of the issues were resolved by changing to Linux bridges, it definitely seems like there is still a problem with the Proxmox networking stack?
Traffic in and out of the physical nics are flowing error free but once the physical passes it to the virtual bridge things start going wonky.

Werner