Fail Over Issue on ESXi vDS Uplinks

[JonM2020]

We have a large ESXi farm on ESXi6.7 latest patch as of now. The issue we are getting is that occasionally the vmnics on the ESXi host will go down just seemingly randomly but the big big issue for us is that the switch port on the affected uplink stays up and continues to send traffic to the guests on the affected ESXi host on the downed vmnic.

This causes the guests on this host to lose connection to the outside world even though they are still up and traffic should be being directed to the other(fail over) uplink on the same dvs switch.

Any advice would be welcome as the config on the vmware side looks fine and our network guys cant see an issue from their side either!

[daphnissov]

Sounds like an opportune time to open a case with GSS.

[JafarTavana]

on the your vDS switches set Network failure detection to Beacon probing.

when you use this method for define which uplink is working or not, vmkernel send a packets out from one physical NIC and then receive it on the other NICs, if one NIC can't get this packet, ESXI know the network connection was lost for receiver NIC(if NIC be in the UP state maybe one physical switches in the path goes down and Beacon Probing can tell the ESXI this port is useless)

[JonM2020]

Hi Jafar

We are using a port channel with the physical uplink pair, do you think beacon probing could help the physical switch detect that the vmnic is down aswell? We believe that that the switch keeps sending traffic to the downed vmnic perhaps because as it has not received a notification from esxi or the physical nic that it is no longer available.

[Scott Vessey]

Do you have “notify switches” set to ”no”?

[JonM2020]

Set to notify.

[JonM2020]

We are using the QFLE drivers for network on these hosts on the latest versions. Bizarrely when I roll back to the legacy BXN2X drivers the issue is resolved as the switch can see that the vmnic is now down and routes traffic to the failover nic, anyone seen anything like this before?

[JafarTavana]

I think if you set both side to active mode, LACP send LACPDU packet by the each port together and if one port don't receive this packet format send port to disable mode.

and in your condition, when your switch send LACPDU from which port goes down to the vDS, it can't replay and Physical switch put the port out of LACP Channel .