We were trying to configure the MLAG on the Arista switch on each uplink group of the CISCO UCS ( UCSC-C240-M3s) server . Initially ESXi 6.0 was installed and when we trying to enable the LACP on the dVswitch , host got disconnected and also stooped pinging and once we disabled the LACP option it came online. VMware support was involved and asked to install all the firmwares but still we had the same issue.
As a trouble-shooting step we installed the windows OS and once we enable the LACP option in the windows NIC Teaming it started same issue as server got disconnected and also stopped pinging.
Another set of ESX servers with same MLAG configuration on the HP Blade Enclosure on Arista was working fine without any issue .
We involved Cisco for the server and Arista from the switch end for the support , from the Cisco we didnt get any proper respond , finally the issue got escalated to the Tier-1 support in Arista and they found the issue is when the Cisco server is configured for trunk mode, the server is sending LACP frames tagged with a vlan id of 0. Currently, in the version of code we are running on this arista switch platform, it’s dropping these frames as it’s not accepting tagged LACPBPDUs. In addition, when the server is configured as an access port, it is sending the LACP frames tagged with the access vlan specified. Even if there is no default vlan specified, the server is still sending the LACP frames tagged with vlan 0. It seems there is no way to have the Cisco server to send the LACP frames un-tagged like HP blade chassis, tintris which are already on mlag.
Requested the Cisco engineer to see the reason for the server is tagging the LACP frames but didnt get the answer so as per the Arista recommendation we upgraded the latest code on the switch to Version4.16.7M which fixed the issue.
same error with EOS 4.17.3F
l2-protocol encapsulation dot1q vlan 1
applied to the port -channel on Arista resolves the issue