I've got some really bad moments with the intel network card X710 10GBit SFP, so that's my issue:
I have some VMs (pfSense) in a vmware cluster composed by 2 hosts ESXi,
that VMs are a set of Master-Slaves(Backup) set with a CARP Sync that permit a sort of software-HA.
I had the first problem in configuration phase when i moved my (configured and ongoing) vSubnet (my LAN) from ethernet interface to SFP 10 GBit.
CARP packets were redunded 3-4 times.
Looking in the network card's datasheet i found it's feature for multicast packs called "VDMq" (that by default should be disabled), in fact it wasn't so I've done it manually, after a driver update.
The problem was gone... until the moment when I configured the second interface 10 GBit/s (of the same physical net-card).
VM's configuration is the same, but the problem of "over-redundancy" came out again, even with the settings that solved the problem the first time it occurred!
So now the hosts CARP intercomunication proceed smoothly, but Masters receive our own packets going in backup mode...
sorry for english nazi-grammar,
hope there'll be soon a solution,
I hope this can be helpful for someone:
to disable VDMq for both interfaces (of the same network adapter) instead of
esxcli system module parameters set --module i40e --parameter-string "VMDQ=0"
(on host ESXi) we set
esxcli system module parameters set --module i40e --parameter-string "VMDQ=0,0"
because the value does not refer to the adapter but to the single port module!