Constant packet loss occurs when you have the following condition
This is a 3rd party (Intel/OEM) driver issue.
The VMDQ loopback feature used by Intel NIC can cause connectivity issues when we have traffic that uses different VLAN for communication (in the specific case the outgoing traffic was using a specific VLAN while the incoming traffic was using a different VLAN). Normally, when you have the same VLAN, the traffic is kept in the same ESXi host. In this case the traffic was using different VLAN and it was supposed to traverse the uplinks, but during the troubleshooting we had 2 issues:
Troubleshoothing steps
pktcap --uplink vmnicX --capture EnsPortReaderRx -w outfile.pcappktcap --uplink vmnicX --capture EnsPortWriterTx -w outfile.pcap pktcap --uplink vmnicX --dir 2 outfile.pcap Further investigation showed that the VMDQ feature is preventing this kind of traffic from working properly even if it is disabled. You may have problem disabling this feature with the intnetcli extension for esxcli as outlined by this Dell KB
https://www.dell.com/support/kbdoc/en-us/000223462/intel-e810-c-adapter-turns-off-vmdq-loopback-via-intnetcli-failed
Also, newer version of the intnetcli extension may not be available for ESXi 7.0 in the intel website, even if it is mentioned.
As mentioned, turning off the VMDQ feature (that is disabled by default with i40en 2.9.2 or later and icen 1.14.2 or later (so you do not even need to disable it with the intnetcli extension beforehand) does not seem to solve this specific problem, so the only alternative is not to use 802.X in the vApp encapsulation or a non Intel NIC (if the application has this specific VLAN requirement) , until the problem is not solved on the driver side: the driver that should prevent this behaviour is still not available.
VMDQ loopback feature is used when SRIOV is enabled, but in this case this was a problem with the ENS code of the driver.