In performance testing we are observing that since starting, VPP is randomly loosing small number of packets (less then 1% of sent packets). Issue is random (per test case) with all packet sizes and also ipv4/ipv6/bd/xconnect setups with multi core and also single core startup configurations.
Issue can be replicated on different test beds. Once the issue occur, we are able leave the VPP in this state and start same traffic from TG. TG is reporting input errors even for very small rates (<0.5Mpps). VPP interface statistics shows all received and sent packets (rx = tx), but TG reports input errors. VPP output 'show error' is clear, 'show hardware' is without errors. Once the VPP is restarted the issue is not present.
We can see this behavior in approx. ~1/50th-1/100th test case. Each test case is doing restart of VPP.