This is a type of test failures in daily jobs, for example [0].
It is affecting only cx7 with mlx5 driver (not e810cq with avf driver), only udp tests (not tcp or other), and interestingly both ASTF (both cps and tput for nat44ed) and STL (det44) traffic profiles. It does not affect corresponding ASTF tests with ip4base routing [3].

When investigated so far, VPP counters suggest the packet is lost somewhere between TG on in-side [1] and VPP [2].
Low scale (h1024) usually passes (losses are visible are also only in out2in direction, but not as small), so it is not easy to get a failure in low enough load.

[0] https://s3-logs.fd.io/vex-yul-rot-jenkins-1/csit-vpp-perf-ndrpdr-weekly-master-2n-icx/84/log.html.gz#s1-s1-s1-s2-s80-t1-k2-k14-k14
[1] https://s3-logs.fd.io/vex-yul-rot-jenkins-1/csit-vpp-perf-ndrpdr-weekly-master-2n-icx/84/log.html.gz#s1-s1-s1-s2-s75-t1-k3-k7-k1-k1-k1-k8-k13-k1-k2
[2] https://s3-logs.fd.io/vex-yul-rot-jenkins-1/csit-vpp-perf-ndrpdr-weekly-master-2n-icx/84/log.html.gz#s1-s1-s1-s2-s75-t1-k3-k7-k1-k1-k1-k8-k14-k1-k1-k1-k1
[3] https://s3-logs.fd.io/vex-yul-rot-jenkins-1/csit-vpp-perf-ndrpdr-weekly-master-2n-icx/84/log.html.gz#s1-s1-s1-s2-s67