netpol test flakes in upstream ci... debugging w/

Looking at some interesting openshift prow results adopting the new netpol tests ....... pasted the logs here https://gist.github.com/jayunit100/79741d33d858ff8baf034f7da63fcabc  

The above screenshot has two test failures for the network policy upstream e2es.    This looks to me like flakey infrastructure, because , TWO TESTs both  exhibit a seemingly random, spot instance where ONE POD cannot talk to ANOTHER POD.  

ON THE LEFT we can see that EGRESS Policies are working:

expected and observed are the same...

e2e-netpol-x-6398/a X X X .

X X X X X

e2e-netpol-x-6398/a X X X . X X X X X

HOWEVER something funny happend if we zoom in :  We can see that one of the pods that WASNT supposed to be blocked DIDNT REACH the pod it was trying to talk to.  Totally random.

We can see the same thing in the other test, on the right: which tests All on All (ALLOW ALL network policy) functionality : 

What do REAL network policy test failures look like ?  for comparizion

more like this:  Where you can see that actually NONE of the egress networkpolicies are applied and all the rows for the