Why the solution obtained by CVX (the final status is solved) is not feasible?

CVX can solve this problem, the status is as follows:

But, flag is 0, which means the last constraint is not satisfied. Why???

What is the numerical value of test1 - test2 ? Constraints are only enforced to within a tolerance. See http://cvxr.com/cvx/doc/solver.html#controlling-precision but also note what is written in http://cvxr.com/cvx/doc/advanced.html#the-successive-approximation-method

It’s basically the same kind of thing( tolerance) as from your dual variable value post from last year Why are the dual variables negative?