The textbook told me that it's hard to get open and short circuit over a broadband at microwave frequency, which is one of the reason why we use s-parameters instead of z-, y-, h-parameters(they all need short circuit to measure). But in the SOLT calibration of Network Analyzer, we do have short and open termination which can cover large frequency range. Is there any contradictory?
No. The SOLT coaxial shorts and opens are offset (i.e. moved from the reference plane) with a delay, and also have a polynomial fit for L and C, respectively. So they are far from ideal, but they are well characterized, thus the VNA can compensate.
The only clarification I would add is that the VNA knows the polynomial fit coefficients since the cal kit and VNA are a matched pair. If the wrong cal kit is used to calibrate the VNA the accuracy is degraded.