Discrete Gaussian sampling on lattices is a fundamental problem in lattice-based cryptography. In this paper, we revisit the Markov chain Monte Carlo (MCMC)-based Metropolis-Hastings-Klein (MHK) algorithm proposed by Wang and Ling and study its complexity under the Geometric Series Assuption (GSA) when the given basis is BKZreduced. We give experimental evidence that the GSA is accurate in this context, and we give a very simple approximate formula for the complexity of the sampler that is accurate over a large range of parameters and easily computable. We apply our results to the dual attack on LWE of [19] and significantly improve the complexity estimates of the attack. Finally, we provide some results of independent interest on the Gaussian mass of a random q-ary lattices.