Our objective is to find a necessary and sufficient condition for consensus over a random network generated by i.i.d. stochastic matrices. We show that the consensus problem in all different types of convergence (almost surely, in probability, and in L-p for every p >= 1) are actually equivalent, thereby obtain the same necessary and sufficient condition for all of them. The main technique we used is based on the stability in a projected subspace of the concerned infinite sequences.