In this paper, we consider a single server discrete time inventory system with postponed demands and server vacation. Arrival process follows DMAP and ordering policy follows policy with geometrically distributed lead time. Due to depletion of inventory in the stock to zero, server may avail vacation immediately or after some random time. Here, we proposed three different vacation policies, namely modified multiple vacation (random idle time before each vacation), single (only one vacation) and multiple (one after another vacation). The demand who arrives during the stock-out period or server vacation period enters the pool of infinite size. These pooled demands are selected one-by-one according to a rule. The vacation time, the server idle time and the inter-selection time follow independent DPH distribution. The joint probability distribution of the number of demands in the pool, the server status, the on-hand inventory level is obtained in the steady-state case. We derived the total expected cost function in the steady-state case for each vacation model. We present some numerical examples and try to justify the best vacation policy for different parameter rates. Also, we analyse the impact of system costs on the total optimal cost of modified multiple vacation policy.