It is well known that giant long period comets originating from the Oort’s cloud could be the most threatening celestial bodies. The warning time could indeed be very short, five years, and the kinetic energy could be sufficient for global and durable effects on Earth, killing all life forms on the surface. Humans might nevertheless be able to survive decades in underground shelters. Based on a model used to determine the minimum number of people to survive on another planet, a classification of long terms shelters is proposed, taking all needs into consideration and also the double redundancy principle. Accumulation shelters corresponds to shelters with lots of resources but a weak autonomy, and therefore a well-established limited lifetime around thirty years. Borderline shelters correspond to long term shelters with strong autonomous capacities but little margins and high risks of collapse due to misconceptions, underestimations and insufficient testing. Complete shelters correspond to ideal shelters with double redundancy for every system and also for the working capacity. Thanks to a high resilience, they could last decades, eventually saving humanity from extinction. The limits of the shelters are discussed, as well as uncertainties. Given the short warning time, it is impossible to test any shelter, and, as a consequence to avoid errors in the design of a complete shelter. The risk is indeed high that some problems are underestimated and a slow but unstoppable degradation of life conditions would lead to the death of the survivals, whatever the preparation and motivation of the survivals.