A server rack holds a number of modular servers configured for liquid cooling by passing a liquid coolant through interiors of the servers. Failure management of the cooling system is by management of the servers in segregated fault domains. Each fault domain comprises a number of the servers serviced by a dedicated coolant circuit that is segregated from the cooling circuits of the other fault domains. Potential liquid coolant leaks in a specific fault domain can be identified by monitoring liquid coolant levels in the respective coolant circuits. Each fault domain can include a separate, dedicated heat exchanger and a separate, dedicated coolant reservoir.