A fault tolerant service requires at least triple redundancy. At least three machines are running identical services on identical parameters. The results are then compared. If all three results are identical, the result is returned. One machine can go down in this configuration without effecting the results. Hot spares are available to replace a faulty machine and join an operational machine pair.
If a service is guaranteed to be error free it is said to be fault tolerant. Ideally, one would wish all services to be fault tolerant but this is an expensive capability and demands large quantities of resources to obtain this level. Only critical services need to be fault tolerant. Most systems work well at highly available levels.