What do services have to do with it, if we are talking about fault tolerance and have reduced it to replicas?

Replicas (for example, in a database) provide fault tolerance at the data level — if one copy is unavailable, another can be used. But services are the logic that uses this data. Service fault tolerance includes not only data replication but also load balancing, automatic failover, retries, timeouts, and error handling.

That is, replicas solve the problem of data availability, while services address the availability and resilience of business logic. Without fault-tolerant services, replicas will do little if the service crashes or hangs. Therefore, when discussing fault tolerance, it is important to consider both services and replicas, as they complement each other.