Raft is a consensus algorithm that allows distributed systems to agree on a sequence of values in the presence of failures. It provides single-key linearizability, which guarantees that every operation appears to take place atomically and in some total linear order that is consistent with the real-time ordering of those operations. However, Raft's read performance can be affected by leader leases, which require an additional round-trip to the majority of replicas on every read operation, resulting in high latency. To improve this, YugabyteDB uses leader leases, where a newly elected leader cannot serve reads until it has acquired a leader lease, and the old leader steps down upon expiry of its lease duration. This approach creates a time window where the old leader is unavailable, but the unavailability window is bounded and occurs only during failure scenarios, making leader leases safe in practice.