A replicated log is an append-only sequence of commands maintained identically across a group of replica servers. Because every replica receives the same commands in the same order, each can execute them on its local state and arrive at the same result. This is the mechanism that consensus algorithms such as Raft use to keep a cluster of machines in agreement.
The Raft paper presents this structure explicitly. In Raft, each server stores a log of entries, where each entry holds a command for the state machine; the leader is responsible for accepting new entries and replicating them to followers. An entry is committed once it is stored on a majority of servers, after which every server applies it to its state machine in log order.
This relationship between the log and the state machine is the basis of state machine replication: if the logs are identical and the commands are deterministic, the replicated state machines stay consistent. Ongaro’s Stanford dissertation develops the surrounding practical concerns, such as compacting the log when it grows and changing cluster membership safely, that any real replicated log must handle.
The replicated log has become a foundational abstraction well beyond Raft. The same idea, an ordered, durable, replicated log of records that consumers replay in order, underpins systems ranging from databases to event-streaming platforms, where the log itself is treated as the source of truth.