Skip to main content

The Raft Log

The Raft Log is the central data structure in Ratis that ensures consistency and fault tolerance. It is an append-only log of operations that are applied to the state machine. The log is replicated across all the servers in a Raft group, and an entry is only considered "committed" when it has been successfully replicated to a majority of the servers.

Raft Log Implementations​

Ratis provides two primary implementations of the RaftLog interface:

1. MemoryRaftLog​

  • Description: This is a simple, in-memory implementation of the Raft Log. It stores all log entries in a list in the Java heap.
  • Use Case: The MemoryRaftLog is primarily used for testing and is not recommended for production use. Since it does not persist data to disk, all log entries will be lost if the server restarts.
  • Performance: It is extremely fast due to its in-memory nature, but it does not provide any durability guarantees.

2. SegmentedRaftLog​

  • Description: This is the production-ready, disk-based implementation of the Raft Log. It stores log entries in a series of segmented files on disk. Each segment has a configurable maximum size, and when a segment is full, a new one is created.
  • Use Case: The SegmentedRaftLog is the default and recommended implementation for production environments. It provides the durability and fault tolerance guarantees that are expected from a Raft-based system.
  • Performance: While not as fast as the MemoryRaftLog, the SegmentedRaftLog is highly optimized for performance. It uses a combination of in-memory caching and efficient disk I/O to provide a good balance between performance and durability.

Key Concepts​

  • Log Index: Each entry in the Raft Log is assigned a unique, monotonically increasing log index.
  • Term: A term is a logical clock in Raft that is used to detect stale leaders. Each term has a single leader, and the term number increases whenever a new leader is elected.
  • Commit Index: The commit index is the highest log index that is known to be replicated on a majority of the servers. Log entries up to the commit index are considered "committed" and can be safely applied to the state machine.
  • Snapshotting: To prevent the Raft Log from growing indefinitely, Ratis uses snapshotting. A snapshot is a point-in-time copy of the state machine's state. Once a snapshot is taken, all the log entries up to that point can be safely discarded.