← notes

Raft log replication

Following an entry from the leader’s log to a committed state-machine update.

raftconsensus

Replication is not commitment

The leader appends an entry locally, sends it to followers, and advances its commit index according to the protocol’s majority and term rules. Having a copy on one follower does not by itself mean the entry is committed.

Storage underneath consensus

The replicated log needs durable local storage. LSM Tree describes one family of storage structures; RocksDB explores background compaction.

Next question

How should a system balance batch size, durable writes, and tail latency? Keep the local storage contract separate from the replication protocol.

Backlinks

Connections will appear here as the garden grows.

Related Notes

Explore the notes garden →