Skip to content

Who gets to decide?

Five lessons on Raft elections, current-term commitment, fencing, safe reads and changing voters.

Start series →

Follow the decisions in order

  1. Part 1 of 5

    Staff

    The old leader is alive. Why won't the cluster accept its writes?

    Explain minority behavior, term-based elections, candidate freshness and the scope of reads under a network partition.

  2. Part 2 of 5

    Staff

    Three servers have the log entry. Why won't Raft commit it?

    Distinguish log replication from commitment of prior-term entries, and connect Raft terms, election restriction and current-term commitment.

  3. Part 3 of 5

    Staff

    The lease expired. Why can its previous owner still write?

    Distinguish lease expiration from the physical ability to write, and enforce non-regression with destination-side fencing.

  4. Part 4 of 5

    Staff

    The leader has applied every entry. Why might its read still be stale?

    Establish linearizable read authority using a quorum-confirmed read barrier and state-machine application, separate from local replay lag.

  5. Part 5 of 5

    Principal

    We added two voters. How could a routine change split the cluster?

    Prove overlapping authority across reconfiguration and plan learner promotion, joint consensus or sequential membership operations with failure recovery.