Skip to main content

Support Priority in Leader Election

Ratis supports assigning priorities to Raft peers, influencing the leader election process to favor higher-priority servers. This mechanism allows administrators to guide the cluster towards a preferred leader, which can be beneficial for performance, resource utilization, or operational reasons (e.g., preferring a server in a specific rack or with more powerful hardware).

How Priority Affects Leader Election​

When a server becomes a candidate and initiates a leader election, the priority assigned to each peer plays a crucial role:

  1. Priority Assignment: Each RaftPeer can be assigned an integer priority. This is typically done during the cluster configuration or dynamically using the Ratis CLI.

    • Configuration: You can set the priority when building RaftPeer objects:

      import org.apache.ratis.protocol.RaftPeer;

      RaftPeer peerWithPriority = RaftPeer.newBuilder()
          .setId("peer1")
          .setAddress("localhost:5000")
          .setPriority(100) // Set a higher priority
          .build();

    • Ratis CLI: Use the ratis-cli peer setPriority command to dynamically adjust peer priorities:

      $ ratis-cli sh peer setPriority -peers <PEER_ID:PEER_ADDRESS,...> -addressPriority <PEER_HOST:PEER_PORT|PRIORITY,...>
      # Example: Set priority 100 for peer at localhost:5000
      $ ratis-cli sh peer setPriority -peers peer1:localhost:5000,peer2:localhost:5001 -addressPriority localhost:5000|100

  2. Higher Priority Veto Power: If a candidate (say, Peer A) requests a vote from another peer (Peer B) that has a higher priority than Peer A, and Peer B rejects Peer A's vote, then Peer A will immediately step down and abort its election attempt. This mechanism ensures that a lower-priority candidate does not win an election if a more suitable, higher-priority candidate is available and active.

  3. Requirement for Higher Priority Acceptance: For a candidate to successfully win an election, it must not only secure a majority of votes from the voting members but also ensure that all higher-priority peers have either voted for it or have not rejected its vote. If there are higher-priority peers that have not yet responded, the candidate will wait for their responses (up to the election timeout). This ensures that the cluster converges on the highest-priority available leader.

Benefits​

  • Preferred Leader Selection: Allows you to designate specific servers as preferred leaders, which can be beneficial for performance optimization (e.g., choosing a server with better hardware or network connectivity).
  • Operational Control: Provides a mechanism to influence leadership during maintenance or recovery scenarios.
  • Predictable Behavior: Helps in making the leader election process more predictable in heterogeneous environments.

References​