Leader Election

A unique feature of OmniPaxos is the guaranteed progress with one quorum-connected server using Ballot Leader Election (BLE). In this section, we will see how the leader election is used in OmniPaxos.

To detect any leader failures and elect a new leader, you must call the function tick() periodically. The number of ticks required before an election timeout is triggered can be configured with the election_tick_timeout field of ServerConfig. For example, if tick() is called every 10ms and the election_tick_timeout is 10, then the timeout will be 100ms.

// Call this periodically
omni_paxos.tick();

If a leader has failed, it will be detected in one election timeout and a new leader is typically elected in the next timeout (if possible).

Note: The leader_priority field in ServerConfig allows the user to give desired servers a higher priority to get elected upon a leader change. Calling try_become_leader() at one of the servers is a manual attempt to become the leader.

In some cases, a network error will require the resending of messages between leader and follower. The tick() function also drives this behavior. The number of ticks required before checking if a message needs to be resent can be configured with the resend_message_tick_timeout field of ServerConfig. When log batching is enabled, tick() also flushes batched entries every flush_batch_tick_timeout ticks (see Storage).

Note: If you prefer not to drive tick() yourself, the Async Runtime wraps OmniPaxos in a background actor that ticks automatically.