Leadership

Hashicorp Raft elects one leader. That process is the only writer of replicated state (membership, locks, leases).

Followers and observers accept the same Runtime API. Mutations they cannot commit locally are forwarded to the leader.

How you see it

• CLI: clusdr leader and the leader / role column on clusdr members
• SDK: Leader()
• Watch: leader.changed

Timeouts

Defaults: 150ms Raft heartbeat and election, 75ms leader lease. Failover with those values is typically a few hundred milliseconds.

If peer RTT is larger, raise the timeouts or you get extra elections. leader_lease_timeout must be ≤ heartbeat_timeout.

Bootstrap

Only the first voter uses clusdr start --bootstrap. Passing --bootstrap again on a node that already bootstrapped is safe (ErrCantBootstrap is ignored).

Observers never become leader. After promote, they can.

Related

• Membership
• Configuration
• clusdr leader