Raft Coordination
What it is
Section titled “What it is”In a multi-node AetheriusDB cluster, every node shares the same backing storage, which means two nodes could in principle try to write the same data at the same time. To prevent that, the cluster grants each node a write lease before it writes: a time-bounded permission covering a slice of the data. A node can only write where it currently holds a valid lease.
Coordination governs who is allowed to write where — it never sits in the read path and never touches the contents of your data. Reads and normal write throughput are unaffected by it.
Why it matters
Section titled “Why it matters”The classic trade-off in shared-storage databases is correctness vs. throughput: either you funnel every write through a single leader (slow) or you risk split-brain (unsafe). Leases give you cluster-wide safety with parallel writes across nodes.
- Split-brain safe — a node with an expired lease cannot corrupt data; its stale writes are rejected.
- Elastic scale-out — a new node joins, claims a share of the write space, and starts serving traffic.
- Graceful failure — a partitioned node’s leases expire automatically, and another node picks up its range without operator intervention.
How you use it
Section titled “How you use it”Coordination is autonomic — most operators never touch it directly. The one operational action you take is draining a node before planned maintenance, so it releases its leases cleanly and peers take over before traffic is cut.
-- Drain a node gracefully before maintenance: it finishes in-flight-- writes, declines new ones, and peers pick up its range within seconds.ALTER CLUSTER DRAIN NODE 'compute-3';