One machine failing
shouldn't be your outage.
Multi-writer runs your database across multiple nodes with consensus replication: a write is confirmed only once the cluster has accepted it, and if a machine dies the survivors keep going — measured in fault drills, not just promised.
Resilience you can put in front of a risk committee.
A write isn't acknowledged until the cluster has accepted it. Across our leader-crash, network-partition and split-brain drills: zero acknowledged writes lost, zero divergence.
When a node dies, the remaining nodes elect a new leader and writes keep flowing. A hardware failure becomes a blip in a graph — not an incident channel.
A node cut off from the cluster refuses writes rather than quietly forking your data. You'll never merge two versions of the truth on Monday morning.
We publish what we test: crashed leaders, partitions, killed majorities, slow nodes. Every claim on this page traces to a drill we ran against real clusters.
Hardware fails on a schedule nobody publishes.
On a single machine, that's a page, a scramble, and an apology to customers. On a multi-writer cluster, leadership moves to a healthy node and your application keeps writing — we've crashed leaders mid-traffic in drills and watched the cluster carry on without losing an acknowledged write.
Multi-writer is in early access: we're running it with design partners today and hardening it in the open. If losing a box can't mean losing your evening, talk to us.