Break the cluster. Keep your data.
A three-node Raft group, simulated in your browser. Elections, replication, majority commit and ReadIndex reads are real code, and the verdict comes from the nodes' logs.
What is simulated
A simplified model for illustration: three nodes, an in-memory network and snapshots reduced to markers. The production implementation is covered by the real test suites in the repository.
- Pre-vote and randomised election timeouts, so a cut-off node cannot disrupt a healthy leader
- Heartbeats and log replication with the consistency check and conflict truncation
- A commit only when a majority holds an entry from the leader's own term
- Linearizable reads through ReadIndex: the leader confirms it is still the leader before it answers
- Check-quorum: a leader that loses its majority steps down
- Crashes keep the durable log; a snapshot interrupted mid-write is discarded on restart
Now try it on real nodes.
We will run the same failures against a real cluster with you and walk through what happens.
The same failure on real processes, recorded in the Admin console. The 15 seconds of failure detection play faster.