Interactive lab
Replication Lab
Every claim in the Database Replication series, running as a cluster you can configure and break. Write to a primary and watch replicas fall behind. Kill the leader mid-write and count what the failover discards. Tune n, w and r until the quorum overlap stops protecting you.
Nothing here is a video. Each node holds real keys and versions, so the anomalies emerge from the simulation rather than being animated for effect.
What each module covers
One node accepts writes and streams them to replicas. Change the acknowledgement mode, the log format, and what happens when the primary dies.
Two partitions replicating to replicas of different speeds, so all three lag anomalies are reachable: reading your own writes, monotonic reads, consistent prefix.
Several leaders accepting writes. Pick a topology and a conflict resolution strategy, then watch what each one costs.
No leader at all. Tune n, w and r, take nodes down, and find where the quorum overlap argument stops holding.