Jepsen 9: The center cannot hold

Distributed systems often claim to save our data durably, provide isolated transactions, and make writes visible to reads. Kyle Kingsbury explores anomalies in three distributed systems—Tendermint, Hazelcast, and Aerospike—and shares general strategies for correctness testing using Jepsen, a distributed system testing harness that applies property-based testing to databases to verify their correctness claims during common failure modes: network partitions, process crashes, and clock skew.