article --summary
The summary
This article walks through deterministic simulation testing, a way to make failures in distributed systems repeatable. The core problem is that partial failures happen alongside live traffic, so the same fault can play out differently each run and be hard to debug.
The technique swaps real clocks, networks, and storage for simulated versions, all driven by a single-threaded execution loop. Because of this, the test can inject hard crashes as well as everyday slowdowns such as network and disk latency spikes, and every run can be replayed exactly the same way.
The author applies the approach to a quorum key-value store as a running example, showing code from Tickloom, a Java implementation. Readers interested in testing distributed software will find it a practical look at how the method is built in practice.
Editorial note: This is a summary of an article first published by Martin Fowler. The full piece is on their site.