for functionality and correctness within its niche.
In distributed consensus algorithms (e.g., Raft, Paxos), engineers use chaos engineering to introduce network partitions, delayed packets, and node failures. appears as a specific test harness that targets hierarchical state machines running across a cluster. Unlike standard chaos tools that randomly kill processes, HSMMaelstrom focuses on attacking state transitions at multiple levels of abstraction simultaneously. HSMMaelstrom
| Metric | Centralized HSMM | HSMMaelstrom | |--------|----------------|---------------| | Throughput (obs/sec) | 1,200 | 3,400 | | 95th percentile latency | 450 ms | 620 ms (due to async) | | Recovery time after crash | N/A (restart all) | 1.2 s | | State prediction F1 | 0.91 | 0.90 (no loss) | for functionality and correctness within its niche