It's not easy to relax: liveness in chained BFT protocols

Modern chained Byzantine Fault Tolerant (BFT) protocols leverage a combination of pipelining and leader rotation to maximize both efficiency and fairness. Unfortunately, this approach compromises liveness. We observe that even simple leader failures such as crashes can prevent the system from making progress, both theoretically, and practically. The root cause is simple: these protocols require a sequence of three or four consecutive honest leaders to commit operations. This paper makes two contributions: first, we show that, in the presence of arbitrary failures, consecutive honest leaders are necessary . When nodes fail by omission however, one can do better. As second contribution, we thus propose Siesta, a novel chained BFT protocol that successfully commit blocks that span multiple non-consecutive leaders. Siesta reduces the expected commit latency of Hotstuff by a factor of three under failures, and the worst-case latency by a factor of eight.