Communications of the ACM
Ironfleet: Proving Safety and Liveness of Practical Distributed Systems
Credit: Getty Images
Distributed systems are notorious for harboring subtle bugs. Verification can, in principle, eliminate these bugs, but it has historically been difficult to apply at full-program scale, much less distributed system scale. We describe a methodology for building practical and provably correct distributed systems based on a unique blend of temporal logic of actions-style state-machine refinement and Hoare-logic verification. We demonstrate the methodology on a complex implementation of a Paxos-based replicated state machine library and a lease-based sharded key-value store. We prove that each obeys a concise safety specification as well as desirable liveness requirements. Each implementation achieves performance competitive with a reference system. With our methodology and lessons learned, we aim to raise the standard for distributed systems from "tested" to "correct."
1. Introduction
Distributed systems are notoriously hard to get right. Protocol designers struggle to reason about concurrent execution on multiple machines, which leads to subtle errors. Engineers implementing such protocols face the same subtleties and, worse, must improvise to fill in gaps between abstract protocol descriptions and practical constraints such as "real logs cannot grow without bound." Thorough testing is considered best practice, but its efficacy is limited by distributed systems' combinatorially large state spaces.
In theory, formal verification can categorically eliminate errors from distributed systems. However, due to the complexity of these systems, previous work has primarily focused on formally specifying,1, 8, 18 verifying,20 or at least bug-checking9 distributed protocols, often in a simplified form, without extending such formal reasoning to the implementations. In principle, one can use model checking to reason about the correctness of both protocols15 and implementations.17 In practice, however, model checking is incomplete—the accuracy of the results depends on the accuracy of the model—and does not scale.1
No entries found