MAG$π$: Types for Failure-Prone Communication

Abstract

Multiparty Session Types (MPST) are a typing discipline for communication-centric systems, guaranteeing communication safety, deadlock freedom and protocol compliance. Several works have emerged which model failures and introduce fault-tolerance techniques. However, such works often make assumptions on the underlying network, e.g., TCP-based communication where messages are guaranteed to be delivered; or adopt centralised reliable nodes and an ad-hoc notion of reliability; or address only a single kind of failure, e.g., node crash failures. In this work, we develop MAGπ-a Multiparty, Asynchronous and Generalised π-calculus, which is the first language and type system to accommodate in unison: (i) the widest range of non-Byzantine faults, including message loss, delays and reordering; crash failures and link failures; and network partitioning; (ii) a novel and most general notion of reliability, taking into account the viewpoint of each participant in the protocol; (iii) the spectrum of network assumptions from the lowest UDP-based network programming to the TCP-based application level. We prove subject reduction and session fidelity; process properties, (deadlock freedom, termination); failure-handling safety and reliability adherence.

MAG$$\pi $$: Types for Failure-Prone Communication

“…Case study ( § 6): we further demonstrate the use of timeouts and undirected branching to model a Domain Name System (DNS) distributed protocol with a cache and in-built load-balancer; we also show the properties it satisfies, including safety and deadlock freedom. Further examples are available in the technical report [23], including a prototype specification of a leader election algorithm used by consensus protocols.…”

“…This ensures that no messages are ignored or lost for reliable communication. Proofs of our theorems, along with any auxiliary results, are given in the technical report [23].…”

“…This is the only requirement since all theorems in § 4 (i ) are parametric on the reliability function R, including fully reliable networks; and (ii ) are proven for (Σ; R)-φ s (Γ ). [23].…”