Speak Now: Safe Actor Programming with Multiparty Session Types (Extended Version)

TL;DR

This paper presents Maty, an actor language that integrates multiparty session types to prevent communication errors, support multiple sessions, and enhance scalability and fault tolerance in distributed applications.

Contribution

Maty is the first actor language supporting static multiparty session typing and multiple concurrent sessions, combining session safety with actor scalability.

Findings

Maty guarantees communication safety, preventing message mismatches and deadlocks.

Implementation in Scala demonstrates expressiveness with real-world benchmarks.

Supports Erlang-style supervision and cascading failure handling.

Abstract

Actor languages such as Erlang and Elixir are widely used for implementing scalable and reliable distributed applications, but the informally-specified nature of actor communication patterns leaves systems vulnerable to costly errors such as communication mismatches and deadlocks. Multiparty session types (MPSTs) rule out communication errors early in the development process, but until now, the many-sender, single-receiver nature of actor communication has made it difficult for actor languages to benefit from session types. This paper introduces Maty, the first actor language design supporting both static multiparty session typing and the full power of actors taking part in multiple sessions. Maty therefore combines the error prevention mechanism of session types with the scalability and fault tolerance of actor languages. Our main insight is to enforce session typing through a flow-sensitive effect system, combined with an event-driven programming style and first-class message handlers. Using MPSTs allows us to guarantee communication safety: a process will never send or receive an unexpected message, nor will a session get stuck because an actor is waiting for a message that will never be sent. We extend Maty to support Erlang-style supervision and cascading failure, and show that this preserves Maty's strong metatheory. We implement Maty in Scala using an API generation approach, and demonstrate the expressiveness of our model by implementing a representative sample of the widely-used Savina actor benchmark suite; an industry-supplied factory scenario; and a chat server.