In heterogeneous swarm systems participants can come and go, the communication topology may change at any time, data replication is asynchronous and partial, and the behavior of local agents typically differs between nodes. These systems are notoriously hard to design and reason about, in particular concerning the observed system behavior that we desire to emerge from the programmed behavior of the individual agents. Nevertheless such systems are used in mission critical contexts, e.g. in factories for manufacturing process orchestration. The reason is that they offer uncompromising availability and resilience of computing services.

This paper presents a set of TypeScript libraries that allow the programmer to model peer-to-peer workflows as state machines, execute them using the Actyx middleware, and check the shape of these machines for conformance to a swarm protocol. The swarm protocol describes an idealized global view onto the cooperation of machines of different roles. It directly corresponds to a diagrammatic representation a product manager would sketch on a whiteboard, thus allowing to verify that the coded state machines correctly implement the product specification. A well-formed swarm protocol also guarantees that conforming machines will achieve eventual consensus on the overall state progression even in the absence of further coordination.

Tool repo: https://github.com/Actyx/machines