Jepsen: Capela dda5892

Summary

Background Much of modern software involves application logic written in some programming language, which stores state in one or more distributed, fault-tolerant databases. The storage system and application logic usually represent data in different ways, requiring additional languages to query and transform data: SQL, stored procedures, etc. There are also usually translation layers between application logic and storage: client libraries, query builders, Object-Document Mappers, and so on. Capela is an unreleased, distributed, general-purpose programming environment which aims to simplify application development by unifying processing and storage in a single distributed system. Capela hopes that their integrated model will make programs shorter and easier to understand for both human engineers and Large Language Models. Capela programs are written in a dialect of Python, augmented with a sophisticated type system. Capela allows programmers to write distributed classes whose instances are automatically replicated and persisted to disk. Method invocation is transactional, with Strong Serializable semantics for pure functions and those which mutate Capela state. Capela also plans to allow functions to perform external I/O, like making calls to HTTP services. Support for external I/O, and its intended consistency model, are still under development. Readers familiar with distributed languages like SR, MPD, Erlang, Oz, or Bloom will recognize elements of Capela. As with Linda’s Tuple Spaces, Capela models distribution not in terms of message passing, but as access to transparently distributed data structures. Like Smalltalk and other image-based languages, Capela persists program state directly, and allows programs to be modified over time. Indeed, Capela feels somewhat like an object-oriented database with stored procedures. More recently, blockchain systems like Ethereum have re-popularized transactional execution over replicated persistent state, viz. smart contracts....