Flo itself is a theory paper, but we are actively using it as the foundation for our high-level Hydro APIs (where we use Rust types to enforce Flo properties): hydro.run/docs/hydrofl...

The goal of Flo is to be a unifying semantics, and to that end we show how the core ideas of existing systems like Flink, LVars, and DBSP work perfectly in the Flo model!

With Flo, we aim to extract the *essence* of streaming systems into formal properties: streaming progress, which formalizes keeping outputs up-to-date, and eager execution, which guarantees determinism in the face of arbitrary scheduling.

Streaming systems have broad “vibes” about responding to events in a timely manner and tolerating scheduling delays gracefully. But these “vibes” have remained informal and each system has brought its own (incompatible) set of guarantees.

Hey, author here, glad you found it interesting!

My talk yesterday at #SFSystemsMeetup was based on this paper. I'll share the slides soon. Thanks a lot @conorpower23.bsky.social @shadaj.me @cwensel.bsky.social and Neil for inviting me! Also thanks Jeff @ Chroma for hosting the event 💙