#differential #dataflow #datalog #graphql #query


A reactive query engine built on Differential Dataflow

2 unstable releases

0.2.0 Jul 17, 2019
0.1.0 Mar 31, 2019

#5 in #graphql-query

MIT license


Declarative Dataflow

A reactive query engine built on differential dataflow.




Interactive use: Declarative accepts queries expressed in a Datalog-inspired binding language and turns them into differential dataflows dynamically and at runtime. This works equally well as a library embedded into a specific application or as a standalone service (e.g. via the included WebSocket server).

Reactive relational queries: Declarative provides a relational query language, with full support for traditional binary joins, worst-case optimal n-way joins, antijoins, various aggregates, predicates, unions, and projections. Queries are made up of rules, which can depend on each other (or recursively on themselves). Query results are updated incrementally as inputs change.

[WIP] Reactive GraphQL queries: Declarative also comes with built-in support for GraphQL-like queries, for a more document-oriented usage model.

Pluggable sinks and sources: Declarative can be extended to read data from and write results back to external systems, such as Kafka or Datomic, as well as static sources such as csv files.

Pluggable frontends: Languages such as Datalog and SQL can be easily implemented on top of Declarative. Well, maybe not easily, but easier than without. A Datalog frontend is provided in Clojure(Script).

Thanks to Differential Dataflow, all these capabilities are provided within the dataflow model and can thus be scaled out to multiple independent workers. Declarative is less efficient and much more opinionated than hand-written Differential Dataflow. In particular, it enforces a fully-normalized, RDF-like data model heavily inspired by systems like Datomic or LogicBlox. Other than that, Declarative is just Differential Dataflow under the hood and can happily co-exist and interact with static, handwritten dataflows.

Included in this repository is the library itself, a server, and a cli.


The library is built using cargo. A sufficiently up-to-date Rust toolchain is enforced via the rust-toolchain file included in this repository.

Declarative Dataflow makes use of the log crate. Logging at a specific level can be enabled by setting the RUST_LOG environment variable to RUST_LOG=declarative_dataflow=<level>.


Crate documentation available on crates.io.

Important architectural decisions are documented in the docs/adr/ sub-directory.

Documentation for this package can be built via cargo doc --no-deps and viewed in a browser via cargo doc --no-deps --open. Please refer to declarative_dataflow::plan::Plan for documentation on the available operators. The tests/ directory contains usage examples.


Declarative Dataflow servers accept a lower-level relational query plan representation, that is meant to be somewhat frontend agnosti. However query plans are rather cumbersome to write manually and do not map to any interesting, higher-level semantics. Currently we therefore provide a Datalog front end written in Clojure.

Please refer to the documentation for an overview of the supported query plans.

Further Reading / Watching

A post on the high-level motivation for this project.

[video] Reactive Datalog For Datomic, recorded at Clojure/conj 2018.

[video] 3DF: Reactive Datalog For Datomic, recorded at :clojureD 2019.

The Clockworks blog has a number of posts on Declarative.


~83K SLoC