Barca

Barca is a hybrid Rust+Python asset orchestrator. It discovers functions decorated with @asset(), @sensor(), and @task(), builds a dependency graph via static analysis (no imports), generates a phased execution plan, and dispatches work to Python workers — all in under 40ms of framework overhead.

from barca import asset

@asset()
def raw_data() -> list[dict]:
    return [{"x": 1}, {"x": 2}, {"x": 3}]

@asset(inputs={"data": raw_data})
def summary(data: list[dict]) -> dict:
    return {"count": len(data), "total": sum(d["x"] for d in data)}

barca get pipeline.py

No config files. No YAML. No daemon. Just functions and a fast binary.

Install

uv add barca

Static analysis, always

The Rust binary parses your Python source with ruff’s AST — it never imports your code to build the dependency graph.

Rust-speed planning

Parsing, DAG construction, and phased execution planning happen in compiled Rust — around 38ms of total framework overhead for a trivial asset.

Content-addressed caching

Every artifact is hashed and stored at {artifacts}/{node}/{run_hash}{ext}, so unchanged assets never re-run.

Artifact-based data passing

Results move between worker batches as serialized files — json, pickle, or parquet — detected and handled automatically.

How it works

1. Rust binary parses Python source using ruff’s AST (pure static analysis, no import)
2. Builds a petgraph DAG from decorator metadata
3. Generates a phased execution plan (phases run sequentially, streams within a phase run in parallel)
4. Maintains a pool of stateless Python workers and dispatches one task at a time via a Unix domain socket
5. Collects results and persists them to .barca/metadata.db (Turso/libSQL)

Node kinds

Kind	Decorator	Cached	Can be input
asset	@asset()	Yes	Yes
sensor	@sensor()	No (always re-runs)	Yes
task	@task()	No (always re-runs)	Yes (to other tasks)