Barca vs Dagster

Last measured: 2026-10-21, with barca 0.28.0 and Dagster 1.13.26, both from PyPI with uv, on Linux (AMD Ryzen AI MAX+ 395). barca ran under Python 3.13; the Dagster project scaffold chose Python 3.14. Setup and commands are in benchmarks/RESULTS.md. It replaces the first comparison, made on 2026-06-10 with barca 0.2.0 and Dagster 1.13.8 on macOS.

Install footprint

Metric	barca 0.28.0	Dagster 1.13.26	Dagster scaffold
Packages installed	1	48	98
Venv size (`du -sm`)	31 MB	81 MB	184 MB
Lockfile entries	2	53	112
Wheel download	13.9 MB	2.0 MB, plus its dependencies

“Dagster 1.13.26” is uv pip install dagster==1.13.26 into an empty venv. “Dagster scaffold” is the project that uvx create-dagster@1.13.26 project myproj --uv-sync creates, which also installs dagster-dg-cli and dagster-webserver. The barca wheel (manylinux x86_64) contains the Rust binary, the web UI that barca serve shows and the Python decorator stubs, and declares no dependencies.

python -c "import dagster" took 365 ms (median of 20), python -c "import barca" 42 ms, and python -c pass 13 ms.

Steps to first output

Barca

uv add barca                    # install
cat > pipeline.py << 'PY'       # write code
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)}
PY
barca get pipeline.py            # run

Dagster

uvx create-dagster@latest project myproj --uv-sync   # scaffold and install
cd myproj && source .venv/bin/activate               # enter project
# write src/myproj/defs/assets.py                    # code
dg launch --assets raw_data,summary                  # run

import dagster as dg

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

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

With --uv-sync, create-dagster did not prompt. The scaffold’s definitions.py loads every module in defs/, so the new file needed no registration. On 2026-06-10 the quickstart also had the reader install pandas and create a data directory and a CSV file; the pipeline above does not need them.

Running the same two-asset pipeline

Metric	barca 0.28.0	Dagster 1.13.26
Command	`barca get pipeline.py`	`dg launch --assets raw_data,summary`
Time, nothing cached	73 ms (67–76)	3.64 s (3.45–3.94)
Time, second run	13 ms (13–15), 0 steps run	not measured
Output lines	1 JSON line on stdout, 2 progress lines on stderr	29 DEBUG lines on stderr
Result access	artifact path in the JSON; `-o value` prints the value	pickled to the instance’s storage directory
Error output	1 JSON line on stdout, 4 lines on stderr	104 lines on stderr

Times are medians of 10 runs, the two tools taking turns run by run; the range is the fastest and slowest run. barca’s first time deletes .barca/ before each run. Dagster wrote its run to a DAGSTER_HOME without a dagster.yaml, and the virtualenv was not activated; that adds a one-line instance notice, a seven-line virtualenv warning and blank lines to the 29 DEBUG lines, 38 lines in all.

Barca output

[barca] importing pipeline.py
[barca] 2/2 steps | done in 0.0s
{"elapsed_seconds":0.0609,"final_output":{"_barca_artifact":{"format":"json","path":".barca/artifacts/pipeline.py--summary/2d10c4….json","size_bytes":24}},"phases":1,"run_id":"…","status":"success","steps":[…],"steps_executed":2,"warnings":[]}

The steps array holds each step’s id, run hash, cache status and time. The current shape is in the CLI contract.

Dagster output (abridged from 29 lines)

… - dagster - DEBUG - __ASSET_JOB - … - RUN_START - Started execution of run for "__ASSET_JOB".
… - dagster - DEBUG - __ASSET_JOB - … - ENGINE_EVENT - Executing steps using multiprocess executor
… - dagster - DEBUG - __ASSET_JOB - … - raw_data - STEP_WORKER_STARTING - Launching subprocess
… (12 more lines for raw_data) …
… - dagster - DEBUG - __ASSET_JOB - … - summary - STEP_WORKER_STARTING - Launching subprocess
… (12 more lines for summary) …
… - dagster - DEBUG - __ASSET_JOB - … - ENGINE_EVENT - Multiprocess executor: parent process exiting after 2.81s
… - dagster - DEBUG - __ASSET_JOB - … - RUN_SUCCESS - Finished execution of run for "__ASSET_JOB".

Error output

The same asset raising ValueError("something went wrong").

Barca 0.28.0

Exit code 1. On stdout, one JSON object with "status": "failed", "error": "ValueError: something went wrong" and "failed_node": "broken.py:oops". On stderr:

[barca] importing broken.py
[barca] 0/1 steps | failed in 0.0s
{"artifact_dir":".barca/artifacts/broken.py--oops","code":1,"error":"step 'broken.py:oops' failed: ValueError: something went wrong","kind":"step_failed","node":"broken.py:oops","remediation":"Fix the error in 'broken.py:oops' (see the traceback) and re-run the same command. …","traceback":"  File \"broken.py\", line 6, in oops\n    raise ValueError(\"something went wrong\")"}
[barca] run failed: step 'broken.py:oops' failed: ValueError: something went wrong; 0 downstream steps skipped (exit 1)

The traceback holds only the frames from the user’s file.

Dagster 1.13.26

Exit code 1 and 104 lines on stderr, ending in Error: Materialization failed. They hold three Stack Trace: blocks. The first:

dagster._core.errors.DagsterExecutionStepExecutionError: Error occurred while executing op "oops"::

ValueError: something went wrong

Stack Trace:
  File ".../dagster/_core/execution/plan/utils.py", line 57, in op_execution_error_boundary
    yield
  File ".../dagster/_utils/__init__.py", line 394, in iterate_with_context
    next_output = next(iterator)
  File ".../dagster/_core/execution/plan/compute_generator.py", line 136, in _coerce_op_compute_fn_to_iterator
    ...
  File ".../dagster/_core/execution/plan/compute_generator.py", line 116, in invoke_compute_fn
    ...
  File ".../src/myproj/defs/broken.py", line 6, in oops
    raise ValueError("something went wrong")

The frame from the user’s file is the last one, below the Dagster frames.

Other things observed on 2026-06-10

These are notes from using Dagster 1.13.8 on 2026-06-10. They have not been checked against a later Dagster release, except where a note dated 2026-10-21 says so.

dg list defs printed a table of every asset with its deps, group and kind, without running anything. barca’s equivalent is barca list, added in 0.2.1.
The Dagster scaffold included a .gitignore covering the files Dagster writes. barca 0.2.0 created .barca/ without one. Note, 2026-10-07: barca 0.18.0 writes .barca/.gitignore containing *, so the directory ignores itself.
A .py file placed in the scaffold’s defs/ folder was picked up without registration. barca 0.2.0 needed the files named on the command line. Note, 2026-10-07: barca 0.18.0 reads every .py file under the project root that imports barca when no file is named; see Discovery.
The Dagster project needed a pyproject.toml, a src layout, definitions.py and a defs/ folder before dg launch ran. barca get file.py runs on a single file.
Each asset printed 14 DEBUG lines per run. No option to reduce this was found during the test.
create-dagster asked “Run uv sync? (y/n)”. No --yes flag was found during the test. Note, 2026-10-21: create-dagster@1.13.26 project myproj --uv-sync ran without asking.
dg commands printed a five-line warning when the project’s virtualenv was not activated, including when run as .venv/bin/dg. Note, 2026-10-21: in 1.13.26 the warning is seven lines and says that adding project_and_activated_venv_mismatch to cli.suppress_warnings turns it off.
Both a dagster command and a dg command were installed.