Coffea-casa — An Analysis Facility for the Columnar Object Framework For Effective Analysis
About Coffea-casa
Coffea-casa is an analysis facility (AF) that provides services for "low latency columnar analysis", enabling rapid processing of HEP data in a column-wise fashion with coffea. Its services, based on Dask and Jupyter, aim to dramatically lower time to insight and provide an easily scalable, user-friendly computational environment.
The facility runs on Kubernetes (JupyterHub via Zero to JupyterHub) and scales Dask workers out onto dedicated resources and onto resources allocated via fair share through HTCondor (e.g. the Nebraska Tier-2). It also handles single sign-on, token-based authentication and authorization for data access (XRootD / XCache), and ships a curated software stack for analysis.
Production instances:
- coffea.casa — CMS AF at UNL (CMS AuthZ / IAM tokens)
- coffea-opendata.casa — open-data instance
- coffea.af.uchicago.edu — ATLAS AF at UChicago
User documentation: https://coffea-casa.readthedocs.io. Examples live in docs/gallery and in coffea-casa-tutorials.
What's in this repository
| Path | Contents |
|---|---|
coffea_casa/ |
coffea_casa Python package (CoffeaCasaCluster, a dask-jobqueue HTCondor cluster for the AF), published to PyPI |
docker/ |
Dockerfiles and environment files for the notebook/scheduler and worker images (details) |
charts/coffea-casa/ |
Helm chart (JupyterHub + coffea-casa configuration) |
docs/ |
Sphinx documentation (ReadTheDocs) and the example notebook gallery |
ci/, tests/, tests-charts/ |
Image structure tests, spawn/smoke tests, chart tests |
The coffea_casa Python package
pip install coffea-casa
Inside a coffea-casa notebook, the Dask JupyterLab extension starts a CoffeaCasaCluster for you. To create one manually:
from coffea_casa import CoffeaCasaCluster
from dask.distributed import Client
cluster = CoffeaCasaCluster() # no workers are submitted until you scale
cluster.adapt(minimum=0, maximum=20)
client = Client(cluster)
See the API docs for all options.
Docker images
Images are published to the OSG Harbor registry under hub.opensciencegrid.org/coffea-casa. They are built on top of the coffea base images (coffeateam/coffea-*-almalinux9):
| Image | Role |
|---|---|
cc-dask-alma9 |
Single-user notebook / Dask scheduler image (JupyterLab, Dask labextension, code-server, …) |
cc-analysis-alma9 |
Slim Dask worker image |
cc-dask-combine-alma9 |
cc-dask-alma9 + ROOT and CMS Combine |
cc-analysis-combine-alma9 |
cc-analysis-alma9 + ROOT and CMS Combine |
cc-dask-alma9 and cc-analysis-alma9 are built for several coffea flavours:
| Flavour | Base image | Python |
|---|---|---|
noml (default) |
coffeateam/coffea-almalinux9-noml |
3.12 |
dak |
coffeateam/coffea-dak-almalinux9 (coffea with dask-awkward) |
3.12 |
full |
coffeateam/coffea-almalinux9 |
3.12 |
0.7 |
coffeateam/coffea-0.7-almalinux9 (legacy coffea 0.7.x) |
3.11 |
Tags have the form <version>-<flavour> (e.g. cc-dask-alma9:2026.10.12-dak); the default noml flavour is also published under the bare <version> tag. Base image versions are bumped automatically by a daily workflow that opens a PR.
See docker/README.md for build arguments and how to build images locally.
Helm chart
helm repo add coffea-casa https://coffeateam.github.io/coffea-casa
helm repo update
helm search repo coffea-casa
See the chart README, the deployment guide and the local (k3d / minikube) deployment guide.
Versioning and releases
Coffea-casa uses calendar versioning (YYYY.MM.DD, with a .N suffix for further releases on the same day). Releasing is automated with GitHub Actions — there are no version numbers to bump by hand:
- Every push to
masteris automatically tagged with the next CalVer tag.- Images whose sources changed are published as
:development/:development-<flavour>. - The Helm chart is published to https://coffeateam.github.io/coffea-casa if
charts/changed. Since Helm requires SemVer, the chart version maps the tag toYYYY.MMDD.N(e.g.2026.10.12→2026.1012.0);appVersionis the exact git tag.
- Images whose sources changed are published as
- Publishing a GitHub Release from a tag (releases):
- builds, tests and publishes all images as
:<tag>/:<tag>-<flavour>; - publishes the
coffea_casapackage to PyPI (version derived from the tag); - republishes the Helm chart.
- builds, tests and publishes all images as
Pull requests build and test images and the chart without publishing anything.
Analysis repositories using coffea-casa
- BTVNanoCommissioning — commissioning studies in the CMS BTV POG based on (custom) NanoAOD samples
- SUEPCoffea_dask — SUEP analysis using fastjet with awkward input from PFNano NanoAOD samples
- hgg-coffea — tools for running the CMS Higgs to two photons analysis on NanoAOD
References
- M. Adamec, G. Attebury, K. Bloom, B. Bockelman, C. Lundstedt, O. Shadura and J. Thiltges, Coffea-casa: an analysis facility prototype, arXiv:2103.01871 (2021).
- PyHEP 2020 coffea-casa proceedings:
- PyHEP 2020 introductory talk: YouTube
Contact us
Questions and feedback are welcome in GitHub Discussions, GitHub Issues, or the IRIS-HEP Slack.
1 maintainer
Metadata
1 maintainer
Release files for coffea-casa 2026.10.13.7
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | |
|---|---|---|---|
| coffea_casa-2026.10.13.7.tar.gz | 8.8 MB | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| coffea_casa-2026.10.13.7-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 8.8 MB
Release files / coffea_casa-2026.10.13.7.tar.gz
| Download URL | coffea_casa-2026.10.13.7.tar.gz |
|---|---|
| Size | 8.8 MB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
f90b14f2e64d65c071eb04b874502fc8c29e63255a636c3eaf4537faa38a4eff
|
|
BLAKE2b-256 checksum How to use checksums |
809ae41559bb8c6c15df0298984cd394a9fd8c724efee328c22a30906f2a02a8
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Oct 13, 2026.
Transparency logRelease files / coffea_casa-2026.10.13.7-py3-none-any.whl
| Download URL | coffea_casa-2026.10.13.7-py3-none-any.whl |
|---|---|
| Size | 17.4 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
57bdd62cf5fb7741d441230e99de5f49c61582689fc248ccfb341f50eab46894
|
|
BLAKE2b-256 checksum How to use checksums |
e52fa9a7fae5d9fe700c0075359c692859ce6f210e20973230b05ae35ac95ed0
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Oct 13, 2026.
Transparency log
