a kedro hook to protect against breaking changes to data

steel-toes is a plugin for the python data pipelining framework kedro. It modifies each datasets filepath when you are developing a new feature to the pipeline, but do not want to wreck pipelines running on other branches.


kedro is a ✨ fantastic project that allows for super-fast prototyping of data pipelines, while yielding production-ready pipelines. kedro promotes collaborative projects by giving each team member access to the exact same data. Team members will often make their own branch of the project and begin work. Sometimes these changes will break existing functionality. Sometimes we make mistakes as we develop, and fix them before merging in. Either case can be detrimental to a teammate working downstream of your changes if not careful.

🥼 Wear the proper PPE during feature development

steel-toes hooks into your catalog to prevent changing downstream data on your teammates while developing in parallel.

on_catalog_created and before_pipeline_run

When your project creates a catalog steel-toes will look to see if branched data exists, if it does it will swap the filepath to the branched path. So you will be able to load the latest data from the perspective of any branch simulaneusly.


After your node is ran, before saving, steel-toes will check if your filepath was swapped, if not it will swap it to the branched filepath before saving.


steel-toes is deployed to pypi and can be pip installed.

pip install steel-toes

For a real kedro project you should add to your requirements.


To add SteelToes to your kedro>0.18.0 project add an instance of the SteelToes hook to your tuple of hooks in src/<project_name>/

from steel_toes import SteelToes

HOOKS = (SteelToes(),) location is typically located in src/<python_package>/


Some datasets have a _filepath attribute that is not meant for saving datasets to and is not needed to be "branched", and should be ignored from steel_toes, for example SQLQueryDataSet.

from kedro.extras.datasets.pandas.sql_dataset import SQLQueryDataSet, SQLTableDataSet
from steel_toes import SteelToes

HOOKS = (SteelToes(ignore_types=[SQLQueryDataSet, SQLTableDataSet]),)

Automatic branch naming

steel_toes will automatically get the branch name from your git branch. All you need to do is create a new branch, and steel-toes will make sure that all the data you write will go to a specific place for that branch. It will not change the filepaths until the dataset exists or just before its written, this way your catalog will still load existing datasets from the dataset specified in the catalog.

git checkout -b my-new-feature origin/main

Override with environment variable

In certain situations such as using kedro docker in production, there is no git branch to pull from. Setting an environment variable before steel-toes initializes will set the branch.

set environment variable in the shell


# run kedro here

set environment variable with python

import os

os.environ["STEEL_TOES_BRANCH"] = "PROD"

# run kedro here

Example filenames

Here is an example of what filepaths look like when I add parquet catalog entries to the spaceflights project, steel_toes will add the branch name automatically just before the file extension.

X_test: data/X_test_main.pq
X_train: data/X_train_main.pq
preprocessed_companies: data/02_intermediate/preprocessed_companies_main.pq
preprocessed_shuttles: data/02_intermediate/preprocessed_shuttles_main.pq
model_input_table: data/03_primary/model_input_table_main.pq
regressor: data/06_models/regressor_main.pickle

Logs on first run

When first running your pipeline with steel-toes it will start the _filepath swap after_node_run, since the swapped file does not yet exist.


At this point catalog.load('preprocessed_shuttles') will not load the branched dataset.

 kedro run
INFO     Kedro project spaceflights                                                     
INFO     STEEL_TOES:after_node_run 'preprocessed_shuttles.pq' -> 'preprocessed_shuttles_main.pq'
INFO     Completed 6 out of 6 tasks                                                     
INFO     Pipeline execution completed successfully.                                     

Logs after dataset exists

Subsequent runs of kedro will swap the dataset to the branched filepath immediately after the catalog has been created.


Now catalog.load('preprocessed_shuttles') will load the branched dataset.

INFO     Kedro project spaceflights                                                            
INFO     STEEL_TOES:after_catalog_created 'preprocessed_shuttles.pq' -> 'preprocessed_shuttles_main.pq'
INFO     Completed 6 out of 6 tasks                                                            
INFO     Pipeline execution completed successfully.                                            

CLI Usage

The CLI provides a handy interface to clean up your branched datasets.

steel-toes --help

Usage: steel-toes [OPTIONS] COMMAND [ARGS]...


  -V, --version  Prints version and exits
  --help         Show this message and exit.

  clean-branch  finds branch datasets and removes them

steel-toes also registers itself as a kedro global cli plugin. You can run kedro clean-branch to clean your branched data.

steel-toes clean-branch --help

Usage: kedro clean-branch [OPTIONS]

  finds branch datasets and removes them

  --dryrun                   Displays the files that would be deleted using
                             the specified command without actually deleting

  -b, --branch TEXT          git branch to clean files from
  -h, --help                 Show this message and exit.

Cleaning up old branches

To clean up your current branch, running kedro clean-branch will remove all the datasets that have been swapped to the current branch. Adding --dryrun will only log what steel-toes intends to do, and will not delete.

kedro clean-branch --dryrun

INFO     STEEL_TOES:after_catalog_created 'preprocessed_shuttles.pq' -> 'preprocessed_shuttles_main.pq'                               
INFO     STEEL_TOES:dryrun-remove | '/home/waylon/git/spaceflights/data/02_intermediate/preprocessed_shuttles_main.pq'                

Dropping the --dryrun flag will delete all the branched datasets.

kedro clean-branch

INFO     STEEL_TOES:after_catalog_created 'preprocessed_shuttles.pq' -> 'preprocessed_shuttles_main.pq'                               
INFO     STEEL_TOES:deleting | '/home/waylon/git/spaceflights/data/02_intermediate/preprocessed_shuttles_main.pq'                


You're Awesome for considering a contribution! Contributions are welcome, please check out the Contributing Guide for more information. Please be a positive member of the community and embrace feedback


We use SemVer for versioning. For the versions available, see the tags, or releases


Waylon Walker

Waylon Walker - Original Author


This project is licensed under the MIT License - see the LICENSE. file for details