Apache Liminals goal is to operationalise the machine learning process, allowing data scientists to quickly transition from a successful experiment to an automated pipeline of model training, validation, deployment and inference in production. Liminal provides a Domain Specific Language to build ML workflows on top of Apache Airflow.

Clone this repo:
  1. 57246ec update notice year 2023 (#100) by Lidor Ettinger · 1 year, 3 months ago master 0.0.5 0.0.5rc2-INCUBATING
  2. e569d89 fix k8s secret doc (#98) by Lidor Ettinger · 1 year, 4 months ago 0.0.5rc1-incubating
  3. 4979c3d Sagemaker example (#90) by Lidor Ettinger · 1 year, 5 months ago
  4. 3d5de7f [LIMINAL-105] add k8s secret cfg (#91) by Lidor Ettinger · 1 year, 5 months ago
  5. 95a6d57 Stream out docker-build cmd stdout by Zion Rubin · 1 year, 10 months ago

Apache Liminal

Apache Liminal is an end-to-end platform for data engineers & scientists, allowing them to build, train and deploy machine learning models in a robust and agile way.

The platform provides the abstractions and declarative capabilities for data extraction & feature engineering followed by model training and serving. Liminal's goal is to operationalize the machine learning process, allowing data scientists to quickly transition from a successful experiment to an automated pipeline of model training, validation, deployment and inference in production, freeing them from engineering and non-functional tasks, and allowing them to focus on machine learning code and artifacts.


Using simple YAML configuration, create your own schedule data pipelines (a sequence of tasks to perform), application servers, and more.

Getting Started

A simple getting stated guide for Liminal can be found here

Apache Liminal Documentation

Full documentation of Apache Liminal can be found here

High Level Architecture

High level architecture documentation can be found here

Example YAML config file

name: MyLiminalStack
owner: Bosco Albert Baracus
  - volume: myvol1
      path: /Users/me/myvol1
  - image: my_python_task_img
    type: python
    source: write_inputs
  - image: my_parallelized_python_task_img
    source: write_outputs
  - image: my_server_image
    type: python_server
    source: myserver
      - endpoint: /myendpoint1
        module: my_server
        function: myendpoint1func
  - pipeline: my_pipeline
    start_date: 1970-01-01
    timeout_minutes: 45
    schedule: 0 * 1 * *
      namespace: TestNamespace
      backends: [ 'cloudwatch' ]
      - task: my_python_task
        type: python
        description: static input task
        image: my_python_task_img
          NUM_FILES: 10
          NUM_SPLITS: 3
          - mount: mymount
            volume: myvol1
            path: /mnt/vol1
        cmd: python -u write_inputs.py
      - task: my_parallelized_python_task
        type: python
        description: parallelized python task
        image: my_parallelized_python_task_img
          FOO: BAR
        executors: 3
          - mount: mymount
            volume: myvol1
            path: /mnt/vol1
        cmd: python -u write_inputs.py
  - service: my_python_server
    description: my python server
    image: my_server_image


  1. Install this repository (HEAD)
   pip install git+https://github.com/apache/incubator-liminal.git
  1. Optional: set LIMINAL_HOME to path of your choice (if not set, will default to ~/liminal_home)
echo 'export LIMINAL_HOME=</path/to/some/folder>' >> ~/.bash_profile && source ~/.bash_profile

Authoring pipelines

This involves at minimum creating a single file called liminal.yml as in the example above.

If your pipeline requires custom python code to implement tasks, they should be organized like this

If your pipeline introduces imports of external packages which are not already a part of the liminal framework (i.e. you had to pip install them yourself), you need to also provide a requirements.txt in the root of your project.

Testing the pipeline locally

When your pipeline code is ready, you can test it by running it locally on your machine.

  1. Ensure you have The Docker engine running locally, and enable a local Kubernetes cluster:

Kubernetes configured

And allocate it at least 3 CPUs (under “Resources” in the Docker preference UI).

If you want to execute your pipeline on a remote kubernetes cluster, make sure the cluster is configured using:

kubectl config set-context <your remote kubernetes cluster>
  1. Build the docker images used by your pipeline.

In the example pipeline above, you can see that tasks and services have an “image” field - such as “my_static_input_task_image”. This means that the task is executed inside a docker container, and the docker container is created from a docker image where various code and libraries are installed.

You can take a look at what the build process looks like, e.g. here

In order for the images to be available for your pipeline, you'll need to build them locally:

cd </path/to/your/liminal/code>
liminal build

You'll see that a number of outputs indicating various docker images built.

  1. Create a kubernetes local volume
    In case your Yaml includes working with volumes please first run the following command:
cd </path/to/your/liminal/code>
liminal create
  1. Deploy the pipeline:
cd </path/to/your/liminal/code>
liminal deploy

Note: after upgrading liminal, it's recommended to issue the command

liminal deploy --clean

This will rebuild the airlfow docker containers from scratch with a fresh version of liminal, ensuring consistency.

  1. Start the server
liminal start
  1. Stop the server
liminal stop
  1. Display the server logs
liminal logs --follow/--tail

Number of lines to show from the end of the log:
liminal logs --tail=10

Follow log output:
liminal logs --follow
  1. Navigate to http://localhost:8080/admin

  2. You should see your pipeline The pipeline is scheduled to run according to the json schedule: 0 * 1 * * field in the .yml file you provided.

  3. To manually activate your pipeline:

  • Click your pipeline and then click “trigger DAG”
  • Click “Graph view” You should see the steps in your pipeline getting executed in “real time” by clicking “Refresh” periodically.

Pipeline activation


More information on contributing can be found here


The Liminal community holds a public call every Monday

Running Tests (for contributors)

When doing local development and running Liminal unit-tests, make sure to set LIMINAL_STAND_ALONE_MODE=True