Awesome Open Source
Awesome Open Source

CI Twitter


What is Couler?

Couler aims to provide a unified interface for constructing and managing workflows on different workflow engines, such as Argo Workflows, Tekton Pipelines, and Apache Airflow.

Couler is included in CNCF Cloud Native Landscape and LF AI Landscape.

Note that while one of ambitious goals of Couler is to support multiple workflow engines, Couler currently only supports Argo Workflows as the workflow orchestration backend. In addition, if you are looking for a Python SDK that provides access to all the available features from Argo Workflows, you might want to check out the low-level Python SDK maintained by the Argo Workflows team.

Who uses Couler?

You can find a list of organizations who are using Couler in If you'd like to add your organization to the list, please send us a pull request.

Why use Couler?

Many workflow engines exist nowadays, e.g. Argo Workflows, Tekton Pipelines, and Apache Airflow. However, their programming experience varies and they have different level of abstractions that are often obscure and complex. The code snippets below are some examples for constructing workflows using Apache Airflow and Kubeflow Pipelines.

Apache Airflow Kubeflow Pipelines

def create_dag(dag_id,
    def hello_world_py(*args):
        print('Hello World')

    dag = DAG(dag_id,
    with dag:
        t1 = PythonOperator(
    return dag

for n in range(1, 10):
    default_args = {'owner': 'airflow',
                    'start_date': datetime(2018, 1, 1)
    globals()[dag_id] = create_dag(

class FlipCoinOp(dsl.ContainerOp):
    """Flip a coin and output heads or tails randomly."""
    def __init__(self):
        super(FlipCoinOp, self).__init__(
            command=['sh', '-c'],
            arguments=['python -c "import random; result = \'heads\' if random.randint(0,1) == 0 '
                       'else \'tails\'; print(result)" | tee /tmp/output'],
            file_outputs={'output': '/tmp/output'})

class PrintOp(dsl.ContainerOp):
    """Print a message."""
    def __init__(self, msg):
        super(PrintOp, self).__init__(
            command=['echo', msg],

# define the recursive operation
def flip_component(flip_result):
    print_flip = PrintOp(flip_result)
    flipA = FlipCoinOp().after(print_flip)
    with dsl.Condition(flipA.output == 'heads'):

    name='pipeline flip coin',
    description='shows how to use graph_component.'
def recursive():
    flipA = FlipCoinOp()
    flipB = FlipCoinOp()
    flip_loop = flip_component(flipA.output)
    PrintOp('cool, it is over. %s' % flipA.output).after(flip_loop)

Couler provides a unified interface for constructing and managing workflows that provides the following:

  • Simplicity: Unified interface and imperative programming style for defining workflows with automatic construction of directed acyclic graph (DAG).
  • Extensibility: Extensible to support various workflow engines.
  • Reusability: Reusable steps for tasks such as distributed training of machine learning models.
  • Efficiency: Automatic workflow and resource optimizations under the hood.

Please see the following sections for installation guide and examples.


  • Couler currently only supports Argo Workflows. Please see instructions here to install Argo Workflows on your Kubernetes cluster.
  • Install Python 3.6+
  • Install Couler Python SDK via the following command:
python3 -m pip install git+ --ignore-installed

Alternatively, you can clone this repository and then run the following to install:

python install

Try Couler with Argo Workflows

Click here to launch the interactive Katacoda environment and learn how to write and submit your first Argo workflow using Couler Python SDK in your browser!


Coin Flip

This example combines the use of a Python function result, along with conditionals, to take a dynamic path in the workflow. In this example, depending on the result of the first step defined in flip_coin(), the template will either run the heads() step or the tails() step.

Steps can be defined via either couler.run_script() for Python functions or couler.run_container() for containers. In addition, the conditional logic to decide whether to flip the coin in this example is defined via the combined use of couler.when() and couler.equal().

import couler.argo as couler
from couler.argo_submitter import ArgoSubmitter

def random_code():
    import random

    res = "heads" if random.randint(0, 1) == 0 else "tails"

def flip_coin():
    return couler.run_script(image="python:alpine3.6", source=random_code)

def heads():
    return couler.run_container(
        image="alpine:3.6", command=["sh", "-c", 'echo "it was heads"']

def tails():
    return couler.run_container(
        image="alpine:3.6", command=["sh", "-c", 'echo "it was tails"']

result = flip_coin()
couler.when(couler.equal(result, "heads"), lambda: heads())
couler.when(couler.equal(result, "tails"), lambda: tails())

submitter = ArgoSubmitter()


This example demonstrates different ways to define the workflow as a directed-acyclic graph (DAG) by specifying the dependencies of each task via couler.set_dependencies() and couler.dag(). Please see the code comments for the specific shape of DAG that we've defined in linear() and diamond().

import couler.argo as couler
from couler.argo_submitter import ArgoSubmitter

def job(name):

#     A
#    / \
#   B   C
#  /
# D
def linear():
    couler.set_dependencies(lambda: job(name="A"), dependencies=None)
    couler.set_dependencies(lambda: job(name="B"), dependencies=["A"])
    couler.set_dependencies(lambda: job(name="C"), dependencies=["A"])
    couler.set_dependencies(lambda: job(name="D"), dependencies=["B"])

#   A
#  / \
# B   C
#  \ /
#   D
def diamond():
            [lambda: job(name="A")],
            [lambda: job(name="A"), lambda: job(name="B")],  # A -> B
            [lambda: job(name="A"), lambda: job(name="C")],  # A -> C
            [lambda: job(name="B"), lambda: job(name="D")],  # B -> D
            [lambda: job(name="C"), lambda: job(name="D")],  # C -> D

submitter = ArgoSubmitter()

Note that the current version only works with Argo Workflows but we are actively working on the design of the unified interface that is extensible to additional workflow engines. Please stay tuned for more updates and we welcome any feedback and contributions from the community.

Community Blogs and Presentations

Alternatives To Couler
Select To Compare

Alternative Project Comparisons
Related Awesome Lists
Top Programming Languages
Top Projects

Get A Weekly Email With Trending Projects For These Topics
No Spam. Unsubscribe easily at any time.
Python (807,639
Machine Learning (37,132
Kubernetes (24,711
Pipeline (15,544
Airflow (1,067
Cloud Native (944
Distributed Computing (630
Argo (592
Workflow Engine (389
Apache Airflow (166
Workflow Management (131
Workflow Automation (122
Kubeflow (115
Tekton Pipelines (39
Argo Workflows (24
Unified Api (7
Unified Interface (3