Skip to main content

Python package to extend Airflow functionality with CWL v1.0 support

Project description

Build Status - Travis CI
Build Status - CWL conformance tests

cwl-airflow

Python package to extend Apache-Airflow 1.9.0 functionality with CWL v1.0 support.

Check it out

(assuming that you already have installed and properly configured python, latest pip, latest setuptools and docker that has access to pull images from the DockerHub; if something is missing or should be updated refer to Installation or Troubleshooting sections)

  1. Install cwl-airflow
    $ pip install cwl-airflow --find-links https://michael-kotliar.github.io/cwl-airflow-wheels/
    
  2. Init configuration
    $ cwl-airflow init
    
  3. Run demo
    $ cwl-airflow demo --auto
    
  4. When all demo wokrflows are submitted program will provide you with the link for Airflow Webserver. It may take some time (usually less then half a minute) for Airflow Webserver to load and display all the data

Airflow Webserver example


Table of Contents


How It Works

Key concepts

  1. CWL descriptor file - YAML or JSON file to describe the workflow inputs, outputs and steps. File should be compatible with CWL v1.0 specification
  2. Job file - YAML or JSON file to set the values for the wokrflow inputs. For cwl-airflow to function properly the Job file should include 3 mandatory and one optional fields:
    • workflow - mandatory field to specify the absolute path to the CWL descriptor file
    • output_folder - mandatory field to specify the absolute path to the folder where all the output files should be moved after successful workflow execution
    • tmp_folder - optional field to specify the absolute path to the folder for storing intermediate results. After workflow execution this folder will be deleted.
    • uid - mandatory field that is used for generating DAG's unique identifier.
  3. DAG - directed acyclic graph that describes the workflow structure.
  4. Jobs folder - folder to keep all Job files scheduled for execution or the ones that have already been processed. The folder's location is set as jobs parameter of cwl section in Airflow configuration file.

What's inside

To build a workflow cwl-airflow uses three basic classes:

  • CWLStepOperator - executes a separate workflow step
  • JobDispatcher - serializes the Job file and provides the worflow with input data
  • JobCleanup - returns the calculated results to the output folder

A set of CWLStepOperators, JobDispatcher and JobCleanup are combined in CWLDAG that defines a graph to reflect the workflow steps, their relationships and dependencies. Automatically generated cwl_dag.py script is placed in the DAGs folder. When Airflow Scheduler loads DAGs from the DAGs folder, the cwl_dag.py script parses all the Job files from the Jobs folder and creates DAGs for each of them. Each DAG has a unique DAG ID that is formed accodring to the following scheme: CWL descriptor file basename-Job file basename-uid field from the Job file

Airflow Webserver example


Installation

Requirements

Ubuntu 16.04.4 (Xenial Xerus)

  • python 2.7 or 3.5 (tested on the system Python 2.7.12 and 3.5.2)
  • docker
    sudo apt-get update
    sudo apt-get install apt-transport-https ca-certificates curl software-properties-common
    curl -fsSL https://download.docker.com/linux/ubuntu/gpg | sudo apt-key add -
    sudo add-apt-repository "deb [arch=amd64] https://download.docker.com/linux/ubuntu $(lsb_release -cs) stable"
    sudo apt-get update
    sudo apt-get install docker-ce
    sudo groupadd docker
    sudo usermod -aG docker $USER
    
    Log out and log back in so that your group membership is re-evaluated.
  • python-dev (or python3-dev if using Python 3.5)
    sudo apt-get install python-dev # python3-dev
    
    python-dev is required in case your system needs to compile some python packages during the installation. We have built python wheels for most of such packages and provided them through --find-links argument while installing cwl-airflow. Nevertheless in case of installation problems you might still be required to install this dependency.

macOS 10.13.5 (High Sierra)

  • python 2.7 or 3.6 (tested on the system Python 2.7.10 and brewed Python 2.7.15 / 3.6.5; 3.7.0 is not supported)
  • docker (follow the link to install Docker on Mac)
  • Apple Command Line Tools
    xcode-select --install
    
    Click Install on the pop up when it appears, follow the instructions. Apple Command Line Tools are required in case your system needs to compile some python packages during the installation. We have built python wheels for most of such packages and provided them through --find-links argument while installing cwl-airflow. Nevertheless in case of installation problems you might still be required to install this dependency.

Both Ubuntu and macOS

  • pip (tested on pip 18.0)

    wget https://bootstrap.pypa.io/get-pip.py
    python get-pip.py # --user
    
  • setuptools (tested on setuptools 40.0.0)

    pip install -U setuptools # --user
    

    --user - optional parameter to install all the packages into your HOME directory instead of the system Python directories. It will be helpful if you don't have enough permissions to install new Python packages. You might also need to update your PATH variable in order to have access to the installed packages (an easy way to do it is described in Troubleshooting section). If installing on macOS brewed Python --user should not be used (explained here)

Install cwl-airflow

$ pip install cwl-airflow --find-links https://michael-kotliar.github.io/cwl-airflow-wheels/ # --user

--find-links - using pre-compiled wheels from This repository allows to avoid installing Xcode for macOS users and python[3]-dev for Ubuntu users

--user - optional parameter to install all the packages into your HOME directory instead of the system Python directories. It will be helpful if you don't have enough permissions to install new Python packages. You might also need to update your PATH variable in order to have access to the installed packages (an easy way to do it is described in Troubleshooting section). If installing on macOS brewed Python --user should not be used (explained here)


Using cwl-airflow

Configuration

Before using cwl-airflow it should be initialized with the default configuration by running the command

$ cwl-airflow init

Optional parameters:

Flag Description Default
-l number of processed jobs kept in history, int 10 x Running, 10 x Success, 10 x Failed
-j path to the folder where all the new jobs will be added, str ~/airflow/jobs
-t timeout for importing all the DAGs from the DAG folder, sec 30
-r webserver workers refresh interval, sec 30
-w number of webserver workers to be refreshed at the same time, int 1
-p number of threads for Airflow Scheduler, int 2

Consider using -r 5 -w 4 to make Airflow Webserver react faster on all newly created DAGs

If you update Airflow configuration file manually (default location is ~/airflow/airflow.cfg), make sure to run cwl-airflow init command to apply all the changes, especially if core/dags_folder or cwl/jobs parameters from the configuration file are changed.

Submitting new job

To submit new CWL descriptor and Job files to cwl-airflow run the following command

cwl-airflow submit WORKFLOW JOB

Optional parameters:

Flag Description Default
-o path to the folder where all the output files should be moved after successful workflow execution, str current directory
-t path to the temporary folder for storing intermediate results, str /tmp
-u ID for DAG's unique identifier generation, str random uuid
-r run submitted workflow with Airflow Scheduler, bool False

Arguments -o, -t and -u doesn't overwrite the values from the Job file set in the fields output_folder, tmp_folder and uid correspondingly. The meaning of these fields is explaned in Key concepts section.

The submit command will resolve all relative paths from Job file adding mandatory fields workflow, output_folder and uid (if not provided) and will copy Job file to the Jobs folder. The CWL descriptor file and all input files referenced in the Job file should not be moved or deleted while workflow is running. The submit command will not execute submitted workflow unless -r argument is provided. Otherwise, make sure that Airflow Scheduler (and optionally Airflow Webserver) is running.

Depending on your Airflow configuration it may require some time for Airflow Scheduler and Webserver to pick up new DAGs. Consider using cwl-airflow init -r 5 -w 4 to make Airflow Webserver react faster on all newly created DAGs.

To start Airflow Scheduler (don't run it if cwl-airflow submit is used with -r argument)

airflow scheduler

To start Airflow Webserver

airflow webserver

Please note that both Airflow Scheduler and Webserver can be adjusted through the configuration file (default location is ~/airflow/airflow.cfg). Refer to the official documentation here

Demo mode

  • To get the list of the available demo workflows

    $ cwl-airflow demo --list
    
  • To submit the specific demo workflow from the list (workflow will not be run until Airflow Scheduler is started separately)

    $ cwl-airflow demo super-enhancer.cwl
    

    Depending on your Airflow configuration it may require some time for Airflow Scheduler and Webserver to pick up new DAGs. Consider using cwl-airflow init -r 5 -w 4 to make Airflow Webserver react faster on all newly created DAGs.

  • To submit all demo workflows from the list (workflows will not be run until Airflow Scheduler is started separately)

    $ cwl-airflow demo --manual
    

    Before submitting demo workflows the Jobs folder will be automatically cleaned.

  • To execute all available demo workflows (automatically starts Airflow Scheduler and Airflow Webserver)

    $ cwl-airflow demo --auto
    

    Before submitting and running demo workflows the Jobs folder will be automatically cleaned.

Optional parameters:

Flag Description Default
-o path to the folder where all the output files should be moved after successful workflow execution, str current directory
-t path to the temporary folder for storing intermediate results, str /tmp
-u ID for DAG's unique identifier generation, str. Ignored when --list or --auto is used random uuid

Troubleshooting

Most of the problems are already handled by cwl-airflow itself. User is provided with the full explanation and ways to correct them through the console output. Additional information regarding the failed workflow steps, can be found in the task execution logs that are accessible through Airflow Webserver UI.

Common errors and ways to fix them

  • cwl-airflow is not found

    Perhaps, you have installed it with --user option and your PATH variable doesn't include your user based Python bin folder. Update PATH with the following command

    export PATH="$PATH:`python -m site --user-base`/bin"
    
  • Fails to install on the latest Python 3.7.0

    Unfortunatelly Apache-Airflow 1.9.0 cannot be properly installed on the latest Python 3.7.0. Consider using Python 3.6 or 2.7 instead.

    macOS users can install Python 3.6.5 (instead of the latest Python 3.7.0) with the following command (explained here)

      brew install https://raw.githubusercontent.com/Homebrew/homebrew-core/f2a764ef944b1080be64bd88dca9a1d80130c558/Formula/python.rb
    
  • Fails to compile ruamel.yaml

    Perhaps, you should update your setuptools. Consider using --user if necessary. If installing on macOS brewed Python --user should not be used (explained here)

    pip install -U setuptools # --user
    
  • Docker is unable to pull images from the Internet.

    If you are using proxy, your Docker should be configured properly too. Refer to the official documentation

  • Docker is unable to mount directory.

    For macOS docker has a list of directories that it's allowed to mount by default. If your input files are located in the directories that are not included in this list, you are better of either changing the location of input files and updating your Job file or adding this directories into Docker configuration Preferences / File Sharing.

  • Airflow Webserver displays missing DAGs

    If some of the Job files have been manually deleted, they will be still present in Airflow database, hence they will be displayed in Webserver's UI. Sometimes you may still see missing DAGs because of the inertness of Airflow Webserver UI.

  • Airflow Webserver randomly fails to display some of the pages

    When new DAG is added Airflow Webserver and Scheduler require some time to update their states. Consider using cwl-airflow init -r 5 -w 4 to make Airflow Webserver react faster for all newly created DAGs. Or manualy update Airflow configuration file (default location is ~/airflow/airflow.cfg) and restart both Webserver and Scheduler. Refer to the official documentation here

  • Workflow execution fails

    Make sure that CWL descriptor and Job files are correct. You can always check them with cwltool (trusted version 1.0.20180622214234)

    cwltool --debug WORKFLOW JOB
    

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for cwl-airflow, version 1.0.13
Filename, size File type Python version Upload date Hashes
Filename, size cwl-airflow-1.0.13.tar.gz (6.2 MB) File type Source Python version None Upload date Hashes View hashes

Supported by

Elastic Elastic Search Pingdom Pingdom Monitoring Google Google BigQuery Sentry Sentry Error logging AWS AWS Cloud computing DataDog DataDog Monitoring Fastly Fastly CDN DigiCert DigiCert EV certificate StatusPage StatusPage Status page