AWS Big Data Blog

What’s new with Amazon MWAA support for startup scripts

Amazon Managed Workflow for Apache Airflow (Amazon MWAA) is a managed service for Apache Airflow that lets you use the same familiar Apache Airflow environment to orchestrate your workflows and enjoy improved scalability, availability, and security without the operational burden of having to manage the underlying infrastructure.

In April 2023, Amazon MWAA added support for shell launch scripts for environment versions Apache Airflow 2.x and later. With this feature, you can customize the Apache Airflow environment by launching a custom shell launch script at startup to work better with existing integration infrastructure and help with your compliance needs. You can use this shell launch script to install custom Linux runtimes, set environment variables, and update configuration files. Amazon MWAA runs this script during startup on every individual Apache Airflow component (worker, scheduler, and web server) before installing requirements and initializing the Apache Airflow process.

In this post, we provide an overview of the features, explore applicable use cases, detail the steps to use it, and provide additional facts on the capabilities of this shell launch script.

Solution overview

To run Apache Airflow, Amazon MWAA builds Amazon Elastic Container Registry (Amazon ECR) images that bundle Apache Airflow releases with other common binaries and Python libraries. These images then get used by the AWS Fargate containers in the Amazon MWAA environment. You can bring in additional libraries through the requirements.txt and files and pass the Amazon Simple Storage Service (Amazon S3) paths as a parameter during environment creation or update.

However, this method to install packages didn’t cover all of your use cases to tailor your Apache Airflow environments. Customers asked us for a way to customize the Apache Airflow container images by specifying custom libraries, runtimes, and supported files.

Applicable use cases

The new feature adds the ability to customize your Apache Airflow image by launching a custom specified shell launch script at startup. You can use the shell launch script to perform actions such as the following:

  • Install runtimes – Install or update Linux runtimes required by your workflows and connections. For example, you can install libaio as a custom library for Oracle.
  • Configure environment variables – Set environment variables for the Apache Airflow scheduler, web server, and worker components. You can overwrite common variables such as PATH, PYTHONPATH, and LD_LIBRARY_PATH. For example, you can set LD_LIBRARY_PATH to instruct Python to look for binaries in the paths that you specify.
  • Manage keys and tokens – Pass access tokens for your private PyPI/PEP-503 compliant custom repositories to requirements.txt and configure security keys.

How it works

The shell script runs Bash commands at startup, so you can install using yum and other tools similar to how Amazon Elastic Compute Cloud (Amazon EC2) offers user data and shell scripts support. You can define a custom shell script with the .sh extension and place it in the same S3 bucket as requirements.txt and You can define an S3 file version of the shell script during the environment creation or update via the Amazon MWAA console, API, or AWS Command Line Interface (AWS CLI). For details on how to configure the startup script, refer to Using a startup script with Amazon MWAA.

During the environment creation or update process, Amazon MWAA copies the, requirements.txt, shell script, and your Apache Airflow Directed Acrylic Graphs (DAGs) to the container images on the underlying Amazon Elastic Container Service (Amazon ECS) Fargate clusters. The Amazon MWAA instance extracts these contents and runs the startup script file that you specified. The startup script is run from the /usr/local/airflow/startup Apache Airflow directory as the airflow user. When it’s complete, the setup process will install the requirements.txt and files, followed by the Apache Airflow process associated with the container.

The following screenshot shows you the new optional Startup script file field on the Amazon MWAA console.

For monitoring and observability, you can view the output of the script in your Amazon MWAA environment’s Amazon CloudWatch log groups. To view the logs, you need to enable logging for the log group. If enabled, Amazon MWAA creates a new log stream starting with the prefix startup_script_exection_ip. You can retrieve log events to verify that the script is working as expected.

You can also use Amazon MWAA local-runner to test this feature on your local development environments. You can now specify your custom startup script in the startup_script directory in the local-runner. It’s recommended that you locally test your script before applying changes to your Amazon MWAA setup.

You can reference files that you package within or your DAGs folder from your startup script. This can be beneficial if you require installing Linux runtimes on a private web server from a local package. It’s also useful to be able to skip installation of Python libraries on a web server that doesn’t have access, either due to private web server mode or for libraries hosted on a private repository accessible only from your VPC, such as in the following example:

export ENVIRONMENT_STAGE=”development”

if [“${MWAA_AIRFLOW_COMPONENT} != “webserver”
pip3 install -r /usr/local/airflow/dags/requirements.txt

The MWAA_AIRFLOW_COMPONENT variable used in the script identifies each Apache Airflow scheduler, web server, and worker component that the script runs on.

Additional considerations

Keep in mind the following additional information of this feature:

  • Specifying a startup shell script file is optional. You can pick a specific S3 file version of your script.
  • Updating the startup script to an existing Amazon MWAA environment will lead to a restart of the environment. Amazon MWAA runs the startup script as each component in your environment restarts. Environment updates can take 10–30 minutes. We suggest using the Amazon MWAA local-runner to test and reduce the feedback loop.
  • You can make several changes to the Apache Airflow environment, such as setting non-reserved AIRFLOW__ environment variables and installing custom Python libraries. For a detailed list of reserved and unreserved environment variables that you can set or update, refer to Set environment variables using a startup script.
  • Upgrading Apache Airflow core libraries and dependencies or Python versions is not supported. This is because there are constraints used for the base Apache Airflow configuration in Amazon MWAA that will lead to version incompatibility with different installs of the Python runtime and dependent library versions. Amazon MWAA runs validations prior to your custom startup script run to prevent Python or Apache Airflow installs from including triggering workflows.
  • A failure during the startup script run results in an unsuccessful task stabilization of the underlying Amazon ECS Fargate containers. This can impact your Amazon MWAA environment’s ability to successfully create or update.
  • The startup script runtime is limited to 5 minutes, after which it will automatically time out.
  • To revert a startup script that is failing or is no longer required, edit your Amazon MWAA environment to reference a blank .sh file.


In this post, we talked about the new feature of Amazon MWAA that allows you to configure a startup shell launch script. This feature is supported on new and existing Amazon MWAA environments running Apache Airflow 2.x and above. Use this feature to install Linux runtimes, configure environment variables, and manage keys and tokens. You now have an additional option to customize your base Apache Airflow image to meet your specific needs.

For additional details and code examples on Amazon MWAA, visit the Amazon MWAA User Guide and the Amazon MWAA examples GitHub repo.

About the Authors

Parnab Basak is a Solutions Architect and a Serverless Specialist at AWS. He specializes in creating new solutions that are cloud native using modern software development practices like serverless, DevOps, and analytics. Parnab works closely in the analytics and integration services space helping customers adopt AWS services for their workflow orchestration needs.

Vishal Vijayvargiya is a Software Engineer working on Amazon MWAA at Amazon Web Services. He is passionate about building distributed and scalable software systems. Vishal also enjoys playing badminton and cricket.