Finest practices for migrating from Apache Airflow 2.x to Apache Airflow 3.x on Amazon MWAA

Apache Airflow 3.x on Amazon MWAA introduces architectural enhancements corresponding to API-based job execution that gives enhanced safety and isolation. Different main updates embody a redesigned UI for higher consumer expertise, scheduler-based backfills for improved efficiency, and assist for Python 3.12. Not like in-place minor Airflow model upgrades in Amazon MWAA, upgrading to Airflow 3 from Airflow 2 requires cautious planning and execution via a migration strategy as a result of elementary breaking modifications.

This migration presents a possibility to embrace next-generation workflow orchestration capabilities whereas offering enterprise continuity. Nevertheless, it’s greater than a easy improve. Organizations migrating to Airflow 3.x on Amazon MWAA should perceive key breaking modifications, together with the elimination of direct metadata database entry from employees, deprecation of SubDAGs, modifications to default scheduling conduct, and library dependency updates. This publish offers greatest practices and a streamlined strategy to efficiently navigate this crucial migration, offering minimal disruption to your mission-critical knowledge pipelines whereas maximizing the improved capabilities of Airflow 3.

Understanding the migration course of

The journey from Airflow 2.x to three.x on Amazon MWAA introduces a number of elementary modifications that organizations should perceive earlier than starting their migration. These modifications have an effect on core workflow operations and require cautious planning to attain a clean transition.

Try to be conscious of the next breaking modifications:

• Removing of direct database entry – A crucial change in Airflow 3 is the elimination of direct metadata database entry from employee nodes. Duties and customized operators should now talk via the REST API as a substitute of direct database connections. This architectural change impacts code that beforehand accessed the metadata database straight via SQLAlchemy connections, requiring refactoring of current DAGs and customized operators.
• SubDAG deprecation – Airflow 3 removes the SubDAG assemble in favor of TaskGroups, Belongings, and Knowledge Conscious Scheduling. Organizations should refactor current SubDAGs to one of many beforehand talked about constructs.
• Scheduling conduct modifications – Two notable modifications to default scheduling choices require an affect evaluation:
  • The default values for catchup_by_default and create_cron_data_intervals modified to False. This transformation impacts DAGs that don’t explicitly set these choices.
  • Airflow 3 removes a number of context variables, corresponding to execution_date, tomorrow_ds, yesterday_ds, prev_ds, and next_ds. It’s essential to exchange these variables with at the moment supported context variables.
• Library and dependency modifications – A big variety of libraries change in Airflow 3.x, requiring DAG code refactoring. Many beforehand included supplier packages may want specific addition to the necessities.txt file.
• REST API modifications – The REST API path modifications from /api/v1 to /api/v2, affecting exterior integrations. For extra details about utilizing the Airflow REST API, see Creating an internet server session token and calling the Apache Airflow REST API.
• Authentication system – Though Airflow 3.0.1 and later variations default to SimpleAuthManager as a substitute of Flask-AppBuilder, Amazon MWAA will proceed utilizing Flask-AppBuilder for Airflow 3.x. This implies clients on Amazon MWAA is not going to see any authentication modifications.

The migration requires creating a brand new setting slightly than performing an in-place improve. Though this strategy calls for extra planning and sources, it offers the benefit of sustaining your current setting as a fallback possibility throughout the transition, facilitating enterprise continuity all through the migration course of.

Pre-migration planning and evaluation

Profitable migration depends upon thorough planning and evaluation of your present setting. This section establishes the muse for a clean transition by figuring out dependencies, configurations, and potential compatibility points. Consider your setting and code in opposition to the beforehand talked about breaking modifications to have a profitable migration.

Surroundings evaluation

Start by conducting an entire stock of your present Amazon MWAA setting. Doc all DAGs, customized operators, plugins, and dependencies, together with their particular variations and configurations. Ensure that your present setting is on model 2.10.x, as a result of this offers the most effective compatibility path for upgrading to Amazon MWAA with Airflow 3.x.

Determine the construction of the Amazon Easy Storage Service (Amazon S3) bucket containing your DAG code, necessities file, startup script, and plugins. You’ll replicate this construction in a brand new bucket for the brand new setting. Creating separate buckets for every setting avoids conflicts and permits continued growth with out affecting present pipelines.

Configuration documentation

Doc all customized Amazon MWAA setting variables, Airflow connections, and setting configurations. Assessment AWS Id and Entry Administration (IAM) sources, as a result of your new setting’s execution function will want equivalent insurance policies. IAM customers or roles accessing the Airflow UI require the CreateWebLoginToken permission for the brand new setting.

Pipeline dependencies

Understanding pipeline dependencies is crucial for a profitable phased migration. Determine interdependencies via Datasets (now Belongings), SubDAGs, TriggerDagRun operators, or exterior API interactions. Develop your migration plan round these dependencies so associated DAGs can migrate on the similar time.

Think about DAG scheduling frequency when planning migration waves. DAGs with longer intervals between runs present bigger migration home windows and decrease danger of duplicate execution in contrast with continuously operating DAGs.

Testing technique

Create your testing technique by defining a scientific strategy to figuring out compatibility points. Use the ruff linter with the AIR30 ruleset to mechanically establish code requiring updates:

ruff test --preview --select AIR30 

Then, evaluation and replace your setting’s necessities.txt file to verify package deal variations adjust to the up to date constraints file. Moreover, generally used Operators beforehand included within the airflow-core package deal now reside in a separate package deal and have to be added to your necessities file.

Check your DAGs utilizing the Amazon MWAA Docker pictures for Airflow 3.x. These pictures make it doable to create and check your necessities file, and ensure the Scheduler efficiently parses your DAGs.

Migration technique and greatest practices

A methodical migration strategy minimizes danger whereas offering clear validation checkpoints. The beneficial technique employs a phased blue/inexperienced deployment mannequin that gives dependable migrations and speedy rollback capabilities.

Phased migration strategy

The next migration phases can help you in defining your migration plan:

• Part 1: Discovery, evaluation, and planning – On this section, full your setting stock, dependency mapping, and breaking change evaluation. With the gathered info, develop the detailed migration plan. This plan will embody steps for updating code, updating your necessities file, making a check setting, testing, creating the blue/inexperienced setting (mentioned later on this publish), and the migration steps. Planning should additionally embody the coaching, monitoring technique, rollback circumstances, and the rollback plan.
• Part 2: Pilot migration – The pilot migration section serves to validate your detailed migration plan in a managed setting with a small vary of affect. Focus the pilot on two or three non-critical DAGs with numerous traits, corresponding to completely different schedules and dependencies. Migrate the chosen DAGs utilizing the migration plan outlined within the earlier section. Use this section to validate your plan and monitoring instruments, and modify each based mostly on precise outcomes. In the course of the pilot, set up baseline migration metrics to assist predict the efficiency of the complete migration.
• Part 3: Wave-based manufacturing migration – After a profitable pilot, you might be prepared to start the complete wave-based migration for the remaining DAGs. Group remaining DAGs into logical waves based mostly on enterprise criticality (least crucial first), technical complexity, interdependencies (migrate dependent DAGs collectively), and scheduling frequency (much less frequent DAGs present bigger migration home windows). After you outline the waves, work with stakeholders to develop the wave schedule. Embody ample validation intervals between waves to verify the wave is profitable earlier than beginning the subsequent wave. This time additionally reduces the vary of affect within the occasion of a migration subject, and offers ample time to carry out a rollback.
• Part 4: Publish-migration evaluation and decommissioning – In any case waves are full, conduct a post-migration evaluation to establish classes discovered, optimization alternatives, and another unresolved gadgets. That is additionally a superb time to supply an approval on system stability. The ultimate step is decommissioning the unique Airflow 2.x setting. After stability is decided, based mostly on enterprise necessities and enter, decommission the unique (blue) setting.

Blue/inexperienced deployment technique

Implement a blue/inexperienced deployment technique for protected, reversible migration. With this technique, you should have two Amazon MWAA environments working throughout the migration and handle which DAGs function wherein setting.

The blue setting (present Airflow 2.x) maintains manufacturing workloads throughout transition. You’ll be able to implement a freeze window for DAG modifications earlier than migration to keep away from last-minute code conflicts. This setting serves because the speedy rollback setting if a difficulty is recognized within the new (inexperienced) setting.

The inexperienced setting (new Airflow 3.x) receives migrated DAGs in managed waves. It mirrors the networking, IAM roles, and safety configurations from the blue setting. Configure this setting with the identical choices because the blue setting, and create equivalent monitoring mechanisms so each environments might be monitored concurrently. To keep away from duplicate DAG runs, ensure that a DAG solely runs in a single setting. This entails pausing the DAG within the blue setting earlier than activating the DAG within the inexperienced setting.Keep the blue setting in heat standby mode throughout all the migration. Doc particular rollback steps for every migration wave, and check your rollback process for not less than one non-critical DAG. Moreover, outline clear standards for triggering the rollback (corresponding to particular failure charges or SLA violations).

Step-by-step migration course of

This part offers detailed steps for conducting the migration.

Pre-migration evaluation and preparation

Earlier than initiating the migration course of, conduct a radical evaluation of your present setting and develop the migration plan:

• Ensure that your present Amazon MWAA setting is on model 2.10.x
• Create an in depth stock of your DAGs, customized operators, and plugins together with their dependencies and variations
• Assessment your present necessities.txt file to know package deal necessities
• Doc all setting variables, connections, and configuration settings
• Assessment the Apache Airflow 3.x launch notes to know breaking modifications
• Decide your migration success standards, rollback circumstances, and rollback plan
• Determine a small variety of DAGs appropriate for the pilot migration
• Develop a plan to coach, or familiarize, Amazon MWAA customers on Airflow 3

Compatibility checks

Figuring out compatibility points is crucial to a profitable migration. This step helps builders deal with particular code that’s incompatible with Airflow 3.

Use the ruff linter with the AIR30 ruleset to mechanically establish code requiring updates:

ruff test --preview --select AIR30 

Moreover, evaluation your code for situations of direct metadatabase entry.

DAG code updates

Primarily based in your findings throughout compatibility testing, replace the affected DAG code for Airflow 3.x. The ruff DAG test utility can mechanically repair frequent modifications. Use the next command to run the utility in replace mode:

ruff test dag/ --select AIR301 --fix –preview

Frequent modifications embody:

• Change direct metadata database entry with API calls:

  # Earlier than (Airflow 2.x) - Direct DB entry
  from airflow.settings import Session
  from airflow.fashions.taskInstance import TaskInstance
  session=Session()
  end result=session.question(TaskInstance)
  
  For Apache Airflow v3.x, make the most of  within the Amazon MWAA SDK.
  Replace core assemble imports with the brand new Airflow SDK namespace:
  # Earlier than (Airflow 2.x)
  from airflow.decorators import dag, job
  
  # After (Airflow 3.x)
  from airflow.sdk import dag, job

• Change deprecated context variables with their trendy equivalents:

  # Earlier than (Airflow 2.x)
  def my_task(execution_date, **context):
      # Utilizing execution_date
  
  # After (Airflow 3.x)
  def my_task(logical_date, **context):
      # Utilizing logical_date

Subsequent, consider the utilization of the 2 scheduling-related default modifications. catchup_by_default is now False, that means lacking DAG runs will not mechanically backfill. If backfill is required, replace the DAG definition with catchup=True. In case your DAGs require backfill, you will need to take into account the affect of this migration and backfilling. Since you’re migrating a DAG to a clear setting with no historical past, enabling backfilling will create DAG runs for all runs starting with the required start_date. Think about updating the start_date to keep away from pointless runs.

create_cron_data_intervals can also be now False. With this alteration, cron expressions are evaluated as a CronTriggerTimetable assemble.

Lastly, consider the utilization of deprecated context variables for manually and Asset-triggered DAGs, then replace your code with appropriate replacements.

Updating necessities and testing

Along with doable package deal model modifications, a number of core Airflow operators beforehand included within the airflow-core package deal moved to the apache-airflow-providers-standard package deal. These modifications have to be included into your necessities.txt file. Specifying, or pinning, package deal variations in your necessities file is a greatest observe and beneficial for this migration.To replace your necessities file, full the next steps:

1. Obtain and configure the Amazon MWAA Docker pictures. For extra particulars, discuss with the GitHub repo.
2. Copy the present setting’s necessities.txt file to a brand new file.
3. If wanted, add the apache-airflow-providers-standard package deal to the brand new necessities file.
4. Obtain the suitable Airflow constraints file to your goal Airflow model to your working director. A constraints file is on the market for every Airflow model and Python model mixture. The URL takes the next type:
   https://uncooked.githubusercontent.com/apache/airflow/constraints-${AIRFLOW_VERSION}/constraints-${PYTHON_VERSION}.txt
5. Create your versioned necessities file utilizing your un-versioned file and the constraints file. For steerage on making a necessities file, see Making a necessities.txt file. Ensure that there aren’t any dependency conflicts earlier than transferring ahead.
6. Confirm your necessities file utilizing the Docker picture. Run the next command contained in the operating container:

   ./run.sh test-requirements

   Tackle any set up errors by updating package deal variations.

As a greatest observe, we suggest packaging your packages right into a ZIP file for deployment in Amazon MWAA. This makes positive the identical actual packages are put in on all Airflow nodes. Consult with Putting in Python dependencies utilizing PyPi.org Necessities File Format for detailed details about packaging dependencies.

Creating a brand new Amazon MWAA 3.x setting

As a result of Amazon MWAA requires a migration strategy for main model upgrades, you will need to create a brand new setting to your blue/inexperienced deployment. This publish makes use of the AWS Command Line Interface (AWS CLI) for example, it’s also possible to use infrastructure as code (IaC).

1. Create a brand new S3 bucket utilizing the identical construction as the present S3 bucket.
2. Add the up to date necessities file and any plugin packages to the brand new S3 bucket.
3. Generate a template to your new setting configuration:

   aws mwaa create-environment --generate-cli-skeleton > new-mwaa3-env.json

4. Modify the generated JSON file:
   1. Copy configurations out of your current setting.
   2. Replace the setting identify.
   3. Set the AirflowVersion parameter to the goal 3.x model.
   4. Replace the S3 bucket properties with the brand new S3 bucket identify.
   5. Assessment and replace different configuration parameters as wanted.

   Configure the brand new setting with the identical networking settings, safety teams, and IAM roles as your current setting. Consult with the Amazon MWAA Consumer Information for these configurations.

5. Create your new setting:

   aws mwaa create-environment --cli-input-json file://new-mwaa3-env.json

Metadata migration

Your new setting requires the identical variables, connections, roles, and pool configurations. Use this part as a information for migrating this info. If you happen to’re utilizing AWS Secrets and techniques Supervisor as your secrets and techniques backend, you don’t have to migrate any connections. Relying your setting’s measurement, you’ll be able to migrate this metadata utilizing the Airflow UI or the Apache Airflow REST API.

1. Replace any customized pool info within the new setting utilizing the Airflow UI.
2. For environments utilizing the metadatabase as a secrets and techniques backend, migrate all connections to the brand new setting.
3. Migrate all variables to the brand new setting.
4. Migrate any customized Airflow roles to the brand new setting.

Migration execution and validation

Plan and execute the transition out of your outdated setting to the brand new one:

1. Schedule the migration throughout a interval of low workflow exercise to reduce disruption.
2. Implement a freeze window for DAG modifications earlier than and throughout the migration.
3. Execute the migration in phases:
   1. Pause DAGs within the outdated setting. For a small variety of DAGs, you should use the Airflow UI. For bigger teams, think about using the REST API.
   2. Confirm all operating duties have accomplished within the Airflow UI.
   3. Redirect DAG triggers and exterior integrations to the brand new setting.
   4. Copy the up to date DAGs to the brand new setting’s S3 bucket.
   5. Allow DAGs within the new setting. For a small variety of DAGs, you should use the Airflow UI. For bigger teams, think about using the REST API.
4. Monitor the brand new setting carefully throughout the preliminary operation interval:
   1. Look ahead to failed duties or scheduling points.
   2. Test for lacking variables or connections.
   3. Confirm exterior system integrations are functioning accurately.
   4. Monitor Amazon CloudWatch metrics to verify the setting is performing as anticipated.

Publish-migration validation

After the migration, completely validate the brand new setting:

• Confirm that each one DAGs are being scheduled accurately in keeping with their outlined schedules
• Test that job historical past and logs are accessible and full
• Check crucial workflows end-to-end to verify they execute efficiently
• Validate connections to exterior techniques are functioning correctly
• Monitor CloudWatch metrics for efficiency validation

Cleanup and documentation

When the migration is full and the brand new setting is steady, full the next steps:

1. Doc the modifications made throughout the migration course of.
2. Replace runbooks and operational procedures to mirror the brand new setting.
3. After a ample stability interval, outlined by stakeholders, decommission the outdated setting:

   aws mwaa delete-environment --name old-mwaa2-env

4. Archive backup knowledge in keeping with your group’s retention insurance policies.

Conclusion

The journey from Airflow 2.x to three.x on Amazon MWAA is a chance to embrace next-generation workflow orchestration capabilities whereas sustaining the reliability of your workflow operations. By following these greatest practices and sustaining a methodical strategy, you’ll be able to efficiently navigate this transition whereas minimizing dangers and disruptions to what you are promoting operations.

A profitable migration requires thorough preparation, systematic testing, and sustaining clear documentation all through the method. Though the migration strategy requires extra preliminary effort, it offers the protection and management wanted for such a major improve.

Concerning the authors