Blog

Apache recently released Airflow 2.3.0. Since its last update, Apache Airflow 2.2.0, this new release has over 700 commits, including 50 new features, 99 improvements, 85 bug fixes and several doc changes.

Airflow 2.3 is a landmark release. Its new dynamic task mapping API, in particular, is a foundational feature that will improve iteratively in each new release of Airflow.

Here is a glimpse of the major updates:

• Dynamic Task Mapping
• Grid View Replaces Tree View
• JSON Serialization for Connections
• Airflow db downgrade and Offline Generation of SQL Scripts

Let’s discuss these updates in detail.

Dynamic Task Mapping

Dynamic task mapping lets Airflow trigger tasks based on unpredictable input conditions. This new feature adds the possibility of creating tasks dynamically at runtime. This is similar to defining tasks in a loop.

A scheduler can carry out this action depending on the results of the previous task rather than having the DAG file collect the data and execute it by itself. The scheduler will make copies of the mapped job, one for each input, just before it is executed. Dynamic Mapping is able to have a task generate the list to iterate over.

Representation

• In the new Grid View, dynamic tasks appear as a single task but their name is appended with a []
• In Graph/Network View, dynamic tasks come up as a single task before execution and are expanded after successful execution.

Benefits

• Implement tasks that change at run time based on the state of previous tasks
• Maintain task atomicity
• Reduce top-level code in your DAG files

Limitations

• We can only expand input parameters or XCOM results, which are lists or dictionaries. An UnmappableXComTypePushed exception is thrown at runtime with any other data type.
• Error handling is applicable to the entire set of tasks so if any one of the dynamic tasks fail, Airflow marks the whole set as failed.

Implementation
Dynamic Task Mapping is simple to implement with two new functions:

• partial(): parameters that remain constant for all tasks. This is for unmapped arguments.
• expand(): This function passes the parameter or parameters that you want to map on. A separate parallel task is created for each input. This is for mapped arguments.
  
  
  

The Grid View

The Grid View replaces the Tree View, with a focus on task history. Grid View makes the DAG display cleaner and provides more useful information than the previous tree view. For example, the grid view can quickly filter tasks upstream, so we can easily monitor the DAG. Also, we can directly know the DAG duration to finish all task instances, which allows us to know the average performance of the DAG.

Benefits

• Focus on task history,
• handle task groups better,
• leave dependency lines and structure of the DAG to Graph View,
• quick performance checks, as it shows duration and other information on the same view,
• additional “play” icon for manual triggers; and
• allows DAG versioning.

JSON Serialization for Connections

You can now create connections using the JSON serialization format.

Previously in general we could only store connections in the Airflow URI format. With this change we can serialize as JSON.

The Airflow URI format can be very tricky to work with and although we have for some time had a convenience method with Connection.get_uri, using JSON is just simpler. JSON serialization can also be used when setting connections in the environment variables.

Airflow DB downgrade and Offline generation of SQL scripts

A new command in Airflow 2.3.0 is used for your chosen version by downgrading or upgrading the database.

The downgrade/upgrade SQL scripts can also be generated for your database. You can also run it against your database manually or just view the SQL queries, which would be run by the downgrade/upgrade command.

airflow db downgrade downgrades the database to your chosen version and airflow db upgrade upgrades your database with the schema changes in the Airflow version you’re upgrading to. Alternatively, you may downgrade/upgrade to a specific Alembic revision ID.

If you want to preview the commands but not execute them, use option --show-sql-only.

Options --from-revision and --from-version may only be used in conjunction with the --show-sql-only option, because when actually running migrations we should always downgrade from the current revision.

Minor New/Updated Features

• New _BranchPythonDecoratedOperator added a decorator for BranchPythonOperator which wraps a Python callable and captures args/kwargs when called for execution.

@task.branch(task_id="branching")
def random_choice():
    return random.choice(options)

• Updated ShortCircuitOperator configurability for respecting downstream trigger rules
• Added a new parameter, ignore_downstream_trigger_rules, which allows users to have the ShortCircuitOperator perform a "hard short" (i.e., blindly skip all downstream tasks; the current behavior) or a "soft short" (i.e., the immediate, downstream task(s) are skipped only and the Scheduler is left to handle the trigger rules appropriately).
• Replaced DummyOperator with EmptyOperator

• Made operator's execution_timeout configurable
  execution_timeout attribute can be globally configured via airflow.cfg. The default value is still None. Users are expected to define an integer value to be passed into timedelta object to set the timeout in terms of seconds by default, via configuration.
• New ALL_SKIPPED trigger rule Trigger Rule when all upstream tasks are in a skipped state Example:

start = DummyOperator(task_id='start')
success = DummyOperator(task_id='success', trigger_rule='all_skipped')
skip = DummyOperator(task_id='skip', trigger_rule='all_skipped')

In Conclusion

With the new release the improvement in Airflow is astounding. It eliminates a growing amount of the tedious, repetitive work that would otherwise be done manually by the developers.

Airflow's new Grid View and dynamic task mapping API not only improve each user's experience but also formalize and incorporate use case-specific logic that would otherwise need to be developed by skilled developers. The new release keeps complete backward compatibility, so it should be able to run all code created in earlier versions. You could claim that Airflow breaks the vicious cycle of writing and maintaining custom code for each use case with each new release.