Airflow Dynamic Task Mapping: Difference between revisions

From NovaOrdis Knowledge Base
Jump to navigation Jump to search
Line 51: Line 51:
</syntaxhighlight>
</syntaxhighlight>


However, Airflow 2.3 and newer allow declaring truly dynamic DAGs, where the structure of the DAG is determined at runtime by data dynamically generated by previous tasks.
However, Airflow 2.3 and newer allow declaring truly dynamic DAGs, where the structure of the DAG is determined at runtime by data dynamically generated by previous tasks, which could not be known in advance when the DAG structure is programmed.





Revision as of 21:51, 18 July 2022

External

Internal

Overview

Many workflows are static, in that all the component task instances and their relationships are statically declared in the workflow definition code:

# noinspection PyPep8Naming
@task
def A():
    ...

@task
def B():
    ...

@task
def C():
    ...

@dag(...)
def some_dag():
    A() >> B() >> C()

However, since the workflow definition code is Python, it allows for a certain degree of flexibility when declaring tasks in a loop:

@dag(...)
def some_dag():
  a = A()
  c = C()
  for i in [1, 2, 3]:
      a >> B() >> c

The resulting DAG looks similar to:

Parallel 1.png

This is not really dynamic, in the sense that the structure of the DAG changes at runtime, but equivalent with declaring the same task, individually, multiple times:

@dag(...)
def some_dag():
  a = A()
  c = C()
  a >> B() >> c
  a >> B() >> c
  a >> B() >> c

However, Airflow 2.3 and newer allow declaring truly dynamic DAGs, where the structure of the DAG is determined at runtime by data dynamically generated by previous tasks, which could not be known in advance when the DAG structure is programmed.




A workflow can create a number of tasks at runtime, based upon current workflow instance state, rather than the DAG author having to know in advance how many tasks would be needed. However, the tasks can be created in the DAG definition itself. Tasks cannot create dynamic tasks.

Questions

  • Can dependencies be declared on dynamic tasks?