PythonOperator¶
Use the PythonOperator to execute Python callables.
Warning
The @task decorator is recommended over the classic PythonOperator to execute Python callables.
@task(task_id="print_the_context")
def print_context(ds=None, **kwargs):
"""Print the Airflow context and ds variable from the context."""
pprint(kwargs)
print(ds)
return "Whatever you return gets printed in the logs"
run_this = print_context()
def print_context(ds=None, **kwargs):
"""Print the Airflow context and ds variable from the context."""
pprint(kwargs)
print(ds)
return "Whatever you return gets printed in the logs"
run_this = PythonOperator(task_id="print_the_context", python_callable=print_context)
Passing in arguments¶
Pass extra arguments to the @task decorated function as you would with a normal Python function.
# Generate 5 sleeping tasks, sleeping from 0.0 to 0.4 seconds respectively
@task
def my_sleeping_function(random_base):
"""This is a function that will run within the DAG execution"""
time.sleep(random_base)
for i in range(5):
sleeping_task = my_sleeping_function.override(task_id=f"sleep_for_{i}")(random_base=i / 10)
run_this >> log_the_sql >> sleeping_task
# Generate 5 sleeping tasks, sleeping from 0.0 to 0.4 seconds respectively
def my_sleeping_function(random_base):
"""This is a function that will run within the DAG execution"""
time.sleep(random_base)
for i in range(5):
sleeping_task = PythonOperator(
task_id=f"sleep_for_{i}", python_callable=my_sleeping_function, op_kwargs={"random_base": i / 10}
)
run_this >> log_the_sql >> sleeping_task
Templating¶
Airflow passes in an additional set of keyword arguments: one for each of the
Jinja template variables and a templates_dict
argument.
The templates_dict argument is templated, so each value in the dictionary
is evaluated as a Jinja template.
@task(task_id="log_sql_query", templates_dict={"query": "sql/sample.sql"}, templates_exts=[".sql"])
def log_sql(**kwargs):
logging.info("Python task decorator query: %s", str(kwargs["templates_dict"]["query"]))
log_the_sql = log_sql()
def log_sql(**kwargs):
logging.info("Python task decorator query: %s", str(kwargs["templates_dict"]["query"]))
log_the_sql = PythonOperator(
task_id="log_sql_query",
python_callable=log_sql,
templates_dict={"query": "sql/sample.sql"},
templates_exts=[".sql"],
)
PythonVirtualenvOperator¶
Use the PythonVirtualenvOperator decorator to execute Python callables
inside a new Python virtual environment. The virtualenv package needs to be installed in the environment
that runs Airflow (as optional dependency pip install apache-airflow[virtualenv] --constraint ...).
Warning
The @task.virtualenv decorator is recommended over the classic PythonVirtualenvOperator
to execute Python callables inside new Python virtual environments.
@task.virtualenv(
task_id="virtualenv_python", requirements=["colorama==0.4.0"], system_site_packages=False
)
def callable_virtualenv():
"""
Example function that will be performed in a virtual environment.
Importing at the module level ensures that it will not attempt to import the
library before it is installed.
"""
from time import sleep
from colorama import Back, Fore, Style
print(Fore.RED + "some red text")
print(Back.GREEN + "and with a green background")
print(Style.DIM + "and in dim text")
print(Style.RESET_ALL)
for _ in range(4):
print(Style.DIM + "Please wait...", flush=True)
sleep(1)
print("Finished")
virtualenv_task = callable_virtualenv()
def callable_virtualenv():
"""
Example function that will be performed in a virtual environment.
Importing at the module level ensures that it will not attempt to import the
library before it is installed.
"""
from time import sleep
from colorama import Back, Fore, Style
print(Fore.RED + "some red text")
print(Back.GREEN + "and with a green background")
print(Style.DIM + "and in dim text")
print(Style.RESET_ALL)
for _ in range(4):
print(Style.DIM + "Please wait...", flush=True)
sleep(1)
print("Finished")
virtualenv_task = PythonVirtualenvOperator(
task_id="virtualenv_python",
python_callable=callable_virtualenv,
requirements=["colorama==0.4.0"],
system_site_packages=False,
)
Passing in arguments¶
Pass extra arguments to the @task.virtualenv decorated function as you would with a normal Python function.
Unfortunately, Airflow does not support serializing var, ti and task_instance due to incompatibilities
with the underlying library. For Airflow context variables make sure that you either have access to Airflow through
setting system_site_packages to True or add apache-airflow to the requirements argument.
Otherwise you won’t have access to the most context variables of Airflow in op_kwargs.
If you want the context related to datetime objects like data_interval_start you can add pendulum and
lazy_object_proxy.
Warning
The Python function body defined to be executed is cut out of the DAG into a temporary file w/o surrounding code. As in the examples you need to add all imports again and you can not rely on variables from the global Python context.
If you want to pass variables into the classic PythonVirtualenvOperator use
op_args and op_kwargs.
If additional parameters for package installation are needed pass them in via the pip_install_options parameter or use a
requirements.txt as in the example below:
SomePackage==0.2.1 --pre --index-url http://some.archives.com/archives
AnotherPackage==1.4.3 --no-index --find-links /my/local/archives
All supported options are listed in the requirements file format.
Virtual environment setup options¶
The virtual environment is created based on the global python pip configuration on your worker. Using additional ENVs in your environment or adjustments in the general pip configuration as described in pip config.
If you want to use additional task specific private python repositories to setup the virtual environment, you can pass the index_urls parameter which will adjust the
pip install configurations. Passed index urls replace the standard system configured index url settings.
To prevent adding secrets to the private repository in your DAG code you can use the Airflow
Connections & Hooks. For this purpose the connection type Package Index (Python) can be used.
In the special case you want to prevent remote calls for setup of a virtual environment, pass the index_urls as empty list as index_urls=[] which
forced pip installer to use the --no-index option.
Caching and reuse¶
Setup of virtual environments is made per task execution in a temporary directory. After execution the virtual environment is deleted again. Ensure that the $tmp folder
on your workers have sufficient disk space. Usually (if not configured differently) the local pip cache will be used preventing a re-download of packages
for each execution.
But still setting up the virtual environment for every execution needs some time. For repeated execution you can set the option venv_cache_path to a file system
folder on your worker. In this case the virtual environment will be set up once and be re-used. If virtual environment caching is used, per unique requirements set different
virtual environment subfolders are created in the cache path. So depending on your variations in the DAGs in your system setup sufficient disk space is needed.
Note that no automated cleanup is made and in case of cached mode. All worker slots share the same virtual environment but if tasks are scheduled over and over on
different workers, it might happen that virtual environment are created on multiple workers individually. Also if the worker is started in a Kubernetes POD, a restart
of the worker will drop the cache (assuming venv_cache_path is not on a persistent volume).
In case you have problems during runtime with broken cached virtual environments, you can influence the cache directory hash by setting the Airflow variable
PythonVirtualenvOperator.cache_key to any text. The content of this variable is uses in the vector to calculate the cache directory key.
Note that any modification of a cached virtual environment (like temp files in binary path, post-installing further requirements) might pollute a cached virtual environment and the operator is not maintaining or cleaning the cache path.
ExternalPythonOperator¶
The ExternalPythonOperator can help you to run some of your tasks with a different set of Python
libraries than other tasks (and than the main Airflow environment). This might be a virtual environment
or any installation of Python that is preinstalled and available in the environment where Airflow
task is running. The operator takes Python binary as python parameter. Note, that even in case of
virtual environment, the python path should point to the python binary inside the virtual environment
(usually in bin subdirectory of the virtual environment). Contrary to regular use of virtual
environment, there is no need for activation of the environment. Merely using python binary
automatically activates it. In both examples below PATH_TO_PYTHON_BINARY is such a path, pointing
to the executable Python binary.
Use the ExternalPythonOperator to execute Python callables inside a
pre-defined environment. The virtualenv package should be preinstalled in the environment where Python is run.
In case dill is used, it has to be preinstalled in the environment (the same version that is installed
in main Airflow environment).
Warning
The @task.external_python decorator is recommended over the classic ExternalPythonOperator
to execute Python code in pre-defined Python environments.
@task.external_python(task_id="external_python", python=PATH_TO_PYTHON_BINARY)
def callable_external_python():
"""
Example function that will be performed in a virtual environment.
Importing at the module level ensures that it will not attempt to import the
library before it is installed.
"""
import sys
from time import sleep
print(f"Running task via {sys.executable}")
print("Sleeping")
for _ in range(4):
print("Please wait...", flush=True)
sleep(1)
print("Finished")
external_python_task = callable_external_python()
def callable_external_python():
"""
Example function that will be performed in a virtual environment.
Importing at the module level ensures that it will not attempt to import the
library before it is installed.
"""
import sys
from time import sleep
print(f"Running task via {sys.executable}")
print("Sleeping")
for _ in range(4):
print("Please wait...", flush=True)
sleep(1)
print("Finished")
external_python_task = ExternalPythonOperator(
task_id="external_python",
python_callable=callable_external_python,
python=PATH_TO_PYTHON_BINARY,
)
Passing in arguments¶
Pass extra arguments to the @task.external_python decorated function as you would with a normal Python function.
Unfortunately Airflow does not support serializing var and ti / task_instance due to incompatibilities
with the underlying library. For Airflow context variables make sure that Airflow is also installed as part
of the virtualenv environment in the same version as the Airflow version the task is run on.
Otherwise you won’t have access to the most context variables of Airflow in op_kwargs.
If you want the context related to datetime objects like data_interval_start you can add pendulum and
lazy_object_proxy to your virtual environment.
Warning
The Python function body defined to be executed is cut out of the DAG into a temporary file w/o surrounding code. As in the examples you need to add all imports again and you can not rely on variables from the global Python context.
If you want to pass variables into the classic ExternalPythonOperator use
op_args and op_kwargs.
PythonBranchOperator¶
Use the PythonBranchOperator to execute Python branching
tasks.
Warning
The @task.branch decorator is recommended over the classic PythonBranchOperator
to execute Python code.
@task.branch()
def branching(choices: list[str]) -> str:
return f"branch_{random.choice(choices)}"
branching = BranchPythonOperator(
task_id="branching",
python_callable=lambda: f"branch_{random.choice(options)}",
)
Argument passing and templating options are the same like with PythonOperator.
BranchPythonVirtualenvOperator¶
Use the BranchPythonVirtualenvOperator decorator to execute Python branching
tasks and is a hybrid of the PythonBranchOperator with execution in a virtual environment.
Warning
The @task.branch_virtualenv decorator is recommended over the classic
BranchPythonVirtualenvOperator to execute Python code.
# Note: Passing a caching dir allows to keep the virtual environment over multiple runs
# Run the example a second time and see that it re-uses it and is faster.
VENV_CACHE_PATH = tempfile.gettempdir()
@task.branch_virtualenv(requirements=["numpy~=1.24.4"], venv_cache_path=VENV_CACHE_PATH)
def branching_virtualenv(choices) -> str:
import random
import numpy as np
print(f"Some numpy stuff: {np.arange(6)}")
return f"venv_{random.choice(choices)}"
# Note: Passing a caching dir allows to keep the virtual environment over multiple runs
# Run the example a second time and see that it re-uses it and is faster.
VENV_CACHE_PATH = Path(tempfile.gettempdir())
def branch_with_venv(choices):
import random
import numpy as np
print(f"Some numpy stuff: {np.arange(6)}")
return f"venv_{random.choice(choices)}"
branching_venv = BranchPythonVirtualenvOperator(
task_id="branching_venv",
requirements=["numpy~=1.24.4"],
venv_cache_path=VENV_CACHE_PATH,
python_callable=branch_with_venv,
op_args=[options],
)
Argument passing and templating options are the same like with PythonVirtualenvOperator.
BranchExternalPythonOperator¶
Use the BranchExternalPythonOperator to execute Python branching
tasks and is a hybrid of the PythonBranchOperator with execution in an
external Python environment.
Warning
The @task.branch_external_python decorator is recommended over the classic
BranchExternalPythonOperator to execute Python code.
@task.branch_external_python(python=PATH_TO_PYTHON_BINARY)
def branching_ext_python(choices) -> str:
import random
return f"ext_py_{random.choice(choices)}"
def branch_with_external_python(choices):
import random
return f"ext_py_{random.choice(choices)}"
branching_ext_py = BranchExternalPythonOperator(
task_id="branching_ext_python",
python=PATH_TO_PYTHON_BINARY,
python_callable=branch_with_external_python,
op_args=[options],
)
Argument passing and templating options are the same like with ExternalPythonOperator.
ShortCircuitOperator¶
Use the ShortCircuitOperator to control whether a pipeline continues
if a condition is satisfied or a truthy value is obtained.
The evaluation of this condition and truthy value is done via the output of a callable. If the callable returns True or a truthy value, the pipeline is allowed to continue and an XCom of the output will be pushed. If the output is False or a falsy value, the pipeline will be short-circuited based on the configured short-circuiting (more on this later). In the example below, the tasks that follow the “condition_is_true” task will execute while the tasks downstream of the “condition_is_false” task will be skipped.
Warning
The @task.short_circuit decorator is recommended over the classic ShortCircuitOperator
to short-circuit pipelines via Python callables.
@task.short_circuit()
def check_condition(condition):
return condition
ds_true = [EmptyOperator(task_id=f"true_{i}") for i in [1, 2]]
ds_false = [EmptyOperator(task_id=f"false_{i}") for i in [1, 2]]
condition_is_true = check_condition.override(task_id="condition_is_true")(condition=True)
condition_is_false = check_condition.override(task_id="condition_is_false")(condition=False)
chain(condition_is_true, *ds_true)
chain(condition_is_false, *ds_false)
cond_true = ShortCircuitOperator(
task_id="condition_is_True",
python_callable=lambda: True,
)
cond_false = ShortCircuitOperator(
task_id="condition_is_False",
python_callable=lambda: False,
)
ds_true = [EmptyOperator(task_id=f"true_{i}") for i in [1, 2]]
ds_false = [EmptyOperator(task_id=f"false_{i}") for i in [1, 2]]
chain(cond_true, *ds_true)
chain(cond_false, *ds_false)
The “short-circuiting” can be configured to either respect or ignore the trigger rule
defined for downstream tasks. If ignore_downstream_trigger_rules is set to True, the default configuration, all
downstream tasks are skipped without considering the trigger_rule defined for tasks. If this parameter is
set to False, the direct downstream tasks are skipped but the specified trigger_rule for other subsequent
downstream tasks are respected. In this short-circuiting configuration, the operator assumes the direct
downstream task(s) were purposely meant to be skipped but perhaps not other subsequent tasks. This
configuration is especially useful if only part of a pipeline should be short-circuited rather than all
tasks which follow the short-circuiting task.
In the example below, notice that the “short_circuit” task is configured to respect downstream trigger
rules. This means while the tasks that follow the “short_circuit” task will be skipped
since the decorated function returns False, “task_7” will still execute as its set to execute when upstream
tasks have completed running regardless of status (i.e. the TriggerRule.ALL_DONE trigger rule).
[task_1, task_2, task_3, task_4, task_5, task_6] = [
EmptyOperator(task_id=f"task_{i}") for i in range(1, 7)
]
task_7 = EmptyOperator(task_id="task_7", trigger_rule=TriggerRule.ALL_DONE)
short_circuit = check_condition.override(task_id="short_circuit", ignore_downstream_trigger_rules=False)(
condition=False
)
chain(task_1, [task_2, short_circuit], [task_3, task_4], [task_5, task_6], task_7)
[task_1, task_2, task_3, task_4, task_5, task_6] = [
EmptyOperator(task_id=f"task_{i}") for i in range(1, 7)
]
task_7 = EmptyOperator(task_id="task_7", trigger_rule=TriggerRule.ALL_DONE)
short_circuit = ShortCircuitOperator(
task_id="short_circuit", ignore_downstream_trigger_rules=False, python_callable=lambda: False
)
chain(task_1, [task_2, short_circuit], [task_3, task_4], [task_5, task_6], task_7)
Passing in arguments¶
Pass extra arguments to the @task.short_circuit-decorated function as you would with a normal Python function.
Templating¶
Jinja templating can be used in same way as described for the PythonOperator.
PythonSensor¶
The PythonSensor executes an arbitrary callable and waits for its return
value to be True.
Warning
The @task.sensor decorator is recommended over the classic PythonSensor
to execute Python callables to check for True condition.
# Using a sensor operator to wait for the upstream data to be ready.
@task.sensor(poke_interval=60, timeout=3600, mode="reschedule")
def wait_for_upstream() -> PokeReturnValue:
return PokeReturnValue(is_done=True, xcom_value="xcom_value")
t8 = PythonSensor(task_id="success_sensor_python", python_callable=success_callable)
t9 = PythonSensor(
task_id="failure_timeout_sensor_python", timeout=3, soft_fail=True, python_callable=failure_callable
)