airflow-mcd

Monte Carlo's alpha Airflow provider.

Installation

Requires Python 3.7 or greater and is compatible with Airflow 1.10.14 or greater.

You can install and update using pip. For instance:

pip install -U airflow-mcd

This package can be added like any other python dependency to Airflow (e.g. via requirements.txt).

Basic usage

Hooks:

• SessionHook

  Creates a pycarlo compatible session. This is useful for creating your own operator built on top of our Python SDK.

  This hook expects an Airflow HTTP connection with the Monte Carlo API id as the "login" and the API token as the "password".

  Alternatively, you could define both the Monte Carlo API id and token in "extra" with the following format:

  {
      "mcd_id": "<ID>",
      "mcd_token": "<TOKEN>"
  }
  

  See here for details on how to generate a token.

Operators:

• BaseMcdOperator

  This operator can be extended to build your own operator using our SDK or any other dependencies. This is useful if you want implement your own custom logic (e.g. creating custom lineage after a task completes).

• SimpleCircuitBreakerOperator

  This operator can be used to execute a circuit breaker compatible rule (custom SQL monitor) to run integrity tests before allowing any downstream tasks to execute. Raises an AirflowFailException if the rule condition is in breach when using an Airflow version newer than 1.10.11, as that is preferred for tasks that can be failed without retrying. Older Airflow versions raise an AirflowException. For instance:

  from datetime import datetime, timedelta
  
  from airflow import DAG
  
  try:
    from airflow.operators.bash import BashOperator
  except ImportError:
    # For airflow versions <= 2.0.0. This module was deprecated in 2.0.0.
    from airflow.operators.bash_operator import BashOperator
  
  from airflow_mcd.operators import SimpleCircuitBreakerOperator
  
  mcd_connection_id = 'mcd_default_session'
  
  with DAG('sample-dag', start_date=datetime(2022, 2, 8), catchup=False, schedule_interval=timedelta(1)) as dag:
      task1 = BashOperator(
          task_id='example_elt_job_1',
          bash_command='echo I am transforming a very important table!',
      )
      breaker = SimpleCircuitBreakerOperator(
          task_id='example_circuit_breaker',
          mcd_session_conn_id=mcd_connection_id,
          rule_uuid='<RULE_UUID>'
      )
      task2 = BashOperator(
          task_id='example_elt_job_2',
          bash_command='echo I am building a very important dashboard from the table created in task1!',
          trigger_rule='none_failed'
      )
  
      task1 >> breaker >> task2
  

  This operator expects the following parameters:

  • mcd_session_conn_id: A SessionHook compatible connection.
  • rule_uuid: UUID of the rule (custom SQL monitor) to execute.

  The following parameters can also be passed:

  • timeout_in_minutes [default=5]: Polling timeout in minutes. Note that The Data Collector Lambda has a max timeout of 15 minutes when executing a query. Queries that take longer to execute are not supported, so we recommend filtering down the query output to improve performance (e.g limit WHERE clause). If you expect a query to take the full 15 minutes we recommend padding the timeout to 20 minutes.
  • fail_open [default=True]: Prevent any errors or timeouts when executing a rule from stopping your pipeline. Raises AirflowSkipException if set to True and any issues are encountered. Recommended to set the trigger_rule param for any downstream tasks to none_failed in this case.

Tests and releases

Locally make test will run all tests. See README-dev.md for additional details on development. When ready for a review, create a PR against main.

When ready to release, create a new Github release with a tag using semantic versioning (e.g. v0.42.0) and CircleCI will test and publish to PyPI. Note that an existing version will not be deployed.

License

Apache 2.0 - See the LICENSE for more information.