Viewflow
Viewflow is a framework built on the top of Airflow that enables data scientists to create materialized views. It allows data scientists to focus on the logic of the view creation in their preferred tool.
Viewflow automatically creates Airflow DAGs and tasks based on SQL, Python, R or Rmd files. Normally, each of these files is responsible for materializing a new view. You write the view definition, Viewflow handles the rest!
One of the major features of Viewflow is its ability to manage tasks' dependencies, i.e., views used to create another view. Viewflow can automatically extract from the code (e.g. SQL query or Python script) the internal and external dependencies. An internal dependency is a view that belongs to the same DAG as a view being created. An external dependency is a view that belongs to a different DAG. The benefits of automatic dependency management are twofold: First, data scientists don't have to manually list dependencies, usually an error-prone process. Second, it makes sure that no view is built on stale data (because all dependent views will be updated beforehand).
Currently, Viewflow supports SQL, Python, R and Rmd views and PostgreSQL/Redshift as a destination. We will continue improving Viewflow by adding new view types (e.g. Jupyter Notebooks) and destinations (e.g., Snowflake, BigQuery, ...).
Do you want more context on why we built and released Viewflow? Check out our announcement blog post: Data Scientists, don’t worry about data engineering: Viewflow has your back.!
Viewflow demo
We created a demo that shows how Viewflow works. The demo creates multiple DAGs: viewflow-demo-1 through viewflow-demo-4. These DAGs create a total of four views in a local Postgres database. Check out the view files in demo/dags/. Some of the following commands are different based on which Airflow version you're using. For new users, Airflow 2 is the best option. However, you can also run the demo using the older Airflow 1.10 version by using the indicated commands.
Run the demo
We use docker-compose to instantiate an Apache Airflow instance and a Postgres database. The Airflow container and the Postgres container are defined in the docker-compose-airflow<version>.yml files. The first time you want to run the demo, you will first have to build the Apache Airflow docker image that embeds Viewflow:
docker-compose -f docker-compose-airflow2.yml build # Airflow 2
docker-compose -f docker-compose-airflow1.10.yml build # Airflow 1.10Then run the docker containers:
docker-compose -f docker-compose-airflow2.yml up # Airflow 2
docker-compose -f docker-compose-airflow1.10.yml up # Airflow 1.10Go to your local Apache Airflow instance on http://localhost:8080. There are four DAGs called viewflow-demo-1 through viewflow-demo-4. Notice how Viewflow automatically generated these DAGs based on the example queries in the subfolders of demo/dags/!
By default, the DAGs are disabled. Turn the DAGs on by clicking on the button Off. This will trigger the DAGs.
Query the views
Once the DAGs have run and all tasks completed, you can query the views created by Viewflow in the local Postgres database created by Docker. You can use any Postgres client (note that Postgres is running locally on port 5432):
psql -h localhost -p 5432 -U airflow -d airflowUse airflow when psql asks you for the user password.
There is a schema named viewflow_raw and a schema named viewflow_demo. The first one contains three tables: users, courses, and user_course. They are considered as the raw data. The second schema, viewflow_demo, is the schema in which the views created by Viewflow are stored.
\dn
+---------------+---------+
| Name | Owner |
|---------------+---------|
| public | airflow |
| viewflow_demo | airflow |
| viewflow_raw | airflow |
+---------------+---------+Viewflow created four views: user_xp (SQL), user_enriched (SQL), course_enriched (SQL) and top_3_user_xp (Python)
\dt viewflow_demo.
+---------------+-----------------+--------+---------+
| Schema | Name | Type | Owner |
|---------------+-----------------+--------+---------|
| viewflow_demo | course_enriched | table | airflow |
| viewflow_demo | top_3_user_xp | table | airflow |
| viewflow_demo | user_enriched | table | airflow |
| viewflow_demo | user_xp | table | airflow |
+---------------+-----------------+--------+---------+You can query these tables to see their data:
select * from viewflow_demo.user_xp;
+-----------+------+-----------------------+
| user_id | xp | __view_generated_at |
|-----------+------+-----------------------|
| 1 | 750 | 2021-03-17 |
| 2 | 200 | 2021-03-17 |
| 3 | 550 | 2021-03-17 |
| 4 | 500 | 2021-03-17 |
| 5 | 650 | 2021-03-17 |
| 6 | 430 | 2021-03-17 |
| 7 | 300 | 2021-03-17 |
| 8 | 280 | 2021-03-17 |
| 9 | 100 | 2021-03-17 |
| 10 | 350 | 2021-03-17 |
+-----------+------+-----------------------+You can also access the tables' comment (both table and columns):
select obj_description('viewflow_demo.user_enriched'::regclass) as view_description;
+---------------------------------------------+
| view_description |
|---------------------------------------------|
| A table with enriched information of users |
+---------------------------------------------+select
column_name,
col_description((table_schema||'.'||table_name)::regclass::oid, ordinal_position) as column_comment
from
information_schema.columns
where
table_schema = 'viewflow_demo'
and
table_name = 'user_enriched';
+--------------------------+-----------------------------------------------+
| column_name | column_comment |
|--------------------------+-----------------------------------------------|
| user_id | The user id |
| xp | The user amount of XP |
| last_course_completed_at | When was the last course completed by a user |
| last_course_completed | Name of the latest completed course by a user |
| number_courses_completed | Number of completed courses by a user |
| __view_generated_at | <null> |
+--------------------------+-----------------------------------------------+And that's it! Congrats on running the demo 🚀 If you want to play more with Viewflow, follow the installation instructions below.
Installation instructions
The current installation process requires you to install Viewflow from the GitHub repository:
RUN pip install git+https://github.com/datacamp/viewflow.git Create a new DAG
Viewflow creates the DAGs automatically based on configuration files.
Here are the steps to create a DAG for the first time.
Create the Viewflow main script
In your Airflow DAG directory (usually $AIRFLOW_HOME/dags), create a python script called viewflow-dags.py that contains the following Python code:
from viewflow import create_dags
DAG = create_dags("./dags", globals(), "<views_schema_name>")This script is executed by Airflow. It calls the main Viewflow function that creates your DAGs. The first parameter is the directory in which your dag folders are located. The third parameter is the schema name in your data warehouse, where your views will be materialized.
Create an Airflow connection to your destination
Viewflow needs to know where to write the views. It uses an Airflow connection that is referred to in the view files by specifying a connection_id. Currently, Viewflow supports Postgres (or Redshift) data warehouses. Please look at the Airflow documentation to create a Postgres connection.
E.g. the demo's connection is managed using environmemt variables declared in demo/.env. This file is the env_file specified in the docker-compose-airflow<version>.yml files and it allows the scheduler and webserver containers to connect to the Postgres server.
Create your DAG directories
In Viewflow, the DAGs are created based on a configuration file and on the SQL and Python files in the same directory.
In $AIRFLOW_HOME/dags/, create a directory called my-first-viewflow-dag. In this directory, create a config.yml file that contains the following yml fields:
default_args:
owner: <email>
retries: 1
schedule_interval: 0 6 * * *
start_date: "2021-01-01"Adapt the values of each element to what suits you. The default_args element contains the Airflow default DAG parameters.
The schedule_interval and start_date elements are the Viewflow counterparts of Airflow's schedule_interval and start_date.
You can now add your SQL and Python files in this directory (see sections below). This will create a new DAG in Airflow called my-first-viewflow-dag that will be triggered every day at 6 AM UTC as of January 1, 2021. All failed tasks will be retried once.
View metadata
Viewflow expects some metadata that must be included in the SQL and Python files (examples follow). Here are the fields that should be included in a yml format:
- owner: The owner of the view (i.e., who is view responsible). The owner appears in Airflow and allows users to know who they should talk to if they have some questions about the view.
- description: What is the view about. Viewflow uses this field as a view comment in the database. The description can be retrieved in SQL (see Section Query the views).
- fields (list): Description of each column of the view. Viewflow uses these fields as column comments in the database. The column descriptions can be retrieved in SQL (see Section Query the views).
- schema: The name of the schema in which Viewflow creates the view. It's also used by Viewflow to create the dependencies.
- connection_id: Airflow connection name used to connect to the database (See Section Create an Airflow connection to your destination).
The newly created view has the same name as the filename of the SQL query, Python script or R(md) script.
SQL views
A SQL view is created by a SQL file. This SQL file must contain the SQL query (as a SELECT statement) of your view and the view metadata. Here's an example:
/*
---
owner: email address of the view owner
description: A description of your view. It's used as the view's description in the database
fields:
email: Description of your column -- used as the view column's description in the database
schema: schema_name_in_your_destination (e.g. viewflow_demo)
connection_id: airflow_destination_connection
---
*/
SELECT DISTINCT email FROM viewflow_raw.usersPython views
Please note that the implementation of the Python view should be considered as beta. It is a newer implementation of the Python view that we use at DataCamp.
A Python view is created based on a Python script. This script must contain at least one function with the view's description metadata in its docstring, which returns a Pandas dataframe.
Here's an example of a Python view:
import pandas as pd
def python_view(db_engine):
"""
---
owner: email address of the view owner
description: A description of your view. It's used as the view's description in the database
fields:
email: Description of your column -- used as the view column's description in the database
schema: schema_name_in_your_destination (e.g. viewflow_demo)
connection_id: airflow_destination_connection
---
"""
df = pd.read_sql_table("users", db_engine, schema="viewflow_raw")
return df[["email"]]Please note that Viewflow expects the Python function that creates the view to have the parameter db_engine (used to connect to the database). You don't have to set db_engine anywhere. Viewflow takes care of setting this variable.
R views
Viewflow handles R scripts similar to the existing SQL and Python files. Additionally, there's an element of automatisation. You simply define the view in R code, Viewflow will automatically read the necessary tables and write the new view to the database. Note that you need to define the new view in the R script with the same name as the R script (which is also the name of the table where the view is materialized in the database).
By default, other tables are expected to be referenced as <schema_name>.<table_name>.
This default behaviour can be changed by adding a new function in dependencies_r_patterns.py and adding a line dependency_function: <your_custom_function> to the metadata of the R script. The script user_xp_duplicate.R illustrates this.
Rmd views
Rmd scripts can be used mostly like R scripts. For Rmd scripts, you do have to explicitly configure the automated reading and writing of tables by adding automate_read_write: True to the metadata. By default, the script is executed as is. The task top_3_user_xp_duplicate.Rmd contains an explanation of the usage of Rmd scripts.
Configuring callbacks
A useful feature is enabling callbacks when a task succeeds, fails, or is retried. This callback can take many forms, e.g. an email or a Slack message. Viewflow allows you to define your own callbacks in viewflow/task_callbacks.py. These callbacks can be configured on multiple levels:
- By default, certain functions defined in viewflow/task_callbacks.py are used (e.g.
on_success_callback_default). - The callbacks can be overwritten for all tasks in a given DAG. E.g. if you have defined 3 custom callback functions in viewflow/task_callbacks.py, you can specify them in the DAG's
config.ymlfile as following:
default_args:
on_success_callback: on_success_callback_custom
on_failure_callback: on_failure_callback_custom
on_retry_callback: on_retry_callback_custom- For the highest level of configurability, you can overwrite the callbacks for a specific task. This option is prioritized, it doesn't matter whether there are callbacks specified in the DAG's
config.ymlfile. The callback functions can simply be added to the metadata of the task's script:
on_success_callback: on_success_callback_custom
on_failure_callback: on_failure_callback_custom
on_retry_callback: on_retry_callback_customOf course, options 1, 2 and 3 can be combined to efficiently configure the callbacks of a multitude of tasks.
Contributing to Viewflow
We welcome all sorts of contributions, be it new features, bug fixes or documentation, we encourage you to create a new PR. To create a new PR or to report new bugs, please read how to contribute to Viewflow.
In the remainder of this section, we show you how to prepare your environment to contribute to Viewflow.
Install Poetry
See https://python-poetry.org/docs/#osx-linux-bashonwindows-install-instructions for comprehensive documentation.
curl -sSL https://raw.githubusercontent.com/python-poetry/poetry/master/get-poetry.py | python
Install the dependencies
You can automatically install the required dependencies by running
poetry installBy default, this will install Airflow 2 and its corresponding dependencies. If you want to use Airflow 1.10, copy the [email protected]/pyproject.toml file to the main directory.
Prepare your environment to run the tests
Postgres
Use docker compose to set up a PostgreSQL database locally:
docker-compose -f docker-compose-test.yml upIf you get a message saying that port 5432 is in use, it means you have a different PostgreSQL server running on your machine. If you used homebrew to install it, you could use brew services stop postgresql to stop the other server.
Import the fixtures into the local database (the password is passw0rd):
psql -U user -W -h localhost -f tests/fixtures/load_postgres.sql -d viewflowRun Pytest
Before you can continue, you will need to set up an Airflow SQLite database.
poetry run airflow db init # Airflow 2
poetry run airflow initdb # Airflow 1.10If running into problems, this link can be helpful. In particular, it's possible you get a Symbol not found: _Py_GetArgcArgv error. This is easily fixed by creating a Python virtual environment as demonstrated in the link, activating this virtual environment and then running poetry install again.
Note for Airflow 1.10.12: if you get an ImportError, it can be helpful to refer to this post.
After setting up the database, run
poetry run pytestIn case the database connection is set up incorrectly, run
poetry run airflow db reset # Airflow 2
poetry run airflow resetdb # Airflow 1.10Viewflow architecture
We built Viewflow around three main components: the parser, the adapter, and the dependency extractor.
The parser transforms a source file (e.g., SQL, Rmd, Python) that contains the view's metadata (e.g., view's owner, view's descriptions, and column's descriptions) and the view's code into a specific Viewflow data structure. The data structure is used by the other components in the Viewflow architecture: the adapter and the dependency creator.
The adapter is the translation layer of Viewflow's views to their corresponding Airflow counterpart. It uses the data structure objects created by the parser to create an Airflow task object (i.e., an Airflow operator).
Finally, the dependency extractor uses the parser's data structure objects to set the internal and external dependencies to the Airflow task object created by the adapter.
This architecture allows us to add new source file types in the future easily (e.g. Python notebook).
Acknowledgments
Today's version of Viewflow is the result of a joint effort of ex and current DataCampers. We would like to thank in particular the following persons who significantly contributed to Viewflow: