Alan Mazankiewicz - Fully Orchestrating Databricks from Airflow

#1about 5 minutes

Exploring the core features of the Databricks workspace

A walkthrough of the Databricks UI shows how to create Spark clusters, run code in notebooks, and define scheduled jobs with multi-task dependencies.

#2about 6 minutes

Understanding the fundamentals of Apache Airflow orchestration

Airflow provides powerful workflow orchestration with features like dynamic task generation, complex trigger rules, and a detailed UI for monitoring DAGs.

#3about 5 minutes

Integrating Databricks and Airflow with built-in operators

The DatabricksRunNowOperator and DatabricksSubmitRunOperator allow Airflow to trigger predefined or dynamically defined jobs in Databricks via its REST API.

#4about 3 minutes

Creating a custom operator for full Databricks API control

To overcome the limitations of built-in operators, you can create a generic custom operator by subclassing BaseOperator and using the DatabricksHook to make arbitrary API calls.

#5about 3 minutes

Implementing a custom operator to interact with DBFS

A practical example demonstrates how to use the custom generic operator to make a 'put' request to the DBFS API, including the use of Jinja templates for dynamic paths.

#6about 2 minutes

Developing advanced operators for complex cluster management

For complex scenarios, custom operators can be built to create an all-purpose cluster, wait for it to be ready, submit multiple jobs, and then terminate it.

#7about 5 minutes

Answering questions on deployment, performance, and tooling

The discussion covers running Airflow in production environments like Kubernetes, optimizing Spark performance on Databricks, and comparing Airflow to Azure Data Factory.

#8about 10 minutes

Discussing preferred data stacks and career advice

The speaker shares insights on their preferred data stack for different use cases, offers advice for beginners learning Python, and describes a typical workday as a data engineer.

today

Machine Learning Engineer