Express — 
Reusable Data Products for GenAI Unifying Databases, PDFs, and Logs
Reusable data products unify databases, PDFs, and logs with metadata, validation, and lineage to enable join-aware RAG retrieval for reliable GenAI applications.
As businesses grow increasingly data-driven, robust ETL (Extract, Transform, Load) solutions become critical to efficiently manage vast datasets. At Nexla, we’ve integrated Spark ETL with Databricks to deliver a flexible, scalable, and high-performance data processing solution. This blog dives deep into how Nexla’s Spark ETL works, its benefits, and the technical details behind its integration with Databricks.
Nexla Spark ETL: A Quick Overview
Nexla’s Spark ETL on Databricks is a powerful solution for handling complex data workflows. Leveraging the computational might of Spark distributed clusters and Databricks, this integration empowers data teams to streamline their data processing at scale. Here’s a breakdown of its core capabilities:
- Multiple Source and Destination Support: Nexla supports file-based sources (like AWS S3, ADLS, GCS) and Databricks sources for input, with any storage and Databricks destinations for output.
- Dynamic Cluster Management: Nexla can either spawn clusters in Databricks or use existing ones. This flexibility allows businesses to scale their resources dynamically depending on workload demands.
- Transformations with No-Code and Spark SQL: Nexla supports No-code transforms, Spark SQL, and even allows running Python or Javascript code for more customized operations.
Key Capabilities
- Efficient Data Handling: Nexla handles data ingestion (500+ connectors available out of the box) and writes data back to its destination with minimal effort. Whether your data is on DBFS or a Databricks warehouse, the integration seamlessly moves data between systems while maintaining performance.
- Leverage Databricks Compute: Once the data is on cloud or another supported source, Spark ETL in Databricks can perform the heavy lifting by reading data into Spark DataFrames, applying transformations, and writing the output back to delta tables or DBFS or choice of storage.
- Cluster Lifecycle Management: Nexla manages the full lifecycle of Databricks clusters. It can spawn clusters, deploy the necessary agents and jars, and even terminate clusters once the job is complete. This ensures efficient resource management and cost savings.
How Nexla’s Spark ETL Works
Source Setup:
When setting up the ETL flow, Nexla offers a user-friendly interface for selecting data sources.You can seamlessly connect to your Databricks cluster for compute without any performance impact.
Transformation Setup:
Nexla allows users to define transformations similar to its standard flows. Basic operations like creating new columns or modifying existing ones can be done no-code or with Spark SQL supporting ANSI-standard SQL for transformations. What’s more, Nexla ensures smooth previews using SQL during pipeline design using the data samples, allowing users to verify the SQL they’ve written before it will be executed on the cluster.
Destination Setup:
Like the source setup, the destination configuration can either point to a cloud storage location or a delta table. Once the destination is defined, the flow is ready for execution.
Pic.1 – Nexla Flow definition example
Pic. 2 – Medallion Architecture example
Bird’s Eye View of Job Execution
Once the flow is set, here’s how Nexla’s Spark ETL executes on Databricks:
- Cluster Setup: Nexla spawns or uses an existing Databricks cluster, using PAT. The cluster is equipped with the necessary JAR files to handle the ETL job without any performance impact
- Data Processing: The job reads data from cloud storage into a Spark DataFrame, applies the defined transformations, and writes the output back to cloud storage or delta table. The entire process leverages Databricks’ computational resources to run efficiently across multiple worker nodes.
- Metrics and Monitoring: Nexla communicates with its backend throughout the job’s lifecycle, collecting metrics and sending them to the UI for monitoring. This provides insights into the job’s performance, helping users track success, failure, or cancellation.
- Cluster Termination: Once the job is complete, Nexla can automatically terminate the Databricks cluster to save on costs. Alternatively, Nexla can work with an existing cluster if provided by the customer, avoiding unnecessary resource allocation.
- Unity Catalog: Nexla fully integrates with unity catalog for all the spark workloads.
Pic. 3 – Nexla and Databricks integration
Conclusion
Nexla’s integration with Databricks represents a huge step forward in scaling ETL processes. The out of box integration with Unity Catalog enables all the new features of Databricks like Data Intelligence Platform, GenAI etc. By leveraging Databricks’ powerful compute environment and Spark’s distributed processing capabilities, Nexla provides a flexible, cloud-native solution for transforming and managing data pipelines. Stay tuned for further updates as Nexla continues to enhance its Spark ETL integration with Databricks!
