Liquid Clustering 101: What Every Databricks Developer Should Know

In the ever-evolving world of data management, Databricks has unveiled a game-changer: Liquid Clustering for Delta Lake. Imagine a dynamic data layout approach that not only simplifies your data decisions but also supercharges your query performance. Dive into this article to unlock the secrets of Liquid Clustering, a feature that promises to redefine how we think about data layout in Delta Lake. Whether you’re a data enthusiast or a seasoned professional, get ready to embark on a journey of discovery and innovation. Let’s dive deep into the world of Databricks Liquid Clustering and explore its transformative potential!

Databricks Liquid Clustering is a revolutionary feature introduced in Databricks Runtime 13.2 and above. It serves as a replacement for traditional table partitioning and ZORDER, aiming to simplify data layout decisions and enhance query performance. With the flexibility to redefine clustering keys without the need to rewrite existing data, Liquid Clustering allows data layouts to evolve in tandem with analytic requirements over time.

How Does Liquid Clustering Work?

Understanding Data Layout Decisions

Liquid Clustering is designed to replace the traditional methods of table partitioning and ZORDER. Instead of being bound by fixed data layouts, Liquid Clustering provides the flexibility to change clustering keys without the need to rewrite existing data. This ensures that as your analytic needs evolve, your data layout can adapt seamlessly.

Simplifying Query Performance

With Liquid Clustering, query performance is optimized. By providing a more flexible approach to data layout, it ensures that queries are faster and more efficient. This is especially beneficial for tables that are often filtered by high cardinality columns, have a significant skew in data distribution, or grow rapidly.

Benefits of Liquid Clustering for Delta Tables

Liquid Clustering offers several advantages:

  1. Flexibility in Data Layout: It allows for the redefinition of clustering keys without rewriting the existing data.
  2. Optimized Query Performance: By simplifying data layout decisions, it ensures faster and more efficient queries.
  3. Enhanced Concurrency: Azure Databricks provides enhanced concurrency for Delta tables with Liquid Clustering enabled.
  4. Evolving Data Layout: As analytic needs change over time, Liquid Clustering ensures that the data layout can adapt accordingly.

Use Cases and Examples of Liquid Clustering

Databricks recommends Liquid Clustering for all new Delta tables. Here are some scenarios that benefit from it:

  • Tables are frequently filtered by high cardinality columns.
  • Tables with a significant skew in data distribution.
  • Rapidly growing tables that need regular maintenance and tuning.
  • Tables with concurrent write requirements.
  • Tables with changing access patterns.
  • Tables where typical partition keys result in too many or too few partitions.

Comparison with other Clustering Methods

While traditional methods like Hive-style partitioning and Z-order indexing have their merits, Liquid Clustering offers a more flexible and efficient approach. For instance, if you’re converting an existing table, Databricks suggests using partition columns as clustering keys for Hive-style partitioning and ZORDER BY columns for Z-order indexing.

Implementation steps for enabling Liquid Clustering

To leverage the benefits of Liquid Clustering, it’s essential to enable it during the table creation process. Here’s how:

  1. Use the CLUSTER BY phrase in the table creation statement. This ensures that the Azure Databricks client manages all layout and optimization operations.
  2. Once enabled, run OPTIMIZE jobs to incrementally cluster data.
-- Create an empty table
CREATE TABLE table1(col0 int, col1 string) USING DELTA CLUSTER BY (col0);

-- Using a CTAS statement
CREATE EXTERNAL TABLE table2 CLUSTER BY (col0)  -- specify clustering after table name, not in subquery
LOCATION ‘table_location’
AS SELECT * FROM table1;

-- Using a LIKE statement to copy configurations
CREATE TABLE table3 LIKE table1;
--Change the Cluster Key
ALTER TABLE table_name CLUSTER BY (new_column1, new_column2);

--disable the cluster Key

Choosing clustering keys for optimal performance

Selecting the right clustering keys is crucial for maximizing query performance. Databricks recommends choosing keys based on commonly used query filters. If two columns are correlated, only one needs to be added as a clustering key. For existing tables, consider using partition columns or ZORDER BY columns as clustering keys.

Row-level concurrency on Databricks

Azure Databricks offers row-level concurrency for clustered tables, reducing conflicts between concurrent write operations. This feature enhances the efficiency of operations like OPTIMIZE, INSERT, MERGE, UPDATE, and DELETE.

Writing data to a clustered table

To write data to a clustered table, ensure you’re using a Delta writer client compatible with all Delta write protocol table features. While most operations don’t automatically cluster data on write, some, like INSERT INTO operations and CTAS statements do. However, it’s essential to run OPTIMIZE frequently to ensure efficient clustering.

Triggering clustering for existing tables

To trigger clustering, use the OPTIMIZE command on your table. Liquid Clustering is incremental, meaning data is only rewritten when necessary. For optimal performance, schedule regular OPTIMIZE jobs. For best performance, Databricks recommends scheduling regular OPTIMIZE jobs to cluster data. For tables experiencing many updates or inserts, Databricks recommends scheduling an OPTIMIZE job every one or two hours. Because liquid clustering is incremental, most OPTIMIZE jobs for clustered tables run quickly.

--Trigger the Liquid clustering job
OPTIMIZE table_name;

Reading data from a clustered table

Reading data from a clustered table is straightforward. Use any Delta Lake client that supports reading deletion vectors and include clustering keys in your query filters for the best results.

SELECT * FROM table_name WHERE cluster_key_column_name = "some_value";

Viewing clustering information for a table

To view the clustering keys of a table, use the DESCRIBE commands. This provides insights into how the table is clustered and aids in making informed decisions.

--See how table is clustered

Limitations and considerations for using Liquid Clustering

While Liquid Clustering offers numerous benefits, there are some limitations:

  • Only columns with collected statistics can be specified as clustering keys.
  • A maximum of four columns can be specified as clustering keys.
  • Structured Streaming workloads don’t support clustering-on-write.

Future Development and Updates for Liquid Clustering

As Databricks continues to innovate, we can expect more updates and features for liquid clustering. With its current capabilities and the promise of future enhancements, Liquid Clustering is set to redefine data layout and query performance in the world of data analytics.

For more databricks performance optimization refer to this article.

FAQs (Frequently Asked Questions):

  1. What is Databricks Liquid Clustering?
    • Databricks Liquid Clustering is a feature in Databricks Runtime 13.2 and above that replaces table partitioning and ZORDER to simplify data layout decisions and optimize query performance.
  2. How does Liquid Clustering improve data access and performance?
    • By providing a flexible approach to data layout and eliminating the need for rewriting data when redefining clustering keys, Liquid Clustering ensures faster and more efficient queries.
  3. What are the benefits of using Liquid Clustering for Delta tables?
    • Benefits include flexibility in data layout, optimized query performance, enhanced concurrency on Azure Databricks, and the ability for data layout to evolve with analytic needs.
  4. Are there any limitations or drawbacks to using Liquid Clustering with Databricks Delta Lake?
    • Yes, there are limitations such as the ability to specify only columns with statistics collected for clustering keys and a maximum of 4 columns as clustering keys.


Databricks Liquid Clustering is a transformative technique in the world of big data analytics. By simplifying data layout decisions and optimizing query performance, it ensures that organizations can efficiently process and analyze vast amounts of data. As data continues to grow and evolve, techniques like Liquid Clustering will be pivotal in ensuring that businesses can derive actionable insights from their data.

+ There are no comments

Add yours

Leave a Reply