Liquid Clustering
Liquid Clustering is a Databricks data engineering capability, introduced July 2023.
Delta Lake data-layout optimization that supersedes Z-ordering and Hive-style partitioning; clustering keys can evolve without rewriting existing data.
- It replaces both partitioning and Z-ordering, and unlike either you can change the clustering keys later without rewriting a single existing file - old ZORDER columns can even be reused as-is.
- Databricks now recommends it for all new tables, over both partitioning and Z-ordering.
- Since June 2025 it can even choose its own keys: automatic liquid clustering, GA on Unity Catalog managed tables, picks clustering keys to suit your queries.
Limitations: Clustering-on-write doesn't support source queries with filters, joins, or aggregations (DBR 15.1 and below); you can't create a liquid-clustered table via Structured Streaming write (DBR 15.4 LTS and below); automatic liquid clustering isn't available for managed Apache Iceberg v2 tables, and needs DBR 18.0 or above on the v3 tables where it is supported; and row-level concurrency isn't supported on managed Apache Iceberg v2 tables, which lack the deletion vectors and row tracking the v3 spec added.
Open in REbricked →- Category
- Data engineering
- Also known as
- CLUSTER BY, liquid, automatic liquid clustering
- Verified
- 2026-09-27