Clustering with liquid clustering, Z-ordering, and deletion vectors

Databricks offers several data layout and optimization strategies to speed up queries on Delta tables. Liquid Clustering is the modern, flexible approach recommended over legacy Z-ordering and partitioning, while deletion vectors optimize row-level DML operations by avoiding costly file rewrites.

1 · Learn the must-know

• 1Liquid Clustering replaces both Hive-style partitioning and Z-ordering, using CLUSTER BY to define clustering keys that can be changed anytime without rewriting existing data.
• 2Z-ordering (OPTIMIZE ... ZORDER BY) co-locates related data within files based on specified columns but requires periodic manual runs and does not incrementally cluster new data as efficiently as Liquid Clustering.
• 3Tables cannot combine Liquid Clustering with Hive-style partitioning or Z-ordering; you must choose one strategy per table.
• 4Deletion vectors mark rows as deleted or updated in an auxiliary file instead of immediately rewriting the entire Parquet data file, improving performance for DELETE, UPDATE, and MERGE operations.
• 5Deletion vectors must eventually be reconciled through OPTIMIZE or VACUUM operations to physically remove marked rows and reclaim storage.
• 6Liquid Clustering automatically triggers incremental clustering during writes and can also be optimized further by running OPTIMIZE, making it the preferred default for new Delta tables.

2 · Check your understanding