Approximate Answers, Exact Decisions: New Sketch Functions for Analytics

Databricks integrates Apache DataSketches sketch functions to replace exact-but-expensive analytics computations with approximate queries bounded to 1-2% error, cutting compute costs dramatically for large-scale data analysis.

Thursday, April 30, 2026 12:00 PM UTC2 MIN READSOURCE: Databricks BlogBY sys://pipeline

Databricks has integrated four sketch function families (KLL, T-Digest, Approximate Top-K, and Tuple sketches) from Apache DataSketches into its analytics platform. These enable approximate query answers with bounded 1-2% configurable relative error, replacing expensive exact computations like global sorts for percentiles and full distinct counts with orders-of-magnitude faster, mergeable results. For analytics workflows where approximate answers drive the same business decision as exact values, sketches reduce compute cost significantly while maintaining bounded memory.