Spanner Columnar Previewed
Google Cloud previewed a Spanner columnar engine for Iceberg lakehouse data — promising Spanner‑scale low latency with vectorized execution and automatic query handling. That feature blurs the line between cloud scale and low‑latency query performance for hybrid data architectures. (x.com)
Google Cloud published the Spanner columnar engine preview on February 26, 2026, in a blog post authored by Jagan R. Athreya and Girish Baliga. (cloud.google.com) The engine implements a dual-storage model that keeps a columnar format alongside Spanner’s row storage and uses vectorized, batch-oriented execution to accelerate scans — Google cites speedups “up to 200×” on live operational data. (cloud.google.com) Spanner’s preview adds automatic query handling that redirects large-scan analytical queries to the columnar representation, plus an on‑demand columnar conversion capability and a new major compaction API to accelerate conversion of existing data. (cloud.google.com) The announcement specifically targets Apache Iceberg lakehouses for “zero‑ETL” serving and notes integration points with BigQuery federated queries and Data Boost to leverage the columnar engine for analytical workloads. (cloud.google.com) The feature is released in Preview (pre‑GA) and is listed under Spanner’s pre‑GA offering terms, with availability tied to Spanner Enterprise and Enterprise Plus editions per Google Cloud documentation and release notes. (docs.cloud.google.com) Google demonstrated the new serving capability using the ClickBench analytical benchmark to highlight scan-heavy workloads, citing the benchmark in its preview materials. (devengoratela.com) For capacity context, Google states Spanner currently handles more than 6 billion queries per second at peak and stores over 17 exabytes of data across customers including Goldman Sachs, Walmart, and Palo Alto Networks—figures Google uses to position the columnar engine for large, mission‑critical fleets. (cloud.google.com)
