+1 to adding this. It's a really awesome idea. It really fits well into the approximate search support that AsterixDB has featured and integrated deeply, ever since the first open source release (pre-apache). Being able to express something that is usually a giant imperative pipeline as a simple declarative expression is very unique.
On Mon, Aug 17, 2026 at 11:35 AM Hongyu Shi <[email protected]> wrote: > > Hi all, Initiating discussion to add cluster-by clause in AsterixDB. > Feature: Cluster-By Clause > > While GROUP BY is the canonical clause for analytic queries to further > understand the trend and nature of the underlying data, our proposed > Cluster-By clause is for users to find the patterns and trends of vector > embeddings by putting similar vector embeddings into some number of > clusters. > > Clustering (CLUSTER BY) functionality for vector embeddings is similar to > grouping (GROUP BY) for structured fields with the difference that > clustering for vector embeddings is a fuzzy and similarity-based clustering > while grouping of structured fields is exact value grouping. Due to the > similarities between the two approaches, clustering functionality is > proposed here to be added to SQL++ syntax using CLUSTER BY and CLUSTER AS, > following the design and syntax of existing GROUP BY and GROUP AS clauses > in SQL++. > APE: > https://cwiki.apache.org/confluence/spaces/ASTERIXDB/pages/406621607/APE+33+CLUSTER+BY > Thanks, Hongyu Shi
