hll_sketch_agg aggregate function

Applies to: check marked yes Databricks SQL check marked yes Databricks Runtime 13.3 LTS and above

This function utilizes the HyperLogLog algorithm to count a probabilistic approximation of the number of unique values in a given column, and outputs the result as a binary representation known as a sketch buffer. This binary representation is suitable for persistence.

Queries can use the resulting buffers to compute approximate unique counts with the hll_sketch_estimate function.

The hll_union and hll_union_agg functions can also combine sketches together by consuming and merging these buffers as inputs.

The implementation uses the Apache Datasketches library. Please see HLL for more information.


hll_sketch_agg ( expr [, lgConfigK ] )

This function can also be invoked as a window function using the OVER clause.


  • expr: An expression of type INT, BIGINT, STRING, or BINARY against which unique counting will occur.

  • lgConfigK: An optional INT constant between 4 and 21 inclusive with default 12. The log-base-2 of K, where K is the number of buckets or slots for the sketch.

Any NULL in expr is ignored.


A non-NULL BINARY buffer containing the HyperLogLog sketch computed as a result of consuming and aggregating all input values in the aggregation group.


> SELECT hll_sketch_estimate(hll_sketch_agg(col, 12))
    FROM VALUES (1), (1), (2), (2), (3) tab(col);

> SELECT hll_sketch_estimate(hll_sketch_agg(col))
    FROM VALUES (1), (1), (2), (2), (3) tab(col);