[SPARK-35558] Avoid redundant computation in retrieval of approximate quantiles

• Type: Improvement
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 3.1.3
• Fix Version/s: 3.2.0
• Component/s: Spark Core
• Assignee: Unassigned
• Reporter: Neoklis
• Created: 28/May/21 18:36
• Updated: 05/Jun/21 19:25
• Resolved: 05/Jun/21 19:25

Description

QuantileSummaries.query() is called repeatedly to obtain the approximate quantiles (e.g., in approximate_percentiles([0, 0.1, 0.5, 0.9])) but performs redundant computation in each invocation and repeated passes over the quantile sketch. These redundant computations can be eliminated by providing a version of query() that accepts the array of percentiles.

Issue Links

• links to: [Github] Pull Request #32700 (alkispoly-db)