Description

Currently, the MODE aggregate function is not supported natively in Comet and falls back to Spark execution.

MODE is a commonly used statistical aggregate that returns the most frequent value in a group. The lack of native support prevents queries using MODE from benefiting from Comet’s execution pipeline.

Motivation

• Enables full native execution for queries involving MODE
• Avoids fallback to Spark, improving performance and consistency
• Brings Comet closer to feature parity with Spark SQL

Proposed Change

Add native support for the MODE aggregate function with:

• Full correctness across all supported data types
• Proper handling of null and edge cases
• Support for partial and final aggregation (mergeable state)
• Integration with Comet’s GroupsAccumulator for grouped queries

Expected Behavior

Queries using MODE should:

• Execute fully within Comet (no fallback)

• Return correct results across all supported types

• Handle edge cases such as:

  • all-null inputs
  • ties (consistent with Spark behavior)