Skip to content

Deprecate native_comet scan (remove from auto scan mode) #2186

Description

@andygrove

What is the problem the feature request solves?

As a step towards removing native_comet scan support, I propose that we stop falling back to this scan when in auto mode and fall back to Spark instead.

The main case where we currently fall back to native_comet is for byte and short fields when COMET_SCAN_ALLOW_INCOMPATIBLE is false.

Describe the potential solution

No response

Additional context

No response

Activity

  1. andygrove commented on Aug 19, 2025

    @andygrove
    MemberAuthor

    A quick experiment shows that there are relatively few tests that fail (or need updating) when we remove native_comet from auto mode:

    PR tests

    fix: native Unsafe row accessors return incorrect results
    

    Spark SQL core-2

    Filters should be pushed down for vectorized Parquet reader at row group level
    filter pushdown - StringPredicate
    SPARK-17091: Convert IN predicate to Parquet filter push-down
    SPARK-34562: Bloom filter push down
    view-with-default-collation.sql
    

    Spark SQL hive-1

    SPARK-30201 HiveOutputWriter standardOI should use ObjectInspectorCopyOption.DEFAULT
    

    Spark SQL hive-3

    Change column collation
    
  2. added this to the 0.11.0 (October) milestone on Aug 19, 2025
  3. removed this from the 0.11.0 (October/November) milestone on Sep 8, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions