Skip to content

Vortex PyArrow Dataset API to implement full filter support #5913

@gatesn

Description

@gatesn

The Dataset API requires implementations to fully apply the filters before returning batches to the caller. The current Vortex implementation instead tends to error if there are unrecognized expressions.

We should:

  1. Attempt to parse the expression as substrait (via Vortex Substrait Rust #5912).
  • If this fails, we run a full scan with post-filtering.
  1. Partition into filters that can / cannot be pushed down
  2. Run any filters that cannot be pushed down using PyArrow prior to returning.

We should also make the Dataset API configurable as to whether or not it uses "view" types like StringView and ListView. We should disable this by default until PyArrow compute has better support for these arrays.

Metadata

Metadata

Assignees

No one assigned

    Labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions