Parquet file reader: open a .parquet file or an in-memory binary and receive an ExArrow.Stream that yields record batches.

The stream interface is identical to ExArrow.IPC.Reader and ADBC streams — use ExArrow.Stream.schema/1, ExArrow.Stream.next/1, and ExArrow.Stream.to_list/1 to consume it.

How Parquet is read (lazy row-group streaming)

Parquet has a footer that is scanned once when the stream is opened, making the schema immediately available via ExArrow.Stream.schema/1. Row groups are then decoded on demand: each call to ExArrow.Stream.next/1 reads and decodes the next row group without touching the rest of the file.

Pushdown options (v0.8+)

Pass a keyword list as the second argument:

• :columns — list of column name strings (projection pushdown)

• :row_groups — list of 0-based row-group indices to read

• :filters — predicate AST evaluated during decode, with row-group statistics pruning when min/max stats allow it:

  {:gt, "score", 0.5}
  {:and, [{:gte, "id", 10}, {:lt, "id", 100}]}
  {:or, [{:eq, "name", "alice"}, {:eq, "name", "bob"}]}

Supported comparison ops: :eq, :ne, :gt, :gte, :lt, :lte. Values may be integers, floats, UTF-8 strings, or booleans.

After open, ExArrow.Parquet.Reader.read_stats/1 reports how many row groups were selected vs skipped.

Examples

{:ok, stream} =
  ExArrow.Parquet.Reader.from_file("/data/events.parquet",
    columns: ["user_id", "score"],
    filters: {:gt, "score", 0.9}
  )

stats = ExArrow.Parquet.Reader.read_stats(stream)
# %{row_groups_total: 12, row_groups_selected: 3, row_groups_skipped: 9}