#arrow #extendr #rstats

arrow_extendr

Enables the use of arrow-rs in R using extendr and nanoarrow

4 releases (major breaking)

new 51.0.0 Apr 11, 2024
50.0.0 Jan 18, 2024
49.0.0 Nov 28, 2023
48.0.1 Nov 28, 2023

#76 in FFI

Download history 6/week @ 2024-01-12 2/week @ 2024-01-19 21/week @ 2024-02-16 21/week @ 2024-02-23 4/week @ 2024-03-08 2/week @ 2024-03-15 8/week @ 2024-03-29 130/week @ 2024-04-05

141 downloads per month

MIT/Apache

28KB
374 lines

arrow_extendr

arrow-extendr is a crate that facilitates the transfer of Apache Arrow memory between R and Rust. It utilizes extendr, the {nanoarrow} R package, and arrow-rs.

Versioning

At present, versions of arrow-rs are not compatible with each other. This means if your crate uses arrow-rs version 48.0.1, then the arrow-extendr must also use that same version. As such, arrow-extendr uses the same versions as arrow-rs so that it is easy to match the required versions you need.

Versions:

  • 50.0.0 (compatible with geoarrow-rs 0.1.0)
  • 49.0.0-geoarrow (not available on crates.io but is the current Git version)
  • 48.0.1
  • 49.0.0

Motivating Example

Say we have the following DBI connection which we will send requests to using arrow. The result of dbGetQueryArrow() is a nanoarrow_array_stream. We want to count the number of rows in each batch of the steam using Rust.

# adapted from https://github.com/r-dbi/DBI/blob/main/vignettes/DBI-arrow.Rmd

library(DBI)
con <- dbConnect(RSQLite::SQLite())
data <- data.frame(
  a = runif(10000, 0, 10),
  b = rnorm(10000, 4.5),
  c = sample(letters, 10000, TRUE)
)

dbWriteTable(con, "tbl", data)

We can write an extendr function which creates an ArrowArrayStreamReader from an &Robj. In the function we instantiate a counter to keep track of the number of rows per chunk. For each chunk we print the number of rows.

use extendr_api::prelude::*;
use arrow_extendr::from::FromArrowRobj;
use arrow::ffi_stream::ArrowArrayStreamReader;

#[extendr]
/// @export
fn process_stream(stream: Robj) -> i32 {
    let rb = ArrowArrayStreamReader::from_arrow_robj(&stream)
        .unwrap();

    let mut n = 0;

    rprintln!("Processing `ArrowArrayStreamReader`...");
    for chunk in rb {
        let chunk_rows = chunk.unwrap().num_rows();
        rprintln!("Found {chunk_rows} rows");
        n += chunk_rows as i32;
    }

    n
}

With this function we can use it on the output of dbGetQueryArrow() or other Arrow related DBI functions.

query <- dbGetQueryArrow(con, "SELECT * FROM tbl WHERE a < 3")
process_stream(query)
#> Processing `ArrowArrayStreamReader`...
#> Found 256 rows
#> Found 256 rows
#> Found 256 rows
#> ... truncated ...
#> Found 256 rows
#> Found 256 rows
#> Found 143 rows
#> [1] 2959

Using arrow-extendr in a package

To use arrow-extendr in an R package first create an R package and make it an extendr package with:

usethis::create_package("my_package")
rextendr::use_extendr();

Next, you have to ensure that nanoarrow is a dependency of the package since arrow-extendr will call functions from nanoarrow to convert between R and Arrow memory. To do this run usethis::use_package("nanoarrow") to add it to your Imports field in the DESCRIPTION file.

Dependencies

~17–26MB
~428K SLoC