Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Re-Encode Offsets When Writing ListArray to IPC #5809

Open
JasonLi-cn opened this issue May 28, 2024 · 0 comments
Open

Re-Encode Offsets When Writing ListArray to IPC #5809

JasonLi-cn opened this issue May 28, 2024 · 0 comments
Labels
enhancement Any new improvement worthy of a entry in the changelog

Comments

@JasonLi-cn
Copy link
Contributor

JasonLi-cn commented May 28, 2024

Describe the bug
When serializing a String ListArray slice(0, 1), the offsets of the inner StringArray is all serialized into the buffer. Imagine that if the inner StringArray is 10k in length, then the offset is (10k + 1) * 4 bytes, incurring significant overhead.

use arrow_array::{Array, ListArray, RecordBatch, StringArray};
use arrow_buffer::{Buffer, ToByteSlice};
use arrow_data::ArrayDataBuilder;
use arrow_ipc::writer::{DictionaryTracker, IpcDataGenerator, IpcWriteOptions};
use arrow_schema::{DataType, Field, Schema};
use std::sync::Arc;

fn main() {
    test_string_list();
}

fn test_string_list() {
    const INNER_STRING_ARRAY_LEN: usize = 8192 * 10;
    let rows: Vec<String> = std::iter::repeat("hello world".to_string())
        .take(INNER_STRING_ARRAY_LEN)
        .collect();
    let string_array = StringArray::from(rows);

    let list_array: ListArray = ArrayDataBuilder::new(DataType::List(Arc::new(Field::new(
        "item",
        DataType::Utf8,
        true,
    ))))
    .len(6)
    .add_buffer(Buffer::from(
        [0_i32, 1, 2, 3, 4, 5, INNER_STRING_ARRAY_LEN as _].to_byte_slice(),
    ))
    .add_child_data(string_array.into_data())
    .null_bit_buffer(None)
    .build()
    .unwrap()
    .into();

    let columns = vec![Arc::new(list_array) as _];

    let schema = Arc::new(Schema::new(vec![Field::new_list(
        "values",
        Field::new("item", DataType::Utf8, true),
        true,
    )]));
    let batch = RecordBatch::try_new(schema, columns).unwrap();
    encode(&batch);

    for i in 0..=6 {
        let slice = batch.slice(0, i);
        encode(&slice);
    }
}

fn encode(batch: &RecordBatch) {
    let option = IpcWriteOptions::default();
    let generator = IpcDataGenerator::default();
    let mut dict_tracker = DictionaryTracker::new(false);

    let (_, encode_data) = generator
        .encoded_batch(&batch, &mut dict_tracker, &option)
        .unwrap();

    println!(
        "batch slice len: {}, ipc_message len: {}, arrow data len: {}",
        batch.num_rows(),
        encode_data.ipc_message.len(),
        encode_data.arrow_data.len()
    );
}
batch slice len: 6, ipc_message len: 192, arrow data len: 1239232
batch slice len: 0, ipc_message len: 172, arrow data len: 0
batch slice len: 1, ipc_message len: 192, arrow data len: 328000
batch slice len: 2, ipc_message len: 192, arrow data len: 328000
batch slice len: 3, ipc_message len: 192, arrow data len: 328000
batch slice len: 4, ipc_message len: 192, arrow data len: 328000
batch slice len: 5, ipc_message len: 192, arrow data len: 328000
batch slice len: 6, ipc_message len: 192, arrow data len: 1239232

To Reproduce

Expected behavior

Additional context

@JasonLi-cn JasonLi-cn added the bug label May 28, 2024
@tustvold tustvold changed the title [Bug] ipc writer reencode_offsets bug Re-Encode Offsets When Writing ListArray to IPC May 31, 2024
@tustvold tustvold added enhancement Any new improvement worthy of a entry in the changelog and removed bug labels May 31, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
enhancement Any new improvement worthy of a entry in the changelog
Projects
None yet
Development

Successfully merging a pull request may close this issue.

2 participants