Skip to content

HTTPFileSystem reads an incorrect amount of data #1895

@bveeramani

Description

@bveeramani

Here's a minimal-ish repro:

from fsspec.implementations.http import HTTPFileSystem

remote_path = "https://huggingface.co/api/datasets/abisee/cnn_dailymail/parquet/3.0.0/train/0.parquet"

expected_data_size = 256540614

filesystem = HTTPFileSystem()
with filesystem.open(remote_path) as file:
    total_read = 0
    while data := file.read(256 * 1024):
        total_read += len(data)

assert (
    total_read == expected_data_size
), f"Data mismatch: {total_read} != {expected_data_size}"
# AssertionError: Data mismatch: 5767168 != 256540614

This issue causes Ray Data's from_huggingface API to break. See ray-project/ray#54101

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions