Arrow IPC streaming error when reading files from CheXpert Plus dataset (Expected to be able to read X bytes for message body, got Y)

[sagarbisht123]

Summary

I'm unable to read files from the AIMI / CheXpert Plus dataset using the Redivis Python client. This workflow worked correctly approximately 2 months ago, but now every file-related API call fails with the same Arrow IPC error.

The issue reproduces across multiple environments, so it does not appear to be specific to my machine.

---

Environment

• Python 3.12
• Latest redivis
• Latest pyarrow
• Also reproduced with multiple older versions of both redivis and pyarrow
• Tested on:
  • Kaggle Notebook
  • RunPod (Ubuntu)

---

Dataset

• Organization: AIMI
• Dataset: chexpert_plus

I have full dataset permissions and can successfully:

• Authenticate
• Query metadata
• Run SQL queries
• Export tables
• Read tabular data

The issue only occurs when accessing files.

---

Minimal Reproduction

import redivis

table = redivis.table("aimi.chexpert_plus:5yyj:v1_0.png_train:s6cj")

file = table.file("patient00001/study1/view1_frontal.png")

data = file.read(as_text=False)

OSError: Expected to be able to read 566984 bytes for message body, got 11853