Giters

facebookresearch / faiss

A library for efficient similarity search and clustering of dense vectors.

Home Page:https://faiss.ai

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Error when reading data in a stream

Nav94 opened this issue · comments

commented

Summary

I am reading a saved faiss model in a streaming fashion during containerized execution in python. I obtain a byte stream of the model and convert it into a faiss::IOReader using:
f = io.BytesIO()
temp = faiss.PyCallbackIOReader(f.read)
temp = faiss.BufferedIOReader(temp)
test_nn = faiss.read_index(temp)

However, I get a runtime error when I am reading a saved faiss model in a stream: Error in faiss::Index* faiss::read_index(faiss::IOReader*, int) at /__w/faiss-wheels/faiss-wheels/faiss/faiss/impl/index_read.cpp:414: Error: 'ret == (1)' failed: read error in : 0 != 1 (Resource temporarily unavailable).

This error does not occur if I just read the file using the filename such as:
test_nn = faiss.read_index("nnscorer_search_index.faiss"),
and am able to run inference with the model this way.

I am not sure if I am missing a step when I am converting the byte object or is there a problem with the streamed data?

Platform

OS: macOS 12.5.1

Faiss version: 1.7.1

Installed from: pip

Faiss compilation options: pip install faiss

Running on:

  • CPU
  • GPU

Interface:

  • C++
  • Python

Reproduction instructions

The original model, "nnscorer_search_index.faiss" was saved using the same faiss version, 1.7.1.

commented

It is likely due to the reader implementation. You should make sure the read function waits until all the data is ready, possibly looping over reads from the underlying channel.