Race condition breaking parsing on multiprocessing

[AI091]

Bug

I am frequently getting this error
Error converting document: [Errno 2] No such file or directory: '/root/.EasyOCR//model/temp.zip'

I think it is the race condition mentioned here JaidedAI/EasyOCR#1335, it seems like an issue on EasyOCR side, but my question is , is there a way around that? otherwise running concurrent instances of docling isn't reliable

[dosubot]

Hi @AI091! I'm Dosu and I’m helping the docling team.

This is a known EasyOCR issue: when running multiple processes, they all use the same temp file for downloads, which leads to race conditions and file not found errors (EasyOCR issue #1335). Docling does not add any built-in locking or unique temp file handling for EasyOCR, so this can break concurrent runs (source).

The most reliable workaround is to pre-download all required EasyOCR models before runtime and disable downloads during processing. You can do this by:

1. Use Docling's EasyOcrModel static method to download models into a persistent directory:

   from docling.models.easyocr_model import EasyOcrModel
   EasyOcrModel.download_models(local_dir="/your/model/dir")

2. In your pipeline options, set download_enabled=False and point model_storage_directory to your pre-downloaded model directory. For example:

   from docling.datamodel.pipeline_options import EasyOcrOptions
   ocr_options = EasyOcrOptions(
       model_storage_directory="/your/model/dir",
       download_enabled=False
   )

3. If you run multiple processes, make sure each process uses its own unique model directory, or ensure all processes only read from the same pre-populated directory and never trigger downloads (source).

If you want to avoid these issues entirely, consider switching to another OCR engine like RapidOCR, which respects cached files and does not re-download models every run (source).

There is currently no built-in fix in Docling for this EasyOCR race condition; a fundamental solution would require changes in EasyOCR or adding synchronization logic in Docling (source).

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}