fix: Improve error handling in blob operations

[shuoweil]

This pull request refactors the blob processing functionalities to introduce comprehensive, two-layer error handling, enhancing robustness and providing clearer feedback on failures.

Key Changes:

1. Remote Function Error Handling (bigframes/blob/_functions.py):

   • Wrapped the core logic of all image and PDF processing remote functions (exif_func, image_blur_func, image_resize_func, image_normalize_func, and their _to_bytes variants) in try...excet blocks.
   • These functions now catch all exceptions and return a structured JSON response containing a status field with the error message, preventing remote function crashes.
   • Added validation for image decoding and encoding steps to gracefully handle corrupted or unsupported file formats.

2. Caller-Side Error Handling (bigframes/operations/blob.py):

   • Updated the public-facing blob methods (exif, image_blur, image_resize, image_normalize, pdf_extract, pdf_chunk) to handle potential failures from the remote UDFs.
   • Added try...except blocks around the UDF execution calls (_df_apply_udf) to catch and re-raise exceptions with more contex.
   • Implemented None checks to ensure that UDFs do not return empty results without raising an error.
   • Corrected JSON parsing logic to properly handle both verbose=True (struct-like JSON) and verbose=False (raw content) responses, fixing TypeError issues when accessing fields.
   • Updated all relevant docstrings to include a Raises section, documenting the RuntimeError that can now be expected on processig failures.

This two-layer approach ensures that errors are handled gracefully at both the remote execution layer and the user-facing API layer, making the blob operations more resilient and easier to debug.

Fixes #<454752361> 🦕

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[tswast]

This is not a refactor: because it is changing user-visible behavior. fix: would be more appropriate, IMO.

[tswast]

Can we update the calling function to be more robust so you don't have to fake this?

[shuoweil]

I will return None on error with verbose=False, update type hints and adjust _output_bq_type to reflect this change.

[tswast]

Same here. This seems to indicate a problem in the calling function.

[sycai]

I would refactor this code block to a separate function. Maybe:

def _serialize(value):
  try:
     return json.dumps(value)
  except (...):
     return str(value)

Then here you can just write:

exif_dict[tag_name] = serialize(value)

Pros: less code indentation, better readability.

[shuoweil]

I appreciate the suggestion for the _serialize helper function to improve readability and reduce indentation. However, this specific refactoring cannot be applied to the code within bigframes/blob/_functions.py.

The functions in _functions.py are deployed as python UDFs by extracting their source code using inspect.getsource(). This process only captures the function body itself, not any external helper functions. If we were to extract _serialize into a separate function, the remote UDF execution environment would not have access to it, leading to runtime failures.

Therefore, while the intent to improve readability is valid, this particular approach is not feasible for these UDF function bodies.

[sycai]

Looks like this code is repeated several times in this change. Maybe define a helper function instead?

[sycai]

It feels like we can refactor this repeated logic into a separate helper function too:

def _apply_udf_or_raise_error(self, ...):
    try:
        res = self._df_apply_udf(...)
    ...

then here you can write:

res = self._apply_udf_or_raise_error(...)

[sycai]

The blob function setup does not support nested function call at the moment. Refactor is not realistic.