Missing _make_exception() function in ml-metadata 1.14, causing TFX pipeline execution issues

[sawanjr]

I encountered an issue while running TFX version 1.14 on a Windows platform, which depends on ml-metadata version 1.14. During the execution of my pipeline, I faced the following error:

ml_metadata.errors.UnimplementedError: MLMD Filtering is not supported in Windows platform since ZetaSQL currently does not compile on Windows.

Upon investigating, I found that the error is raised by the _make_exception() function, which is missing in the ml-metadata 1.14 source code. However, this function is present in the official TFX GitHub repository, specifically in the metadata_store.py file. Here's the missing function:

def _make_exception(msg, error_code):
"""Makes an exception with MLMD error code.

Args:
msg: Error message.
error_code: MLMD error code.

Returns:
An exception.
"""

try:
exc_type = errors.exception_type_from_error_code(error_code)
return exc_type(msg)
except KeyError:
return errors.UnknownError(msg)

This discrepancy causes the TFX pipeline to fail when running on Windows, since the current ml-metadata 1.14 package lacks this necessary function.

Steps to reproduce:

Use TFX 1.14 with ml-metadata 1.14 on Windows.
Run a TFX pipeline that involves the ImportExampleGen component.
Observe the error raised during pipeline execution.
Proposed Solution: Please include the missing _make_exception() function in the ml-metadata 1.14 package, or update the version of ml-metadata bundled with TFX 1.14 to include this function.