API coerces byte-valued attributes to strings

[dimaqq]

Describe your environment

OS: any
Python version: any
SDK version: 1.30.0
API version: 1.30.0

What happened?

OTLP proto was expanded to allow for bytes values in attributes about 3 years ago.

#4118 / #4128 implemented that in the OTLP exporter.

Unfortunately, the #4218 alone is not enough.

The bytes-valued attributes are coerced to strings way before they reach the exporter.

It happens here:

opentelemetry-python/opentelemetry-api/src/opentelemetry/attributes/__init__.py

Lines 110 to 121 in afec2dd

	def _clean_attribute_value(
	value: types.AttributeValue, limit: Optional[int]
	) -> Optional[types.AttributeValue]:
	if value is None:
	return None
	if isinstance(value, bytes):
	try:
	value = value.decode()
	except UnicodeDecodeError:
	_logger.warning("Byte attribute could not be decoded.")
	return None

Steps to Reproduce

• set attribute with bytes value that is internally string-safe
• export
• see stringValue instead of bytesValue in the output

Expected Result

produces bytesValue: <bytes>

Actual Result

produces stringValue: <str> or drops the attribute with a warning

Additional context

#4118 added encoding of bytes attributes in OTLP

Would you like to implement a fix?

None

[wasup-yash]

i would be happy to resolve this too, do you have any proposed sol in mind that could be done ?