Log message is not logged to file for exception with unpickleable content

[micuda]

Hi @Delgan ,
at first, thank You for this library. =)

When I try log an exception with unpickleable content the log message is not logged to file. When you logging to e.g. stderr, the behaviour is correct.

In my case I think it is related to pydantic/pydantic#1667. But I also think that should be handled in loguru, because you never know what kind of exception will you log.

Is it possible write to file the same output as for non-enqueued sink?

Related line: https://github.com/Delgan/loguru/blob/0.5.1/loguru/_handler.py#L174

---

An ideal bug report includes:

• The Python version you are using

$ python -V
Python 3.8.2

• The Loguru version you are using (you can find it with print(loguru.version))

$ python -c 'import loguru;print(loguru.__version__)'
0.5.1

• Your operating system name and version (Linux, MacOS, Windows)

$ lsb_release -a 2> /dev/null | grep -i description
Linux Mint 19.2 Tina

• Your development environment and local setup (IDE, Terminal, project context, any relevant information that could be useful)

Because I couldn't create more minimal example, for reproducing the issue you have to:

$ pip install fastapi==0.54.1 httpie==2.2.0 loguru==0.5.1 pydantic==1.5.1 uvicorn==0.11.5

• Some minimal reproducible example

Create file test_log_exception_with_unpickleable_content_fails.py with content:

#!/usr/bin/env python3
from pathlib import Path

from loguru   import logger
from pydantic import BaseModel
from fastapi  import FastAPI, Depends

CURRENT_DIR = Path(__file__).resolve(strict = True).parent

logger.remove()
logger.add(CURRENT_DIR / 'log_file.log', enqueue = True)

app = FastAPI()

@app.middleware('http')
async def test_middleware(request, call_next):
  try:
    response = await call_next(request)

  finally:
    logger.exception('log message')

class Profile(BaseModel):
  first_name: str
  last_name:  str
  age:        int

@app.get('/test', response_model = Profile)
def get_profile():
  return { 'first_name': 'John', 'last_name': 'Doe' }


if __name__ == '__main__':
  import uvicorn

  uvicorn.run('test_log_exception_with_unpickleable_content_fails:app',
              host = '127.0.0.1',
              port = 8000)

Run server:

$ ./test_log_exception_with_unpickleable_content_fails.py

or

$ uvicorn test_log_exception_with_unpickleable_content_fails:app --port 8000

Make a request:

$ http :8000/test

[Delgan]

Hey @micuda, thanks for the detailed bug report! It's always nice to have a reproducible example, it saves a lot of time. 👍

I agree that it would be better for loguru not to panic because of a non-pickable exception.

Basically, I need to make the RecordException serializable here:

loguru/loguru/_recattrs.py

Lines 61 to 67 in 6f86f48

	class RecordException(namedtuple("RecordException", ("type", "value", "traceback"))):
	def __repr__(self):
	return "(type=%r, value=%r, traceback=%r)" % (self.type, self.value, self.traceback)
	def __reduce__(self):
	exception = (self.type, self.value, None) # tracebacks are not pickable
	return (RecordException, exception)

I could call pickle.dumps(self.value) and replace it with None if it fails, but this means the exception will be serialized twice which seems to be a waste of resources. Another alternative is to implement __getstate__() and __setstate__() but this does not work well for namedtuple. In the end, I'm not yet sure how to solve this issue, but I will think about it.

Edit: I figured it out, I can just a new class method.

[Delgan]

So I slightly refactored the __reduce__() function for ExceptionRecattr. If the exception value can't be pickled, it will be replaced with None. This prevents errors in loguru. Thanks for the report. :)

[micuda]

Good job! Thank You. =)