Python logging: 'Operation not supported' after upgrading to DBRT 6.1

[zachary_jones]

My organization has an S3 bucket mounted to the databricks filesystem under

/dbfs/mnt

. When using Databricks runtime 5.5 and below, the following logging code works correctly:log_file = '/dbfs/mnt/path/to/my/bucket/test.log' logger = logging.getLogger('test-logger') logger.setLevel(logging.INFO) handler = logging.FileHandler(str(log_file)) handler.setLevel(logging.INFO) logger.addHandler(handler)

logger.info('test')

After upgrading to Databricks runtime 6.1, the above code produces a logging error "OSError: [Errno 95] Operation not supported". Here's the stack trace that is printed:

Traceback (most recent call last):
  File "/databricks/python/lib/python3.7/logging/__init__.py", line 1038, in emit
    self.flush()
  File "/databricks/python/lib/python3.7/logging/__init__.py", line 1018, in flush
    self.stream.flush()
OSError: [Errno 95] Operation not supported

The strange thing is that regular Python file I/O works fine with the same file. (i.e. I can

open()

and

write()

to that filepath successfully.) Any idea what's going on?

[SanghyunLee]

> The strange thing is that regular Python file I/O works fine with the same file. (i.e. I can open() and write() to that filepath successfully.) Any idea what's going on?

If you try to do a further operation after opening the file it will throw the same error. I'm also experiencing the same issue.

[Etch · Accepted solution]

According to the limitations in the Docs, starting from runtime 6.0, random writes to dbfs are no longer supported.

[lycenok]

Probably it's worth to try to rewrite the emit ... https://docs.python.org/3/library/logging.html#handlers

This works for me:

class OurFileHandler(logging.FileHandler):

 def emit(self, record):
   # copied from https://github.com/python/cpython/blob/master/Lib/logging/__init__.p
   if self.stream is None:
     self.stream = self._open()
   try:
     msg = self.format(record)
     stream = self.stream
     # issue 35046: merged two stream.writes into one.
     stream.write(msg + self.terminator)
     self.flush()
   except RecursionError:  # See issue 36272
     raise
   except Exception:
     self.handleError(record)

# logger must be defined 
ch = logging.FileHandler(log_file_path)
ch = OurFileHandler(log_file_path)
formatter = logging.Formatter(
    '%(asctime)s: %(name)s (%(levelname)s): %(message)s'
)
ch.setFormatter(formatter)
logger.addHandler(ch)