tensorflow: "import tensorflow" configures Python logging if tensorboard is not installed

System information

• Have I written custom code (as opposed to using a stock example script provided in TensorFlow): No
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): macOS 10.15.7, Python 3.8 from MacPorts, Python venv
• Mobile device (e.g. iPhone 8, Pixel 2, Samsung Galaxy) if the issue happens on mobile device: N/A
• TensorFlow installed from (source or binary): binary wheel tensorflow-2.4.1-cp38-cp38-macosx_10_11_x86_64.whl from PyPI (using python -m pip install tensorflow)
• TensorFlow version (use command below): v2.4.0-49-g85c8b2a817f 2.4.1
• Python version: 3.8.8
• Bazel version (if compiling from source): N/A
• GCC/Compiler version (if compiling from source): N/A
• CUDA/cuDNN version: N/A
• GPU model and memory: Radeon Pro 560X 4 GB, Intel UHD Graphics 630 1536 MB

Describe the current behavior

After uninstalling tensorboard from the Python virtual environment with pip uninstall tensorboard, executing import tensorflow at the prompt causes logging to be configured: in particular, a StreamHandler handler is added to the root logger.

As a result, in a large Python GUI (PyQt5) application, an import tensorflow at application startup time resulted in many log messages unrelated to tensorflow being printed to the console; without importing tensorflow, those log messages weren’t visible. (Note that the application had set the level of the root logger to logging.DEBUG, and added its own root-level loggers, under the assumption that it would be the only part of the Python environment working with the configuration for the root-level logger.)

Describe the expected behavior

Following usual logging best practices (libraries should only emit log messages but not configure logging; log configuration should be left to the application using the library), import tensorflow does not affect the root logger.

Standalone code to reproduce the issue

Here’s an interpreter session demonstrating the issue

Python 3.8.8 (default, Feb 22 2021, 09:21:28) 
[Clang 12.0.0 (clang-1200.0.32.28)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> import logging
>>> root_logger = logging.root
>>> root_logger.handlers
[]
>>> import tensorflow
WARNING:root:Limited tf.compat.v2.summary API due to missing TensorBoard installation.
WARNING:root:Limited tf.compat.v2.summary API due to missing TensorBoard installation.
WARNING:root:Limited tf.compat.v2.summary API due to missing TensorBoard installation.
WARNING:root:Limited tf.summary API due to missing TensorBoard installation.
>>> root_logger.handlers  # expected an empty list
[<StreamHandler <stderr> (NOTSET)>]

Additional information

This looks like a shallow bug and an easy fix. Here’s the troublesome code: https://github.com/tensorflow/tensorflow/blob/c023a829167aa4a233a36479cd338ef9784b6990/tensorflow/compat_template.__init__.py#L43

Diagnosis: because the Python logging system hadn’t already been configured at the point of the import tensorflow execution, and because tensorboard wasn’t available,_logging.warning was then called as above: that then implicitly configured logging by calling logging.basicConfig.

A simple solution would be to set up a logger in the normal way (logger = logging.getLogger(__name__)) and use logger.warning instead of logging.warning; that way, implicit configuration of the logging machinery is avoided.

We were able to work around this problem in our application by making sure that we configured logging (adding a null handler to the root logger) before importing tensorflow, so that logging.basicConfig was no longer invoked.

Another workaround was to ensure that tensorboard was installed, but since we don’t generally need the functionality provided by tensorboard, our packaging system leaves tensorboard out as a explicit dependency of tensorflow. If the recommendation is that tensorboard should be treated as a required dependency of tensorflow, that would be good to know.