--host does not work anymore

[Daniel-01]

Issue description

Using Tensorboard 2.8.0:

tensorboard --logdir my_dir --host 0.0.0.0
TensorFlow installation not found - running with reduced feature set.
Error: Must not specify both --host and --bind_all.

Note that I did NOT specify --bind_all. Yet it complains that I have done so.
The above command works fins in tensorboard 2.5.0 and was the most convinent way to make the board accessible from a remote computer.

[bileschi]

Sorry for the trouble:
Tried to reproduce locally and the command appears to work for me. Both in clean venv with 2.8.0 and at HEAD

$ tensorboard --logdir ~/tensorboard-logdirs/scalars_demo --host 0.0.0.0

...
NOTE: Using experimental fast data loading logic. To disable, pass
    "--load_fast=false" and report issues on GitHub. More details:
    https://github.com/tensorflow/tensorboard/issues/4784

TensorBoard 2.8.0 at http://0.0.0.0:6006/ (Press CTRL+C to quit)

Is it possible that your shell is somehow adding the --bind_all arg?

Can you try the following in a clean virtual env

make a clean virtual env:

$ deactivate
$ python3 -m venv /tmp/tb-venv
$ pip3 install tensorboard

Test the fresh TensorBoard install

$ tensorboard --version
$ tensorboard --logdir my_dir --host 0.0.0.0

[Daniel-01]

Yes, that works.
Must be something weird in the docker l am using: nvcr.io/nvidia/pytorch:22.01-py3
Maybe Nvidia adds --bind_all automatically somehow?
Sorry for the false alarm.

[bileschi]

No worries! Glad you're back into a working state. Closing the issue for now, please reopen if this is not actually complete.

[huxuan]

For folks using the nvidia NGC Pytorch docker image, the source code of the built-in tensorboard seems to be changed by some reason. The following sed command will revert the inconsistent behavior.

sed -i "s/\"--bind_all\", default=True,/\"--bind_all\",/g" /opt/conda/lib/python3.8/site-packages/tensorboard/plugins/core/core_plugin.py

[guunergooner]

• I followed @huxuan suggestion and modified the default parameters of bind_all to solve this problem

root@localhost:~$ pip show tensorboard
Name: tensorboard
Version: 2.9.0
Summary: TensorBoard lets you watch Tensors Flow
Home-page: https://github.com/tensorflow/tensorboard
Author: Google Inc.
Author-email: packages@tensorflow.org
License: Apache 2.0
Location: /usr/local/lib/python3.8/dist-packages
Requires: absl-py, google-auth, google-auth-oauthlib, grpcio, markdown, numpy, protobuf, requests, setuptools, tensorboard-data-server, tensorboard-plugin-wit, werkzeug, wheel
Required-by: tensorflow

root@localhost:~$ tensorboard --version
2.9.0

root@localhost:~$ grep bind_all /usr/local/lib/python3.8/dist-packages/tensorboard/plugins/core/core_plugin.py
            default=None,  # like localhost, but prints a note about `--bind_all`
network on both IPv4 and IPv6, see `--bind_all`, with which this option is
            "--bind_all", default=False,
        elif flags.host is not None and flags.bind_all:
            raise FlagsError("Must not specify both --host and --bind_all.")

[aisensiy]

No one is trying to fix this? I have to sed it myself??

[bileschi]

Is this an actual issue with TensorBoard, or is it an issue with some vended version of TensorBoard?