About why jupyter_tensorboard=0.1.8 and tensorboard>=1.12 are incompatible

About

I debugged an issue I got with the jupyter-tensorboard plugin.

Versions

$ pip list | grep tensor
jupyter-tensorboard  0.1.8
tensorboard          1.12.0
tensorflow           1.12.0

Error

[E 2018-12-21 16:09:55.174 SingleUserLabApp web:1670] Uncaught exception POST /user/erik.sundell/api/tensorboard?1545408595205 (10.132.0.6)
    HTTPServerRequest(protocol='https', host='jupyter.allting.com', method='POST', uri='/user/erik.sundell/api/tensorboard?1545408595205', version='HTTP/1.1', remote_ip='10.132.0.6')
    Traceback (most recent call last):
      File "/opt/conda/lib/python3.6/site-packages/tornado/web.py", line 1590, in _execute
        result = method(*self.path_args, **self.path_kwargs)
      File "/opt/conda/lib/python3.6/site-packages/tornado/web.py", line 3006, in wrapper
        return method(self, *args, **kwargs)
      File "/opt/conda/lib/python3.6/site-packages/jupyter_tensorboard/api_handlers.py", line 36, in post
        .new_instance(data["logdir"], reload_interval=reload_interval)
      File "/opt/conda/lib/python3.6/site-packages/jupyter_tensorboard/tensorboard_manager.py", line 137, in new_instance
        purge_orphaned_data=purge_orphaned_data)
      File "/opt/conda/lib/python3.6/site-packages/jupyter_tensorboard/tensorboard_manager.py", line 35, in create_tb_app
        tensorboard.assets_zip_provider)
      File "/opt/conda/lib/python3.6/site-packages/tensorboard/backend/application.py", line 158, in standard_tensorboard_wsgi
        reload_task)
    TypeError: TensorBoardWSGIApp() takes from 4 to 5 positional arguments but 6 were given

Hypothesis

This repo is overriding a function called TensorBoardWSGIApp in the application class imported from tensorboard, this can be seen below (notice the scroll wheel).

https://github.com/lspvic/jupyter_tensorboard/blob/bb33ce64098dcb6de2833af4003f4696fd360fb6/jupyter_tensorboard/tensorboard_manager.py#L100-L114

But when the internal methods of tensorboard.application runs, such as the standard_tensorboard_wsgi method, it results in an error as it is passing 6 arguments as would make sense with their new function signature, see:

  return TensorBoardWSGIApp(flags.logdir, plugins, loading_multiplexer,
                            reload_interval, flags.path_prefix,
                            reload_task)




def TensorBoardWSGIApp(logdir, plugins, multiplexer, reload_interval,
                       path_prefix='', reload_task='auto'):

Solution idea

We investigate if it is enough to simply add the missing reload_task='auto' to the function signature we define and override the old one with.

Preliminary investigation

jupyter-tensorboard - currently not matching tensorboard as it should

https://github.com/lspvic/jupyter_tensorboard/blob/bb33ce64098dcb6de2833af4003f4696fd360fb6/jupyter_tensorboard/tensorboard_manager.py#L100-L114

tensorboard

def TensorBoardWSGIApp(logdir, plugins, multiplexer, reload_interval, path_prefix=""):
    path_to_run = parse_event_files_spec(logdir)
    if reload_interval >= 0:
        # We either reload the multiplexer once when TensorBoard starts up, or we
        # continuously reload the multiplexer.
        start_reloading_multiplexer(multiplexer, path_to_run, reload_interval, reload_task)
    return TensorBoardWSGI(plugins, path_prefix)

Conclusion

Adding reload_task to the TensorBoardWSGIApp function signature, add passing it forward to start_reloading_multiplexer as the last argument would resolve this issue.

UPDATE: Okay, I need to iterate as we have apparently also overridden start_reloading_multiplexer.

Action

~Creating a PR to fix it.~

UPDATE: This was not a as easy I had hoped, I need more experience with these repositories to continue - but lack the time to invest in it. Anyhow. We are calling application.reload_multiplexer from start_reloading_multiplexer but that isn’t available since this commit that is part of the 1.12.0 release of tensorboard.

Current status

The jupyter-tensorboard package requires tensorboard(-gpu)<1.12 as far as I understand.

[asianzhang]

I have the same problem; is this package going to be updated anytime soon?

[lspvic]

jupyter-tensorboard 0.1.9 has tf 1.2 support, please upgrade via pip install -U jupyter-tensorboard