Tensorboard error after upgrading to 1.4: trying to access flag before flags were parsed

问题

Since upgrading to TF 1.4 I am getting this error when I try to run tensorboard:

Traceback (most recent call last):
  File "/opt/python/3.6.3/bin/tensorboard", line 11, in <module>
    sys.exit(main())
  File "/opt/python/3.6.3/lib/python3.6/site-packages/tensorboard/main.py", line 39, in main
    return program.main(default.get_plugins(),
  File "/opt/python/3.6.3/lib/python3.6/site-packages/tensorboard/default.py", line 71, in get_plugins
    debugger = debugger_plugin_loader.get_debugger_plugin()
  File "/opt/python/3.6.3/lib/python3.6/site-packages/tensorboard/plugins/debugger/debugger_plugin_loader.py", line 46, in get_debugger_plugin
    if FLAGS.debugger_data_server_grpc_port is None:
  File "/opt/python/3.6.3/lib/python3.6/site-packages/absl/flags/_flagvalues.py", line 509, in __getattr__
    raise _exceptions.UnparsedFlagAccessError(error_message)
absl.flags._exceptions.UnparsedFlagAccessError: Trying to access flag --debugger_data_server_grpc_port before flags were parsed.

I am getting this error even when just typing tensorboard with no --logdir specified but also when I do specify a log dir. I notice this has been reported in github as of 5 days ago (https://github.com/tensorflow/nmt/issues/176), but I am surprised not to see more folks reporting this.

I also noticed that I was not able to run a Tensorflow RNN tutorial for the same reason last week, with the error also indicating flags were accessed before being parsed. Has anyone run into this and can you tell me if there's a fix?

回答1:

As instructed in this Github issue, the quick fix is to upgrade Tensorboard to the nightly build:
pip install --upgrade tb-nightly

As also explained this issue will be fixed as soon as soon as TensorFlow nightly 20171122 is released though.

来源：https://stackoverflow.com/questions/47399965/tensorboard-error-after-upgrading-to-1-4-trying-to-access-flag-before-flags-wer