Don’t show stack traces to ordinary users

[wchargin]

Don’t show stack traces to ordinary users

If TensorBoard dies with an exception, that exception is typically
displayed as a stack trace plus error message. This isn’t very friendly:

$ tensorboard
Traceback (most recent call last):
  File "/usr/local/google/home/wchargin/virtualenv/tf-nightly-20190201-py2.7/bin/tensorboard", line 10, in <module>
    sys.exit(run_main())
  File "/usr/local/google/home/wchargin/virtualenv/tf-nightly-20190201-py2.7/local/lib/python2.7/site-packages/tensorboard/main.py", line 62, in run_main
    app.run(tensorboard.main, flags_parser=tensorboard.configure)
  File "/usr/local/google/home/wchargin/virtualenv/tf-nightly-20190201-py2.7/local/lib/python2.7/site-packages/absl/app.py", line 294, in run
    flags_parser,
  File "/usr/local/google/home/wchargin/virtualenv/tf-nightly-20190201-py2.7/local/lib/python2.7/site-packages/absl/app.py", line 351, in _run_init
    flags_parser=flags_parser,
  File "/usr/local/google/home/wchargin/virtualenv/tf-nightly-20190201-py2.7/local/lib/python2.7/site-packages/absl/app.py", line 213, in _register_and_parse_flags_with_usage
    args_to_main = flags_parser(original_argv)
  File "/usr/local/google/home/wchargin/virtualenv/tf-nightly-20190201-py2.7/local/lib/python2.7/site-packages/tensorboard/program.py", line 182, in configure
    loader.fix_flags(flags)
  File "/usr/local/google/home/wchargin/virtualenv/tf-nightly-20190201-py2.7/local/lib/python2.7/site-packages/tensorboard/plugins/core/core_plugin.py", line 461, in fix_flags
    raise ValueError('A logdir or db must be specified. '
ValueError: A logdir or db must be specified. For example `tensorboard --logdir mylogdir` or `tensorboard --db sqlite:~/.tensorboard.db`. Run `tensorboard --helpfull` for details and examples.

or:

$ tensorboard --logdir . --host wat
Traceback (most recent call last):
  File "/usr/local/google/home/wchargin/virtualenv/tf-nightly-20190201-py2.7/bin/tensorboard", line 10, in <module>
    sys.exit(run_main())
  File "/usr/local/google/home/wchargin/virtualenv/tf-nightly-20190201-py2.7/local/lib/python2.7/site-packages/tensorboard/main.py", line 62, in run_main
    app.run(tensorboard.main, flags_parser=tensorboard.configure)
  File "/usr/local/google/home/wchargin/virtualenv/tf-nightly-20190201-py2.7/local/lib/python2.7/site-packages/absl/app.py", line 300, in run
    _run_main(main, args)
  File "/usr/local/google/home/wchargin/virtualenv/tf-nightly-20190201-py2.7/local/lib/python2.7/site-packages/absl/app.py", line 251, in _run_main
    sys.exit(main(argv))
  File "/usr/local/google/home/wchargin/virtualenv/tf-nightly-20190201-py2.7/local/lib/python2.7/site-packages/tensorboard/program.py", line 209, in main
    server = self._make_server()
  File "/usr/local/google/home/wchargin/virtualenv/tf-nightly-20190201-py2.7/local/lib/python2.7/site-packages/tensorboard/program.py", line 245, in _make_server
    return self.server_class(app, self.flags)
  File "/usr/local/google/home/wchargin/virtualenv/tf-nightly-20190201-py2.7/local/lib/python2.7/site-packages/tensorboard/program.py", line 298, in __init__
    super(WerkzeugServer, self).__init__(host, flags.port, wsgi_app)
  File "/usr/local/google/home/wchargin/virtualenv/tf-nightly-20190201-py2.7/local/lib/python2.7/site-packages/werkzeug/serving.py", line 577, in __init__
    self.address_family), handler)
  File "/usr/lib/python2.7/SocketServer.py", line 417, in __init__
    self.server_bind()
  File "/usr/local/google/home/wchargin/virtualenv/tf-nightly-20190201-py2.7/local/lib/python2.7/site-packages/tensorboard/program.py", line 373, in server_bind
    super(WerkzeugServer, self).server_bind()
  File "/usr/lib/python2.7/BaseHTTPServer.py", line 108, in server_bind
    SocketServer.TCPServer.server_bind(self)
  File "/usr/lib/python2.7/SocketServer.py", line 431, in server_bind
    self.socket.bind(self.server_address)
  File "/usr/lib/python2.7/socket.py", line 228, in meth
    return getattr(self._sock,name)(*args)
socket.gaierror: [Errno -2] Name or service not known

We can improve on this wall of text by adding a top-level exception
handler that just prints the error message (the last line in the above
block). We may want to also save the stack trace to a file somewhere in
case the user does want to debug it (e.g., they’re a TensorBoard
developer).

(c/o @manivaradarajan)

[nfelt]

My $0.02 is that I agree we should avoid stacktraces in cases where we already know it's not a bug in TensorBoard (like bad CLI arguments), but I dislike the notion that we should suppress all stacktraces by default. It makes it that much more annoying to debug actual issues, and means an extra roundtrip (potentially many extra roundtrips) with users to get info about what actually failed.

I personally think that as long as we don't needlessly emit stacktraces for common user errors, it's fine to emit them when encountering an actual unexpected failure mode. Many of our users are at least somewhat technical, and I know that as a technical user if something encounters an unexpected bug I'd rather see a stacktrace that might actually help me figure out the issue than some generic "TensorBoard failed unexpectedly, oops!" error.