Initializing client with sync_tensorboard=True crashes

wandb --version && python --version && uname

• Weights and Biases version: 0.8.18
• Python version: Python 3.5.3
• Operating System: Linux

Description

Trying to syncronize tensorboard to weights and biases client, and the program keeps crashing with the following log:

    wandb.init(sync_tensorboard=True)
  File "/home/giacaglia/tacotron2/venv/lib/python3.5/site-packages/wandb/__init__.py", line 889, in init
    util.get_module("wandb.tensorboard").patch()
AttributeError: 'NoneType' object has no attribute 'patch'

The following are the libraries installed in our machine:

absl-py==0.8.1
anykeystore==0.2
apex==0.1
argh==0.26.2
astor==0.8.0
audioread==2.1.8
cachetools==3.1.1
certifi==2019.11.28
chardet==3.0.4
Click==7.0
configparser==4.0.2
cryptacular==1.5.5
cycler==0.10.0
decorator==4.4.1
defusedxml==0.6.0
docker-pycreds==0.4.0
gast==0.2.2
gitdb2==2.0.6
GitPython==3.0.5
google-auth==1.7.1
google-auth-oauthlib==0.4.1
google-pasta==0.1.8
gql==0.1.0
graphql-core==2.2.1
grpcio==1.25.0
h5py==2.10.0
hupper==1.9.1
idna==2.8
inflect==0.2.5
joblib==0.14.0
Keras-Applications==1.0.8
Keras-Preprocessing==1.1.0
librosa==0.6.0
llvmlite==0.30.0
Markdown==3.1.1
matplotlib==2.1.0
numba==0.46.0
numpy==1.13.3
nvidia-ml-py3==7.352.0
oauthlib==3.1.0
opt-einsum==3.1.0
PasteDeploy==2.0.1
pathtools==0.1.2
pbkdf2==1.3
Pillow==6.2.1
plaster==1.0
plaster-pastedeploy==0.7
promise==2.2.1
protobuf==3.10.0
psutil==5.6.7
pyasn1==0.4.7
pyasn1-modules==0.2.7
pyparsing==2.4.5
pyramid==1.10.4
pyramid-mailer==0.15.1
python-dateutil==2.8.1
python3-openid==3.1.0
pytz==2019.3
PyYAML==5.2
repoze.sendmail==4.4.1
requests==2.22.0
requests-oauthlib==1.3.0
resampy==0.2.2
rsa==4.0
Rx==1.6.1
scikit-learn==0.21.3
scipy==1.0.0
sentry-sdk==0.13.5
shortuuid==0.5.0
six==1.13.0
smmap2==2.0.5
SQLAlchemy==1.3.11
subprocess32==3.5.4
tensorboard==1.10.0
tensorboardX==1.1
tensorflow-estimator==1.14.0
tensorflow-gpu==1.10.0
termcolor==1.1.0
torch==1.0.0
transaction==2.4.0
translationstring==1.3
Unidecode==1.0.22
urllib3==1.25.7
velruse==1.1.1
venusian==3.0.0
wandb==0.8.18
watchdog==0.9.0
WebOb==1.8.5
Werkzeug==0.16.0
wrapt==1.11.2
WTForms==2.2.1
wtforms-recaptcha==0.3.2
zope.deprecation==4.4.0
zope.interface==4.7.1
zope.sqlalchemy==1.2

What I Did

Just running a simple training of our neural network. When running wandb.init(project=‘project_name’) it works, but not with sync_tensorboard=True

[vanpelt]

This is fixed in the 0.10.x series of our library.

[EricCousineau-TRI]

It looks like this was triggered by the change in 72b8ef982d2, using util.get_module("wandb.tensorboard"), which very aggressively eats import errors: https://github.com/wandb/client/blob/4c46d8d093eb3a52748fb28bb1d7386efd9375e3/wandb/util.py#L122-L126

However, probably most relevant, @giacaglia, is I saw tensorboardX==1.1 in your freeze output. I can reproduce your error with tensorboardX==1.4, but not tensorboardX==1.5 (that was in my original repro, but I have now edited it).

So it looks like this is generally resolved by upgrading to tensorboardX>=1.5, which is now explicitly called via merge of #1098

I think the solution is to change the relevant line to util.get_module("wandb.tensorboard", required=True), so that it fails fast and shows the true error.