Error communicating with backend
See original GitHub issueHi,
Sorry for creating this as a new issue, but technically it is. I added a comment to #1287 about this, but since it is a different problem, I thought it would be best to track it in a new issue. Copying the comment from the previous thread:
I am on version 0.10.4 of the client and I faced a similar error which I’m guessing is network error. It happened on one of the multiple similar runs.
From what I can tell, it seems something went wrong with login/init. Can the client circumvent this without crashing? Something simple I can think of is just allowing the user to increase the timeout, so the client just keeps polling the backend till it connects, rather than stopping the run. I’m not sure if this has problems I haven’t thought about.
Thanks!
Here’s the stack trace:
wandb: ERROR Error communicating with backend
Traceback (most recent call last):
File "runners/mme/mme.py", line 64, in <module>
config=args, reinit=True, project=project)
File "/home/grad3/samarthm/bitbucket-misc/ssda_mme/utils/ioutils.py", line 398, in init
wandb.init(*args, **kwargs)
File "/home/grad3/samarthm/anaconda3/envs/pytorch3conda/lib/python3.7/site-packages/wandb/sdk/wandb_init.py", line 460, in init
run = wi.init()
File "/home/grad3/samarthm/anaconda3/envs/pytorch3conda/lib/python3.7/site-packages/wandb/sdk/wandb_init.py", line 378, in init
raise UsageError(error_message)
wandb.errors.error.UsageError: Error communicating with backend
Issue Analytics
- State:
- Created 3 years ago
- Comments:48 (14 by maintainers)
Top GitHub Comments
Issue-Label Bot is automatically applying the label
bug
to this issue, with a confidence of 0.67. Please mark this comment with 👍 or 👎 to give our bot feedback!Links: app homepage, dashboard and code for this bot.
Hmm I was wrong, I got this error again with 0.10.8
After some debugging, I found that
wandb.init(...)
works, but when I use thepytorch_lightning.loggers.Wandb(...).experiment
it fails, giving the sameError communicating with backend
.This might be related to some hard coded arguments in the
pytorch_lightning
wandb.init
call.I ended up using the following script. One could inherit the logger, but I’ll just initialize
_experiment
variable in pl wandblogger.If it fails, it successfully initializes on the second try.