Trouble authenticating pull from Nvidia Docker registry

[Frank Fineis]

Hello,

I'm trying to pull containers from NVIDIA's Docker registry (see https://ngc.nvidia.com/registry/nvidia-tensorflow), which requires an API key. I've created an account with them, acquired a key, and have been able to successfully authenticate via docker login nvcr.io, followed by docker pull nvcr.io/nvidia/tensorflow:18.08-py3. 

Haven't had the same luck in pulling the containers with Singularity though. According to the documentation on authentication, I made a recipe file with the "Username" and "Password" fields.

Bootstrap: docker

Username: $oauthtoken

Password: [long API key]

Registry: nvcr.io

Namespace: nvidia

From: tensorflow:18.08-py3

but I keep getting denied:

ERROR Error getting token for repository nvidia/tensorflow, please check your credentials.

No JSON object could be decoded

I've tried putting the username and/or key in quotes as well. I've also tried exporting $SINGULARITY_DOCKER_USERNAME and $SINGULARITY_DOCKER_PASSWORD and using singularity pull, but to no avail. Any advice?


Thanks!
- Frank

[Frank Fineis]

Hey solution to this -

Thanks to Vanessa, she recommended just using docker2singularity. There was likely something going on on Nvidia's end, so this is a great work around.

[Reidy, Chris - (chrisreidy)]

What I do to obtain Nvidia images (also with Vanessa’s help) is use sregistry from python.

So you pip install sregistry (one time)

Then ‘export SREGISTRY_NVIDIA_TOKEN=”token value” ‘

Then ‘sregistry pull nvidia://theano:18.01 ‘  for example

Which places the image in ~/.singularity/shub

 

Chris

 

*****************************

Chris Reidy

HPC Systems Administrator, Principal

Supercomputing

University of Arizona

*****************************

 

--
You received this message because you are subscribed to the Google Groups "singularity" group.
To unsubscribe from this group and stop receiving emails from it, send an email to singularity...@lbl.gov.