Docker Community Forums

Share and learn in the Docker community.

Docker build failed to register layer; read-only file system

Hi everyone,

I just downloaded Docker, on a Windows 10 machine and am having issues tagging a Docker image to my repository on DockerHub. I have the Dockerfile and requirements.txt file in a directory on my local machine named ‘Dockerfile’. I have been navigating to that directory in terminal and running

docker build -t sicily33/sicily-tf .

whereby sicily33 is my username and sicily-tf is the repository. The Dockerfile in question is an nvidia base image with CUDA/cnn libraries so I can run tensorflow-gpu on an NVIDIA Tesla (Data Serve) GPU on the HPC server at my university. I currently get an error in the build process which is:

ERROR 1/7 FROM docker.io/nvidia/cuda:11.0-cudnn8-devel-ubuntu18.04@sha256:11777cee30f0bbd7cb4a3da562fdd0 1135.8s

The build then never actually completes, because of this final error:

1/7 FROM docker.io/nvidia/cuda:11.0-cudnn8-devel-ubuntu18.04@sha256:11777cee30f0bbd7cb4a3da562fdd0926adb2af02069dad7cf2e339ec1dad036:
failed to register layer: Error processing tar file(exit status 1): write /usr/local/cuda-11.0/targets/x86_64-linux/lib/libcusolverMg.so.10.6.0.245: read-only file system

The image has been tested from another respository (an end to end Linux system, with the Linux version of Docker), and I can pull it from that repository using singularity on the HPC server. I’m not sure if this is because I am building with a Windows system, all I wanted to do was push the file to my repository so it could be hosted from there. I have found little information on the read-only file system, especially because I’m not running multiple containers. I have tried

  • Restarting my Docker

  • Redownloading the files

  • Resetting to factory defaults and cleaning/purging any data

Any advice would be amazing!! I’m so stuck