CUDA not available at build time

Some deep learning libraries refuse to get installed if there is no CUDA-capable device available. Of course, you can fix this by defining a default runtime in your docker configuration. However, this is not an option when your build system does not even have a GPU.

Therefore, an alternative is to skip the CUDA device check by inserting the FORCE_CUDA environment variable in your docker build as follows:


CUDA architectures for PyTorch

If your container needs to build PyTorch (e.g., if you cannot use a PyTorch base image), then you can supply the list of NVIDIA architectures via the TORCH_CUDA_ARCH_LIST environment variable. By using ARG, you can define a default value and still override it at build time via the --build-arg option:

ARG TORCH_CUDA_ARCH_LIST="Kepler;Kepler+Tesla;Maxwell;Maxwell+Tegra;Pascal;Volta;Turing"

Interactive timezone prompt

In case you get the following prompt for configuring your timezone data (which will fail unattended builds):

Configuring tzdata

Please select the geographic area in which you live. Subsequent configuration
questions will narrow this down by presenting a list of cities, representing
the time zones in which they are located.

 1. Africa      4. Australia  7. Atlantic  10. Pacific  13. Etc
 2. America     5. Arctic     8. Europe    11. SystemV
 3. Antarctica  6. Asia       9. Indian    12. US
Geographic area:

Then you can precede your command with DEBIAN_FRONTEND=noninteractive.

In case of apt-get, use something like this:

RUN DEBIAN_FRONTEND=noninteractive apt-get install -y ...

Or like this:

RUN export DEBIAN_FRONTEND=noninteractive && \
    apt-get install -y ...

Finally, the brute force method is to define it globally via the ENV command:

ENV DEBIAN_FRONTEND noninteractive

Missing shared library

A common error that you will encounter is installations via pip failing due to a shared library not being present, similar to this:

error: cannot open shared object file: No such file or directory

In order to remedy the problem, you need to determine the package that contains this shared library. In case of Ubuntu (or Debian) you can use the Ubuntu Packages Search:

On that page, scroll to the Search the contents of packages section and enter the file name of the missing shared library (e.g., If you know the distribution (e.g., bionic release) and architecture (e.g., amd64) then you can restrict the search further. Finally, click on Search.

From the results page, select the appropriate package name and include that in your docker build.