Customization of base images¶
This requires knowledge of Docker concepts and skills in creating custom Dockerfiles.
When building the base image with
a base image is created with:
- Python 2.7 and Python 3.6
- No CUDA support
To build a CUDA-enabled base image, add
-c 1 to the command line:
./bin/dssadmin build-container-exec-base-image -c 1
- This image contains CUDA 9.0 and CuDNN 7.0. If you require other CUDA versions, you would have to create a custom image.
- You can also build it as a separate base image (see
After each upgrade of DSS, you must rebuild all base images.
If you don’t use the
-t flag, DSS builds a base image with this naming scheme:
dku-exec-base-DSS_INSTALL_ID : dss-DSS_VERSION
- DSS_INSTALL_ID is the identifier of the DSS installation, found in the
- DSS_VERSION is the version of DSS.
If you don’t specify anything in the “base image” field of the DSS containerized execution configuration, this tag will automatically be used.
You can build other base images by appending the
-t IMAGE_NAME:IMAGE_VERSION flag to the
./bin/dssadmin build-container-exec-base-image command.
There are cases where you would want to install additional system packages, generally because they are required by your code environment.
DSS does not automatically provide support for doing this, so the generic mechanism is to:
- Build a base image with the regular DSS mechanisms.
- Write a custom Dockerfile that starts from the built base image, and add the required package.
- Build this custom Dockerfile, and output a custom tag.
- Enter this custom tag in the DSS containerized execution configuration.
After each upgrade of DSS, you must rebuild all base images, including custom ones.