| Docker Images | Build Status |
|---|---|
aswf/ci-* |
|
aswftesting/ci-* |
More information:
Changes are documented here
These images are for Continuous Integration testing of various project managed by the ASWF.
Each image (apart from ci-common) is available for multiple VFX Platform Years.
aswf/ci-common:1.0: A base CentOS-7 image with devtoolset-6, clang-7 and cuda.aswf/ci-base:20XX: Based onaswf/ci-commonwith most most VFX Platform requirements pre-installed.aswf/ci-openexr:20XX: Based onaswf/ci-common, comes with all OpenEXR upstream dependencies pre-installed.aswf/ci-ocio:20XX: Based onaswf/ci-common, comes with all OpenColorIO upstream dependencies pre-installed.aswf/ci-opencue:20XX: Based onaswf/ci-common, comes with all OpenCue upstream dependencies pre-installed.aswf/ci-openvdb:20XX: Based onaswf/ci-common, comes with all OpenVDB upstream dependencies pre-installed.aswf/ci-usd:20XX: Based onaswf/ci-common, comes with all USD upstream dependencies pre-installed.aswf/ci-vfxall:20XX: Based onaswf/ci-common, comes with most VFX packages pre-installed.
The VFXPLATFORM_VERSION is the calendar year mentioned in the VFX Platform, e.g. 2019.
The ASWF_VERSION is a semantic version made of the VFXPLATFORM_VERSION as the major version number, and a minor version number to indicate minor changes in the Docker Image that still point to the same calendar year version, e.g. 2019.0 would be followed if necessary by a 2019.1 version.
The minor version here does not point to a calendar month or quarter, it is solely to express that the image has changed internally. We could also have a patch version.
The most precise version tag is the ASWF_VERSION of the image, e.g. aswf/ci-base:2019.0, but it is recommended to use the VFXPLATFORM_VERSION as the tag to use in CI pipelines, e.g. aswf/ci-openexr:2019.
The latest tag is pointing to the current VFX Platorm year images, e.g. aswf/ci-openexr:latest points to aswf/ci-openexr:2019.0 but will be updated to point to aswf/ci-openexr:2020.0 in the calendar year 2020.
There is another dockerhub organisation with copies of the aswf docker images called aswftesting, images published there are for general testing and experimentations. Images can be pushed by any fork of the official repo as long as the branch is called testing. Images in this org will change without notice and could be broken in many unexpected ways!
To get write access to the aswftesting dockerhub organisation you can open a jira issue there.
As of April 2020 there are full 2018, 2019 and 2020 VFX Platform compliant images.
In order to decouple the building of packages (which can take a lot of time, such as clang, Qt and USD) from the management of the CI Images, the packages are built and stored into "scratch" docker images that can be "copied" into the CI images at image build time by Docker.
Storing these CI packages into docker images has the additional benefit of being completely free to store on the docker hub repository.
The main negative point about this way of storing build artifacts is that tarballs are not available directly to download. It is very trivial to generate one and the provided download-package.sh script can be used to generate a local tarball from any package.
Also, CI packages are built using experimental docker syntax that allows cache folders to be mounted at build time, and is built with docker buildx. The new Docker BuildKit system allows the building of many packages in parallel in an efficient way with support for ccache.
Check aswfdocker for python utility usage.
To build packages and images locally follow the instructions to install the aswfdocker python utility.
Packages require a recent Docker version with buildx installed and enabled.
To build all packages:
aswfdocker --verbose build -t PACKAGE --group-name common,base,vfx --group-version 2018,2019,2020To build a single package, e.g. USD:
# First list the available CI packages to know which package belong to which "group":
aswfdocker packages
# Then run the build
aswfdocker --verbose build -t PACKAGE --group-name vfx --group-version 2019 --target usdImages can be build with recent Docker versions but do not require buildx but it is recommended to speed up large builds.
To build all images:
aswfdocker --verbose build -t IMAGE --group-name base,vfx1,vfx2,vfxall --group-version 2018,2019,2020To build a single image:
# First list the available CI images to know which package belong to which "group":
aswfdocker images
# Then run the build
aswfdocker --verbose build -t IMAGE --group-name vfx1 --group-version 2019 --target openexr