Rook is composed of golang project and can be built directly with standard golang
tools,
and storage software (like Ceph) that are built inside containers. We currently support
three different platforms for building:
- Linux: most modern distros should work although most testing has been done on Ubuntu
- Mac: macOS 10.6+ is supported
An Intel-based machine (recommend 2+ cores, 8+ GB of memory and 128GB of SSD). Inside your build environment (Docker for Mac or a VM), 6+ GB memory is also recommended.
The following tools are need on the host:
- curl
- docker (1.12+) or Docker for Mac (17+)
- git
- make
- golang
- rsync (if you're using the build container on mac)
You can build the Rook binaries and all container images for the host platform by simply running the
command below. Building in parallel with the -j
option is recommended.
make -j4
Developers may often wish to make only images for a particular backend in their testing. This can
be done by specifying the IMAGES
environment variable with make
as exemplified below. Possible
values for are as defined by subdir names in the /rook/images/
dir. Multiple images can be separated by a space.
make -j4 IMAGES='ceph' build
Run make help
for more options.
Official Rook builds are done inside a build container. This ensures that we get a consistent build, test and release environment. To run the build inside the cross container run:
> build/run make -j4
The first run of build/run
will build the container itself and could take a few
minutes to complete.
If you're running the build container on the Mac using Docker for Mac, the build container will rely on rsync to copy source to and from the host. To reset the build container and it's persistent volumes, you can run the below command. You shouldn't have to do this often unless something is broken or stale with your build container:
build/reset
You can also run the build for all supported platforms:
make -j4 build.all
Or from the cross container:
build/run make -j4 build.all
Currently, only amd64
platform supports this kind of 'cross' build. In order to make build.all
succeed, we need to follow the steps specified by the below section Building for other platforms
first.
We suggest to use native build
to create the binaries and container images on arm{32,64}
platform, but if you do want to build those arm{32,64}
binaries and images on amd64
platform
with build.all
command, please make sure the multi-arch feature is supported. To test run the
following:
> docker run --rm -ti arm32v7/ubuntu uname -a
Linux bad621a75757 4.8.0-58-generic #63~16.04.1-Ubuntu SMP Mon Jun 26 18:08:51 UTC 2017 armv7l armv7l armv7l GNU/Linux
> docker run --rm -ti arm64v8/ubuntu uname -a
Linux f51ea93e76a2 4.8.0-58-generic #63~16.04.1-Ubuntu SMP Mon Jun 26 18:08:51 UTC 2017 aarch64 aarch64 aarch64 GNU/Linux
In order to build container images for these platforms we rely on cross-compilers and QEMU. Cross compiling is much faster than QEMU and so we lean heavily on it.
In order for QEMU to work inside docker containers you need to do a few things on the linux host. If you are using a recent Docker for Mac build you can skip this section, since they added support for binfmt and multi-arch docker builds.
On an Ubuntu machine with a 4.8+ kernel you need to run install the following:
DEBIAN_FRONTEND=noninteractive apt-get install -yq --no-install-recommends \
binfmt-support
qemu-user-static
You also need to run the following on every boot:
docker run --rm --privileged hypriot/qemu-register
you can install a systemd unit to help with this if you'd like, for example:
cat <<EOF > /etc/systemd/system/update-binfmt.service
[Unit]
After=docker.service
[Service]
Type=oneshot
ExecStartPre=/usr/bin/docker pull hypriot/qemu-register
ExecStart=/usr/bin/docker run --rm --privileged hypriot/qemu-register
RemainAfterExit=yes
[Install]
WantedBy=multi-user.target
EOF
systemctl enable update-binfmt.service
Doing a complete build of Rook and the dependent packages can take a long time (more than an hour on a typical macbook). To speed things up we rely heavily on image caching in docker. Docker support content-addressable images by default and we use that effectively when building images. Images are factored to increase reusability across builds. We also tag and timestamp images that should remain in the cache to help future builds.
To prune the number of cached images run make prune
. There are two options that control the level of pruning performed:
PRUNE_HOURS
- the number of hours from when an image was last used (a cache hit) for it to be considered for pruning. The default is 48 hrs.PRUNE_KEEP
- the minimum number of cached images to keep in the cache. Default is 24 images.
Every PR and every merge to master triggers the CI process in Jenkins. On every commit to PR and master the Jenkins CI will build, run unit tests, and run integration tests. If the build is for master or a release, the build will also be published. If any of the CI stages fail, then the process is aborted and no artifacts are published. On every successful build Artifacts are pushed to an s3 bucket. On every successful master build, images are pushed to dockerhub.com.
During the integration tests phase, all tests under /tests/integration are run. It may take a while to run all Integration tests. Based on the changes in the PR, it may not be required to run the full CI. For faster CI options, add the following tags somewhere in the body of the main PR message.
Jenkins will skip the build process completely if [skip ci]
is found in the PR message.
This option should be used when updating documentation, yaml, or other files that
do not change runtime code. If your PR is marked as WIP and will not be ready for some time to run integration tests,
this tag should also be added, then removed later when the CI should be triggered.
This option should not be used if there are any changes to .go
files, CI scripts, etc.
In case there are known CI failures that are blocking a merge, project maintainers can add this flag to unblock the CI and merge the PR. This should be used with extreme care so regressions are not introduced.
Jenkins will only run the tests for an individual storage provider if specified. There is a keyword for each storage provider that has integration tests:
[test cassandra]
[test ceph]
[test cockroachdb]
[test edgefs]
[test nfs]
Since the majority of effort for a PR generally focuses on a single storage provider it will save time in the CI if you only trigger the tests for your provider. If this key phrase is not found in the PR, all tests will be executed.
These options should not be used if there are any changes to code which is shared with other storage providers. If there is any risk of affecting another storage provider, all tests should be executed in the CI.
In master and release builds, all test suites will run on all supported versions of K8s. In PR builds, all the test suites will run on some version of K8s. For bigger changes, it is recommended to run all test suites on all versions of K8s by using the tag:
[test full]