This document gives a detailed breakdown of the various build processes and options for building Strimzi from source.
- Developer Quick Start
- Build Pre-Requisites
- Using an IDE
- Build and deploy Strimzi from source
- Build details
- Helm Chart
- Running system tests
- DCO Signoff
- Building container images for other platforms with Docker
buildx
To build Strimzi from a source you need a Kubernetes or OpenShift cluster available. If you do not have an existing Kubernetes cluster, you can install Minikube to have access to a cluster on your local machine.
You will also need access to several command line utilities. See the pre-requisites section for more details.
To build this project you must first install several command line utilities and a Kubernetes or OpenShift cluster.
make
- Make build systemmvn
(version 3.5 and above) - Maven CLIhelm
(version 3) - Helm Package Management System for Kubernetesasciidoctor
andasciidoctor-pdf
- Documentation generation.- Use
gem
to install the latest version for your platform.
- Use
yq
- (version 4.2.1 and above) YAML manipulation tool.- Warning: There are several different
yq
YAML projects in the wild. Use this one. You need version 4.2.1 or above.
- Warning: There are several different
docker
- Docker command line clientshellcheck
- ShellCheck is a GPLv3 tool that gives warnings and suggestions for bash/sh shell scripts.
In order to use make
these all need to be available on your $PATH
.
The make
build uses GNU versions of find
, sed
and other utilities and is not compatible with the BSD versions
available on macOS. When using macOS, you have to install the GNU versions of find
and sed
. When using brew
, you
can do brew install gnu-sed findutils grep coreutils
.
This command will install the GNU versions as gcp
, ggrep
, gsed
and gfind
and our make
build will automatically pick them up and use them.
The build requires bash
version 4+ which is not shipped with macOS but can be installed via homebrew. You can
run brew install bash
to install a compatible version of bash
. If you wish to change the default shell to the
updated bash run sudo bash -c 'echo /usr/local/bin/bash >> /etc/shells'
and chsh -s /usr/local/bin/bash
The mvn
tool might install the latest version of OpenJDK during the brew install. For builds on macOS to succeed,
OpenJDK version 17 needs to be installed. This can be done by running brew install openjdk@17
. For maven to read the
new Java version, you will need to edit the ~/.mavenrc
file and paste the following
line export JAVA_HOME=/Library/Java/JavaVirtualMachines/openjdk-17.jdk/Contents/Home
.
You may come across an issue of linking from the above step. To solve this run this command:
sudo ln -sfn /usr/local/opt/openjdk@17/libexec/openjdk.jdk /Library/Java/JavaVirtualMachines/openjdk-17.jdk
.
If this throws an error that it cannot find the file or directory, navigate into /Library/Java/
(or however deep you
can) and create a new folder named JavaVirtualMachines
followed by creating a file named openjdk-17.jdk
. The folder
structure after everything is said and done should look like /Library/Java/JavaVirtualMachines/openjdk-17.jdk
. After
doing that run the command at the beginning again and this should link the file and allow you to use maven with OpenJDK
version 17.
When running the tests, you may encounter OpenSSL
related errors for parts that you may not have even worked on, in
which case you need to make sure you are using OpenSSL
and not LibreSSL which comes by default with macOS.
In order to run the integration tests and test any changes made to the operators you will need a functioning Kubernetes or OpenShift cluster. This can be a remote cluster, or a local development cluster.
The default Minikube setup should allow the integration tests to be run without additional configuration changes.
Sometimes however, updates to Minikube may prevent the test cluster context being correctly set. If you have errors due
to oc
commands being called on your Minikube cluster then explicitly set the TEST_CLUSTER
environment variable
before running the make
commands.
export TEST_CLUSTER=minikube
The project includes a lot of code generation which takes place within the maven build process, that is why just importing the project might end up with a lot of Cannot resolve symbol
errors.
Follow these steps to import the project in IntelliJ IDEA, run code generation and trigger a reindex in IntelliJ.
- After cloning the repo, open the folder in IntelliJ IDEA.
- From the toolbar in the Maven tool window, click on
Generate Sources and Update Folders For All Projects
button to initiate the code generation. - Restart the IDE via Find Action: press
Ctrl+Shift+A
and type Restart IDE.
Afterwards IntelliJ should no longer have any Cannot resolve symbol
errors.
Note: After running the Maven build in the terminal you might need to reload the project from the Maven tool window.
The build also uses a Java annotation processor. Some IDEs (such as IntelliJ's IDEA) by default don't run the annotation
processor in their build process. You can run mvn clean install -DskipTests
to run the annotation processor
as part of the maven
build, and the IDE should then be able to use the generated classes. It is also possible to
configure the IDE to run the annotation processor directly.
Eclipse users may find the m2e-apt plugin useful for the automatic configuration of Eclipse projects for annotation processing.
To build Strimzi from a source the operator and Kafka code needs to be compiled into container images and placed in a location accessible to the Kubernetes/OpenShift nodes. The easiest way to make your personal Strimzi builds accessible, is to place them on Docker Hub or Quay.io. The instructions below use Docker Hub, however other build options (including options for limited or no network access) are available in the sections below this quick start guide.
-
If you don't have one already, create an account on Docker Hub. Then log your local Docker client into Docker Hub using:
docker login
This sets Docker Hub as the
docker push
registry target. -
Make sure that the
DOCKER_ORG
andDOCKER_REGISTRY
environment variables are set to the same value as your username on the Docker Registry, and the Docker Registry you are using.export DOCKER_ORG=docker_hub_username export DOCKER_REGISTRY=docker_registry_name #defaults to docker.io if unset
By default, the
docker_push
target will build the images under the strimzi organisation ( e.g.strimzi/operator:latest
) and attempt to push them to the strimzi repositories on the Docker Hub. Only certain users are approved to do this, so you should push to your own Docker Hub organisation (account) instead. To do this, make sure that theDOCKER_ORG
andDOCKER_REGISTRY
environment variables are set to the same value as your username on the Docker Registry, and Docker Registry you are using before running themake
commands.
When the Docker images are build, they will be labeled in the form:docker_registry_name/docker_hub_username/operator:latest
in your local repository and pushed to your Docker Hub account under the same label. -
Now build the Docker images and push them to your repository on Docker Hub:
make clean make all
Once this completes you should have several new repositories under your Docker Hub account:
docker_registry_name/docker_hub_username/kafka
docker_registry_name/docker_hub_username/kaniko-executor
docker_registry_name/docker_hub_username/maven-builder
docker_registry_name/docker_hub_username/operator
The tests run during the build can be skipped by setting the MVN_ARGS
environment variable and passing that to the
make command:
make clean
make MVN_ARGS='-DskipTests' all
-
To use the newly built images, update the
packaging/install/cluster-operator/060-Deployment-strimzi-cluster-operator.yaml
to obtain the images from your repositories on Docker Registry rather than the official Strimzi images:
Linuxsed -Ei -e "s#(image|value): quay.io/strimzi/([a-z0-9-]+):latest#\1: $DOCKER_REGISTRY/$DOCKER_ORG/\2:latest#" \ -e "s#(image|value): quay.io/strimzi/([a-zA-Z0-9-]+:[0-9.]+)#\1: $DOCKER_REGISTRY/$DOCKER_ORG/\2#" \ -e "s#([0-9.]+)=quay.io/strimzi/([a-zA-Z0-9-]+:[a-zA-Z0-9.-]+-kafka-[0-9.]+)#\1=$DOCKER_REGISTRY/$DOCKER_ORG/\2#" \ packaging/install/cluster-operator/060-Deployment-strimzi-cluster-operator.yaml
macOS
sed -E -i '' -e "s#(image|value): quay.io/strimzi/([a-z0-9-]+):latest#\1: $DOCKER_REGISTRY/$DOCKER_ORG/\2:latest#" \ -e "s#(image|value): quay.io/strimzi/([a-zA-Z0-9-]+:[0-9.]+)#\1: $DOCKER_REGISTRY/$DOCKER_ORG/\2#" \ -e "s#([0-9.]+)=quay.io/strimzi/([a-zA-Z0-9-]+:[a-zA-Z0-9.-]+-kafka-[0-9.]+)#\1=$DOCKER_REGISTRY/$DOCKER_ORG/\2#" \ packaging/install/cluster-operator/060-Deployment-strimzi-cluster-operator.yaml
This updates
060-Deployment-strimzi-cluster-operator.yaml
, replacing all the image references (inimage
andvalue
properties) with ones with the same name but with the repository changed.Note: please ensure you don't commit these changes accidentally.
-
The installation files assume you're installing into the namespace
myproject
. If you wish to use a different one, you'll need to replace it in the installation files.
Linuxsed -Ei "s/myproject/<desired_namespace>/g" packaging/install/cluster-operator/*RoleBinding*.yaml
macOS
sed -E -i '' -e "s/myproject/<desired_namespace>/g" packaging/install/cluster-operator/*RoleBinding*.yaml
This updates the RoleBinding and ClusterRoleBinding files to ensure that the roles binding to the service account use the correct namespace.
Note: please ensure you don't commit these changes accidentally.
-
Then deploy the Cluster Operator by running the following (replace
myproject
with your desired namespace if necessary):# Running against Kubernetes kubectl -n myproject create -f packaging/install/cluster-operator # Running against OpenShift oc -n myproject create -f packaging/install/cluster-operator
-
Finally, you can deploy the cluster custom resource running:
# Running against Kubernetes kubectl -n myproject create -f packaging/examples/kafka/kafka-ephemeral.yaml # Running against OpenShift oc -n myproject create -f packaging/examples/kafka/kafka-ephemeral.yaml
Strimzi includes a Makefile
with various Make targets to build the project.
Commonly used Make targets:
docker_build
for building Docker imagesdocker_tag
for tagging existing imagesdocker_push
for pushing images to a Docker registry
Note: If you are having trouble running any of the Make commands it may help to run
mvn clean
and thenmvn install -DskipTests
before running the commands again.
Strimzi currently developed and tested with Java 17.
The docker_build
target will build the Docker images provided by the Strimzi project. You can build the Docker images
by calling make docker_build
from the root of the Strimzi repository (this will build the Java code as well). Or you
can build only the Docker images without the Java code by calling make docker_build
from the docker-images
subdirectory (this expects that you have built the Java code before).
The docker_build
target will always build the images under the strimzi
organization. This is necessary in order
to be able to reuse the base image you might have just built without modifying all Dockerfiles. The DOCKER_TAG
environment variable configures the Docker tag to use (default is latest
).
Target docker_tag
tags the Docker images built by the docker_build
target. This target is automatically called as
part of the docker_push
target, but can be called separately if you wish to avoid pushing images to an external
registry.
To configure the docker_tag
and docker_push
targets you can set following environment variables:
DOCKER_ORG
configures the Docker organization for tagging/pushing the images (defaults to the value of the$USER
environment variable)DOCKER_TAG
configured Docker tag (default islatest
)DOCKER_REGISTRY
configures the Docker registry where the image will be pushed (default isdocker.io
)
When building the Docker images you can use an alternative JRE or use an alternate base image.
The build assumes the docker
command is available on your $PATH
. You can set the DOCKER_CMD
environment variable
to use a different docker
binary or an alternative implementation such as podman
.
The docker images can be built with an alternative container OS version by adding the environment
variable ALTERNATE_BASE
. When this environment variable is set, for each component the build will look for a
Dockerfile in the subdirectory named by ALTERNATE_BASE
. For example, to build docker images based on alpine,
use ALTERNATE_BASE=alpine make docker_build
. Alternative docker images are an experimental feature not supported by
the core Strimzi team.
The make all
target can be used to trigger both the docker_build
and the docker_push
targets described above. This
will build the Docker images, tag them and push them to the configured repository.
The build can be customised by:
- building fewer Kafka versions
- customising the Maven settings
Other build options:
As part of the Docker image build several versions of Kafka will be built, which can increase build times. Which Kafka versions are to be built are defined in the kafka-versions.yaml file. Unwanted versions can be commented out to speed up the build process.
Running make
invokes Maven for packaging Java based applications (that is, Cluster Operator, Topic Operator, etc).
The mvn
command can be customized by setting the MVN_ARGS
environment variable when launching make all
. For
example:
MVN_ARGS=-DskipTests make all
will compile test code, but not run unit or integration testsMVN_ARGS=-DskipITs make all
will compile test code and run unit tests, but not integration testsMVN_ARGS=-Dmaven.test.skip=true make all
won't compile test code and won't run unit or integration tests the integration tests.
If you do not want to have the docker daemon running on your local development machine, you can build the container images in your Minikube VM by setting your docker host to the address of the VM's daemon:
eval $(minikube docker-env)
The images will then be built and stored in the cluster VM's local image store and then pushed to your configured Docker registry.
You can avoid the docker_push
step and sed
commands above by configuring the Docker Host as above and then running:
make docker_build
make DOCKER_ORG=strimzi docker_tag
This labels your latest container build as strimzi/operator:latest
and you can then deploy the standard CRDs without
changing the image targets. However, this will only work if all instances of the imagePullPolicy:
setting are set
to IfNotPresent
or Never
. If not, then the cluster nodes will go to the upstream registry (Docker Hub by default)
and pull the official images instead of using your freshly built image.
The strimzi-kafka-operator
Helm Chart can be installed directly from its source.
`helm install strimzi-operator packaging/helm-charts/helm3/strimzi-kafka-operator`
The chart is also available in the release artifact as a tarball.
System tests has its own guide with more information. See Testing Guide document for more information.
The project requires that all commits are signed-off, indicating that you certify the changes with the developer
certificate of origin (DCO) (https://developercertificate.org/). This can be done using git commit -s
for each commit
in your pull request. Alternatively, to signoff a bunch of commits you can use git rebase --signoff _your-branch_
.
You can add a commit-msg hook to warn you if the commit you just made locally has not been signed off. Add the following
line to you .git/hooks/commit-msg
script to print the warning:
./tools/git-hooks/signoff-warning-commit-msg $1
The Checkstyle plugin runs on all pull requests to the Strimzi repository. If you haven't compiled the code via maven, before you submit the PR, then formatting bugs can slip through and this can lead to annoying extra pushes to fix things. In the first instance you should see if your IDE has a Checkstyle plugin that can highlight errors in-line, such as this one for IntelliJ.
You can also run the Checkstyle plugin for every commit you make by adding a pre-commit hook to your local Strimzi git
repository. To do this, add the following line to your .git/hooks/pre-commit
script, to execute the checks and fail
the commit if errors are detected:
./tools/git-hooks/checkstyle-pre-commit
Docker and Podman supports building images for different platforms.
If you want to build Strimzi images for other platforms, you can set the environment variable DOCKER_PLATFORM
to pass additional build options and run the build.
For example, following can be used to build Strimzi images for Linux on Arm64 / AArch64:
DOCKER_PLATFORM="--platform linux/arm64" make all
Following steps are needed to add support for new Kafka releases:
- Edit the
kafka-versions.yaml
file- Add a new record with the new Kafka version
- Make sure only one version is set as default
- If needed, set up the new
kafka-thirdparty-libs
directory - Run
make all
to regenerate / update all files - Update the versions in
attributes.adoc
indocumentation/
- Update all example in
packaging/examples/